Stimulated Raman scattering microscopy: fundamentals, instrumentation, and chemical imaging applications

1 History of Raman and coherent Raman scattering (CRS)

Light–matter interaction refers to the fundamental process by which optical photons interact with the electronic and vibrational states of molecules, leading to phenomena, such as reflection, refraction, absorption, scattering, and emission. Raman scattering is a specific light–matter interaction process in which photons undergo inelastic scattering by exchanging energy with molecular vibrations, resulting in scattered photons whose frequency, and thus energy, are shifted relative to the incident photons. The early history of Raman scattering and coherent Raman scattering (CRS) is summarized in Table 1. In 1923, Smekal predicted the occurrence of inelastic light scattering arising from discrete quantum energy exchanges, providing the earliest theoretical foundation for what later became known as the Raman effect (Smekal, 1923). In 1925, Kramers and Heisenberg developed the quantum formulation of light scattering known as the Kramers–Heisenberg dispersion formula, which established the theoretical framework for the Raman effect (Kramers and Heisenberg, 1925). In 1928, Raman and Krishnan reported the first experimental observation of Raman scattering (Raman and Krishnan, 1928). In 1962, 2 years after the invention of the laser (Maiman, 1960), Woodbury and Ng used a ruby laser with a nitrobenzene Kerr cell and observed an unexpected near-infrared (NIR) emission (Woodbury and Ng, 1962), which was later recognized as the first observation of stimulated Raman scattering (SRS) (Eckhardt et al., 1962). In 1963, Hellwarth formulated the semiclassical theory of SRS (Hellwarth, 1963). In 1964, Yajima and Takatsuji reported the first experimental observation of coherent anti-Stokes Raman scattering (CARS) (Yajima and Takatsuji, 1964). In 1965, Maker and Terhune formulated a unified theoretical framework for nonlinear light–matter interactions, integrating SRS, CARS, and four-wave mixing (FWM) within a single third-order (χ 3) nonlinear optical model (Maker and Terhune, 1965). In 1977, Owyoung and Jones first demonstrated SRS spectroscopy using low-power continuous-wave dye lasers, with power levels more than six orders of magnitude lower than those used in earlier pulsed SRS experiments, establishing a key foundation for high-resolution nonlinear Raman spectroscopy (Owyoung and Jones, 1977). In 1982, Duncan and colleagues reported the first CARS microscope using visible picosecond lasers to achieve chemically selective CARS imaging (Duncan et al., 1982). In 1999, Zumbusch and co-workers reported laser-scanning CARS microscopy using near-infrared ultrafast lasers in a collinear geometry, achieving three-dimensional (3D), high-resolution, label-free chemical imaging of biological specimens, marking the beginning of modern CARS microscopy (Zumbusch et al., 1999). In 2007, Ploetz et al. introduced femtosecond SRS microscopy, using a 1-kHz femtosecond laser-amplifier system to generate a white-light probe and a picosecond pump, and demonstrated a broadband hyperspectral implementation of SRS (Ploetz et al., 2007). In 2008, Freudiger et al. reported SRS microscopy using high-repetition-rate picosecond lasers combined with high-frequency lock-in detection, enabling rapid, high-sensitivity imaging of single Raman bands in biological specimens with SRS contrast that is inherently free of the nonresonant background (Freudiger et al., 2008). In 2009, Nandakumar et al. independently reported picosecond SRS microscopy with a quantitative vibrational response (Nandakumar et al., 2009). Since then, CRS microscopy, encompassing both CARS and SRS, has advanced rapidly in sensitivity, speed, and biological applications.

TABLE 1. Chronological overview of the early history of Raman scattering and coherent Raman scattering (CRS), including coherent anti-Stokes Raman scattering (CARS) and stimulated Raman scattering (SRS).

2 Principles of Raman and coherent Raman scattering (CRS)

2.1 Classical description of Raman scattering

Molecules consist of atoms connected by chemical bonds that can be modeled as elastic springs. Because atoms are not rigidly fixed, they undergo continuous motion about their equilibrium positions. These motions give rise to vibrational modes, each corresponding to a specific pattern of bond stretching or bending at a characteristic frequency. The vibrational energy associated with each mode is quantized, meaning that a molecule can only occupy discrete vibrational energy levels. The vibrational energy diagrams of molecules are illustrated in Figure 1A. When external electromagnetic energy is applied, the molecular dipole can be driven at or near these vibrational resonances, allowing transitions between vibrational states. When light irradiates a molecule, the energy of light photons may be reflected, transmitted, absorbed, or scattered. The single-photon energy, E p h, of monochromatic light is given by (Colthup et al., 1990):where ν i and λ i are the linear frequency and wavelength of the incident light, respectively. h = 6.626 × 10 − 34 m 2 k g / s is the Planck constant, and c = 3 × 10 8 m / s is the speed of light in vacuum. For instance, photons with a wavelength of λ i = 785 n m have a linear frequency of ν i = 3.821 × 10 14 H z and single-photon energy of E p h = 2.53 × 10 − 19 J.

E p h = h ν i = h c λ i ,

The Raman effect refers to an inelastic scattering process between photons and molecules, during which a small fraction of the single-photon energy is transferred to or obtained from molecular vibrations. For the Stokes process, a photon loses an energy E v to excite a vibrational mode, so the Stokes-scattered photons have slightly reduced energy:

E Stokes = E p h − E v ,

Since the Stokes-scattered photons have reduced photon energy (E Stokes < E p h), they exhibit longer wavelengths (λ S t o k e s > λ i). In contrast, if the molecules lose vibrational energy, this energy can be transferred to the photons, leading to the emission of anti-Stokes photons with slightly increased energy:

E anti ‐ Stokes = E p h + E v .

Since the anti-Stokes-scattered photons have increased photon energy, they exhibit shorter wavelengths. According to the Boltzmann distribution, at room temperature the Stokes signal of a substance is much stronger than its anti-Stokes signal in spontaneous Raman scattering. Therefore, Raman spectroscopy typically measures the Stokes-scattered photons.

A classical description of the Raman effect adopts the concept of polarizability. When a molecule is placed in the electric field of the light, the electrons and nuclei of the molecule will experience oppositely directed forces applied by the electric field of the light. As a result, the molecule is polarized and an induced dipole moment μ is generated, which can be formulated aswhere E is the strength of the electric field of the incident light, and α is the polarizability, representing the deformability of the electron cloud of the molecule induced by the field. In classical electrodynamics, a dipole moment is a vector quantity that characterizes the separation of positive and negative charges in a molecule. The electric field (E) of the incident light oscillates at the optical frequency ν i, which can be represented as where E 0 is the maximum amplitude of the field. The induced dipole moment of the molecule will oscillate following the same frequency of the field. Combining the previous two equations, the induced dipole moment is expressed as

μ = α E ,

E = E 0 ⁡ cos 2 π ν i t ,

μ = α E 0 ⁡ cos 2 π ν i t .

The polarizability α can be expanded in a Taylor series aswhere α 0 is the equilibrium polarizability, Q is a normal coordinate, which is a single variable that describes the collective displacement of atoms along one characteristic vibrational mode, and ∂ α ∂ Q is the rate of change of polarizability with respect to Q. In the harmonic approximation, higher order terms not shown in the above equation are ignored. The normal coordinate Q varies periodically,where ν v is the frequency of the chemical bond vibration, Q 0 is the maximum amplitude for Q. Combining the previous three equations, we get

α = α 0 + ∂ α ∂ Q Q + ⋯ ,

Q = Q 0 ⁡ cos 2 π ν v t ,

μ = α 0 E 0 ⁡ cos 2 π ν i t + ∂ α ∂ Q Q 0 E 0 ⁡ cos 2 π ν v t cos 2 π ν i t .

Using trigonometric identity, we further write

μ = α 0 E 0 ⁡ cos 2 π ν i t + 1 2 ∂ α ∂ Q Q 0 E 0 cos 2 π ν i − ν v t + cos 2 π ν i + ν v t .

From this expression, the induced dipole moment μ is seen to contain three frequency components: ν i, ν i − ν v, and ν i + ν v, which correspond to Rayleigh scattering, Stokes Raman scattering, and anti-Stokes Raman scattering, respectively (Figure 1A). If a vibrational mode of a molecule causes a change in polarizability (∂ α ∂ Q ≠ 0), the mode is Raman active and produces Raman scattering. If a vibrational mode does not cause a change in polarizability (∂ α ∂ Q = 0), it is Raman inactive and produces no Raman scattering. This condition is known as the Raman selection rule, which requires

∂ α ∂ Q ≠ 0 .

For example, the asymmetric stretching vibrational mode of carbon dioxide (CO₂) is Raman inactive because the opposite changes in the polarizability of the two C-O bonds cancel each other, resulting in no net change in the overall molecular polarizability. Although this asymmetric stretch mode in CO₂ is Raman inactive due to its zero net polarizability change, it is strongly infrared (IR) active because this vibration generates a transient dipole moment, satisfying the IR selection rule (∂ μ ∂ Q ≠ 0) (Colthup et al., 1990).

The Raman signal intensity can be expressed using the Raman cross section σ, which quantifies the probability that an incident photon will undergo inelastic scattering via the Raman effect, as follows:where N is the number of molecules, I 0 is the incident light intensity, and d σ / d Ω is the differential Raman cross section in units of c m 2 s r − 1 m o l e c u l e − 1. Here Ω denotes the solid angle and s r (steradian) is the SI unit of solid angle. Typical values of the Raman cross section d σ / d Ω are extremely small, on the order of 10 − 29 c m 2 / s r for some common liquids, and 10 − 30 c m 2 / s r for biomolecules such as glucose. Consequently, the spontaneous Raman signal is intrinsically weak, posing a fundamental sensitivity limitation for many Raman spectroscopy and microscopy applications (Ferraro et al., 2003).

I R a m a n ∝ N I 0 d σ d Ω ,

2.2 Theory of coherent Raman scattering (CRS)

Under intense laser excitation, the molecular response becomes nonlinear, and higher-order terms in the light–matter interaction must be retained. In nonlinear optics, macroscopic polarization can be expanded in powers of the electric field as (Boyd, 2020)

P t = ε 0 χ 1 E t + χ 2 E 2 t + χ 3 E 3 t + ⋯ .

In this equation, ε 0 is the vacuum permittivity. The first-order susceptibility χ 1 represents the linear optical response that determines the refractive index (n) of the medium. The second-order susceptibility χ 2 describes the nonlinear processes such as second-harmonic generation (SHG), sum-frequency generation (SFG), and difference-frequency generation (DFG), which occur only in non-centrosymmetric media. The third-order susceptibility χ 3 accounts for effects including four-wave mixing (FWM), self-phase modulation (SPM), cross-phase modulation (XPM), third-harmonic generation (THG), and coherent Raman scattering (CRS). To obtain a high instantaneous electric field and thereby enhance the nonlinear signal strength, ultrafast pulsed lasers with picosecond or femtosecond pulse durations are typically used for these nonlinear optics experiments, instead of continuous-wave lasers.

In a degenerate CRS process, two laser beams are employed: a pump beam and a Stokes beam. When the two laser beams are spatially and temporally overlapped within the sample, their interaction drives the occurrence of CRS, as well as other nonlinear processes. The two most widely used CRS modalities for microscopic imaging are coherent anti-Stokes Raman scattering (CARS) and stimulated Raman scattering (SRS). Figure 1B illustrates the energy diagrams of CARS and SRS.

From the diagram, the optical frequency of the CARS emission is equal towhere ω p and ω S are the angular frequencies of the pump and Stokes beams, respectively. ω v is resonant frequency of the vibrational mode (ω v = 2 π ν v). In CARS microscopy, this anti-Stokes frequency ω a s is detected. The resonance condition to achieve maximized CARS signal is

ω a s = 2 ω p − ω S ,

ω p − ω S = ω v .

The intensity of the CARS signal is given by

I C A R S ∝ χ 3 2 I p 2 I S .

Here, I p and I S are the optical intensities of the pump and the Stokes beams, respectively. The Raman-related third-order susceptibility χ 3 is composed of a resonant term and a nonresonant term, as follows:

χ 3 ω v = χ N R 3 + χ R 3 ω v .

For a single Raman vibrational mode, the molecular resonance is often approximated by a Lorentzian line shape, corresponding to a damped harmonic oscillator with a center frequency ω v and a linewidth Γ v. Under this assumption, the resonant term χ R 3 for one vibrational mode can be expressed aswhere i = − 1 is the imaginary unit, A represents the Raman mode strength, and Γ v is the linewidth, which is the half-width at half-maximum (FWHM) of the Lorentzian Raman profile. This expression shows that χ R 3 is a complex quantity, containing both a real part, Re χ R 3, and an imaginary part, Im χ R 3. It is noted that the imaginary part Im χ R 3 follows the Lorentzian function identical to the Raman spectral line profile.

χ R 3 = A ω v − ω p − ω S − i Γ v ,

These equations reveal that CARS signals contain both nonresonant and resonant parts, as well as their coherently mixed component. Therefore, CARS imaging suffers from a strong nonresonant background, which reduces the chemical selectivity and specificity. Extensive efforts have been made to suppress this nonresonant background and to retrieve the resonant CARS signals.

Like CARS, SRS originates from the third-order nonlinear polarization induced by the interaction of the pump and Stokes fields through χ 3. When the frequency difference between the two laser beams satisfies the Raman resonance condition (ω p − ω S = ω v), a vibrational coherence is generated in the medium. This coherence acts back on the optical fields, producing third-order polarization components at both the pump and Stokes frequencies. These nonlinear polarizations interfere coherently with the original optical fields, resulting in small intensity changes: stimulated Raman loss (SRL) on the pump beam, and stimulated Raman gain (SRG) on the Stokes beam, expressed aswhere g R ω v is the Raman gain coefficient, proportional to the imaginary part of the resonant susceptibility:

I S R L ∝ − g R ω v I p I S , I S R G ∝ + g R ω v I p I S ,

g R ω v ∝ Im χ R 3 .

Accordingly, the overall SRS signal intensity (measured through either SRL or SRG) is also proportional to Im χ R 3, as well as the number of molecules and the laser intensities:

I S R S ∝ N Im χ R 3 I p I S .

Since the imaginary part of the resonant susceptibility Im χ R 3 reproduces the Lorentzian Raman line shape, SRS spectra directly reflect the spontaneous Raman spectra, but with significantly higher signal strength due to the stimulated amplification process. Importantly, the nonresonant background term χ N R 3, which is present in CARS, does not contribute to SRS, making SRS nonresonant-background free. In addition, the linear dependence of the SRS signal on the pump and Stokes laser intensities enables straightforward quantitative vibrational imaging of chemical species.

3 Instrumentation and data of Raman-based spectroscopy and microscopy

Raman-based spectroscopy and microscopy rely on optical and electronic instrumentation to excite molecular vibrations and to acquire digitized spectra and images. The instrument architecture and acquisition strategy therefore determine both imaging performance and the structure of the resulting datasets. These datasets enable quantitative chemical imaging but also require systematic processing to suppress laser noise and experimental artifacts and to extract chemically meaningful information. In SRS microscopy, data processing commonly includes spectral and spatial filtering, background and baseline correction, field-of-view alignment, and quantitative normalization across Raman bands (Lin and Cheng, 2023; Zhou et al., 2025). With increasing emphasis on low-dose imaging, high-speed acquisition, and hyperspectral measurements, denoising and reconstruction in SRS microscopy have extended from conventional algorithms to machine learning and deep learning frameworks, often based on convolutional neural networks (CNNs) (Manifold et al., 2019; Abdolghader et al., 2021; Li et al., 2021c; Ranjan et al., 2022). Deep-learning-based unmixing, segmentation, and classification further enable automated extraction of quantitative chemical maps and high-throughput analysis of large hyperspectral datasets, with U-Net architectures widely used for segmentation and generative adversarial networks (GANs) applied for image translation and virtual staining (Hollon et al., 2020; Zhang et al., 2021; Liu et al., 2024; Kondepudi et al., 2025). In this section, we focus on the instrumentation and acquisition strategies for Raman spectroscopy, confocal Raman microscopy, and CARS and SRS microscopy. We also summarize the common data types encountered in Raman and SRS imaging. Specific applications of AI-based methods in SRS image processing and bioimaging studies are illustrated in Section 5.

3.1 Raman spectroscopy and confocal Raman microscopy

Raman spectroscopy is an analytical technique that utilizes the Raman effect to probe the molecular composition and structure of materials (Ferraro et al., 2003). The typical optical configuration of a Raman spectrometer is illustrated in Figure 2A. A narrowband laser line is focused onto the sample to induce Raman scattering, and the emitted Stokes-shifted signals at longer wavelengths are collected and dispersed by a grating-based spectrograph for spectral detection using an array photodetector. The excitation laser line must be sufficiently suppressed by optical filters (including a notch filter) before the scattered light from the sample enters the spectrograph. Raman spectroscopy can also be implemented using optical fibers, offering a flexible and portable configuration suitable for in vivo or in situ measurements (Figure 2B). Such fiber-based Raman systems have been widely applied in biomedical research and clinical diagnostics, including noninvasive cancer detection. For field and point-of-care applications, handheld Raman spectrometers have been developed using integrated optical modules, including the excitation laser, optical filters, and a compact detector (Figure 2C). These systems enable rapid, on-site spectral analysis without the need for bulky laboratory instruments. However, there is often a trade-off between portability and sensitivity, as the compact optical design and reduced laser power may limit signal strength compared with the benchtop systems. A representative Raman spectrum of oleic acid is shown in Figure 2D, covering Raman shifts from 500 to 3,500 cm^-1. The spectral profile with characteristic Raman peaks provides a unique molecular fingerprint that can be used to identify specific chemical species.

To achieve Raman imaging, confocal Raman microscopy has been developed, which enables 3D chemical imaging of material samples by scanning the sample stage, allowing high-resolution mapping of molecular distribution of the specimens. Because the Raman signal is dispersed by a grating-based spectrograph for detection, the excitation beam path must remain fixed. As a result, confocal Raman imaging relies on mechanical two-dimensional (2D) sample-stage scanning, which inherently limits imaging speed compared with laser-beam-scanned optical microscopes.

3.2 Instrumentation for CARS and SRS microscopy

While the Raman effect can be excited using a continuous-wave laser, coherent Raman scattering (CRS) microscopy, including CARS and SRS, requires two synchronized ultrafast pulsed laser beams. In CRS, the frequency (color or wavelength) difference between the pump and Stokes photons is tuned to match the resonant frequency of a Raman-active vibrational mode, which drives coherent excitation of molecular vibration and enhances the signal intensity by several orders of magnitude compared with spontaneous Raman scattering. As illustrated in Figure 3A, most modern SRS microscopes use a dual-beam, diode-pumped solid-state (DPSS) ultrafast laser system. In this configuration, a DPSS femtosecond or picosecond oscillator pumps an optical parametric oscillator (OPO) to generate a broadly tunable near-infrared (NIR) output, typically 700–1,300 nm, that serves as the pump beam, while part of the oscillator output is used directly as the Stokes beam, usually at 1,045 nm. The key advantages of this design are its wide spectral tunability, which covers Raman shifts across the entire 400-4,000 cm^-1 range, and the intrinsically low timing jitter between the two beams, since both originate from the same master oscillator. Since its introduction around 2008, this dual-beam solid-state laser system has become the standard light source for benchtop SRS microscopes used in biomedical and chemical imaging applications (Freudiger et al., 2008; Nandakumar et al., 2009; Cheng et al., 2021).

Both picosecond and femtosecond lasers can be used for SRS microscopic imaging. Transform-limited picosecond pulses with 2–7 ps duration provide a narrow spectral bandwidth of approximately 5 cm^-1 on average, which is ideal for rapid imaging a single Raman band. In contrast, femtosecond pulses of 100–200 fs are about ten times broader in spectral bandwidth, typically spanning ∼30-60 cm^-1, and are therefore not directly usable for single-band excitation without additional spectral control. However, femtosecond sources are preferred for other multiphoton modalities, such as two-photon fluorescence (TPF) and second-harmonic generation (SHG) microscopy, due to their much higher peak power. To adapt femtosecond lasers for SRS, spectral focusing is often applied to impose linear chirps on both the pump and Stokes pulses, improving spectral resolution without losing laser power or imaging speed (Hellerer et al., 2004; Rocha-Mendoza et al., 2008; Fu et al., 2012). Since a femtosecond laser source can be used to do multimodal multiphoton imaging, including SRS, TPF, and SHG, it is often an economical choice for many laboratories. In recent years, fiber-based ultrafast sources have also emerged as compact and alignment-free alternatives to solid-state lasers. These developments are summarized in Section 4.2.

Figure 3B presents a representative schematic of the optical configuration and detection scheme of SRS microscopy. The tunable OPO output pulse trains and the fixed beam pulse trains must first be temporally synchronized through an optical delay line, because the two pulse trains are not by default synchronized within the laser body. Modern SRS systems use a modulation-transfer detection strategy to suppress the laser noise and to achieve near shot-noise-limited sensitivity. The Stokes beam, typically fixed at 1,045 nm, is amplitude-modulated at a radio-frequency typically ranging from a few megahertz (MHz) to 20 MHz using an electro-optic modulator (EOM). For example, with a laser repetition rate of 80 MHz and an EOM modulation at 10 MHz, on average every four pulses are left on followed by four pulses being turned off, periodically. The pump beam remains unmodulated before reaching the sample. The two beams are then combined collinearly by a dichroic mirror and directed into a laser-scanning microscope module. A pair of galvanometer mirrors raster-scan the beams in 2D across the sample. A one-million-pixel image (1,000 × 1,000) can be acquired within about 10 s using linear galvos, while resonant scanners enable video-rate SRS imaging. The beams are tightly focused through a high-numerical-aperture (NA) objective onto the sample to induce a strong third-order nonlinear response, and a condenser with a slightly higher NA collects the transmitted light to minimize background from thermal-lens effects.

Within the focal volume, coherent interaction between the pump and Stokes photons and the molecules in the sample results in stimulated Raman loss (SRL) on the pump beam and stimulated Raman gain (SRG) on the Stokes beam via the third-order nonlinear susceptibility (χ³) of the material. Most SRS microscopes detect SRL on the pump beam because silicon photodiodes have high responsivity across a broad wavelength range below ∼950 nm. After the sample, optical filters with high optical density (OD) block the Stokes beam so that only the pump beam can reach the detector. The pump beam is detected by a large-area, high-dynamic-range silicon photodiode, and the small SRL signal is demodulated by a high-sensitivity lock-in amplifier referenced to the same modulation frequency as the EOM. This electronic demodulation scheme isolates the SRL signal from the strong pump beam and the broadband laser noises and yields a clean, high-sensitivity SRS signal. The demodulated output is then digitally sampled by a data acquisition board and displayed through microscope control software such as ScanImage or LabVIEW. Lock-in detection at MHz frequencies is essential to reject laser noise and to achieve the high signal-to-noise ratio (SNR) required for quantitative SRS imaging.

3.3 Single-band, multicolor, and hyperspectral SRS imaging

Because molecular vibrations give rise to a rich Raman spectrum covering a wide range of frequencies (∼400–4,000 cm^-1), SRS microscopy naturally has a spectral dimension that enables visualization of multiple chemical contrasts associated with various Raman-active vibrational bands. In picosecond SRS microscopy, images are typically acquired at a single Raman band, providing chemical contrast from one selected molecular vibrational mode. By tuning the wavelength of either the pump or the Stokes beam, SRS images can be sequentially collected at multiple Raman bands, producing multicolor images of the same sample that distinguish different molecular components. In addition, it is feasible to continuously tune the excitation wavelengths over a wide range of Raman shifts to enable hyperspectral imaging. Femtosecond SRS microscopy can also perform single-band imaging through spectral focusing, which introduces matched dispersion-based chirping to both the pump and Stokes pulses to achieve high spectral resolution (Fu et al., 2013). When the relative delay between the chirped pulses is scanned, hyperspectral SRS data can be obtained over a narrow range of Raman shifts determined by the bandwidth of the femtosecond pulses. In addition, fiber-based laser systems have been developed that can rapidly sweep the output wavelengths across a broad spectral range, enabling fast and broadband hyperspectral SRS imaging (Ozeki et al., 2012).

Depending on the goal of the studies, different types of SRS datasets, including single-band, multicolor, and hyperspectral images, can be acquired to balance spectral resolution, imaging speed, and SNR. Figure 4A shows representative multicolor SRS images of cultured SKOV-3 cancer cells on a coverglass, illustrating the chemical contrast achievable through vibrationally selective imaging. The image acquired at 2,930 cm^-1, corresponding to the CH₃ symmetric stretching vibration, highlights the protein-rich structures within the cells. The signal obtained at 2,854 cm^-1 probes the CH₂ symmetric stretching vibration, providing strong contrast from lipid-rich compartments. In contrast, the image recorded at 2,790 cm^-1 represents a non-CH background where biomolecular C-H vibrations are absent, confirming that the contrast obtained at the resonant frequencies arises from genuine SRS interactions. The line intensity profiles in Figure 4B further demonstrate the different contrasts of lipids, proteins, and the non-CH background.

Figure 5A shows a representative hyperspectral SRS dataset acquired from cultured HeLa cells, illustrating both the data structure and the chemical imaging capabilities available in the high-wavenumber C-H stretching region. The hyperspectral stack consists of 21 sequential SRS images; each recorded at a distinct Raman shift spanning 2,776-3,089 cm^-1. Every frame contains a full 512 × 512-pixel field of view, enabling spatially resolved spectral analysis across the entire field of view. This dataset was acquired using a femtosecond SRS microscope with spectral focusing for chirping the femtosecond pulses and stepwise tuning of the laser wavelength. With a pixel dwell time of 4.8 μs, the acquisition achieves high-speed imaging while maintaining a high SNR. Across the spectral dimension, distinct intracellular components emerge with varying contrast: at Raman shifts below ∼2,820 cm^-1, the images show minimal signals due to the absence of resonant C-H vibrations; near 2,854 cm^-1, strong SRS signals from lipid-rich structures, particularly lipid droplets, become prominent; and in the 2,900–2,990 cm^-1 range, contrast from protein-rich regions such as cytoplasm and the nucleus increases markedly. This evolution of image contrast demonstrates how hyperspectral SRS encodes biochemical distributions as a function of vibrational resonance frequencies. The plots in Figure 5B display SRS spectra extracted from three intracellular regions: lipid droplets, cytosol, and nuclei. Lipid droplets show a dominant resonance at 2,854 cm^-1 arising from the CH₂ groups in fatty-acyl chains, whereas cytosolic and nuclear spectra feature stronger contributions near 2,930 cm^-1, corresponding to CH₃ groups enriched in protein structures, consistent with the results shown in Figure 4. These region-specific signatures highlight how hyperspectral SRS data enable straightforward extraction of molecular information through band integration or full spectral analysis. This dataset demonstrates the structure of a typical hyperspectral SRS measurement: a stack of spatial images sampled at discrete Raman shifts, each reflecting a specific vibrational resonance. Such datasets support pixelwise spectral extraction, spectral unmixing, ratiometric mapping, and machine-learning-based classification to interrogate cellular chemical composition with submicron resolution.

3.4 Label-free SRS imaging of native molecules

Label-free SRS imaging achieves chemical contrasts relying on the native chemical bonds in the molecules of the sample with no exogenous labels or dyes. Chemically selective imaging of multiple biomolecules, such as lipids, proteins, DNAs, RNAs, polysaccharides, glycogen, cellulose, cholesterol, and water, has been demonstrated in cells and biological tissues (Cheng and Xie, 2015). Characteristic vibrational modes of chemical bonds such as carbon-carbon (C-C or C=C), carbon-hydrogen (C-H), carbon-oxygen (C-O), and oxygen-hydrogen (O-H) have been widely used for label-free imaging of biomolecules. Most biomolecular vibrations exhibit Raman frequencies from 400 to 1800 cm^-1 and 2,800–3,500 cm^-1. In addition, some small-molecule drugs can be directly imaged through their unique Raman-active vibrational modes, which exhibit distinct vibrational frequencies separated from other molecular components in the cellular or tissue samples. SRS imaging of the native drug molecules allows direct monitoring of drug delivery, localization, and metabolism without chemical modification. More broadly, any substance possessing unique Raman peaks can, in principle, be visualized in a label-free manner by SRS, enabling molecularly specific imaging and quantitative analysis of complex biological and chemical systems. However, it is noted that detection sensitivity remains a limitation in SRS imaging of chemical analytes. For instance, when molecular concentrations in the sample fall far below the millimolar (mM) range, it becomes difficult to achieve sufficient SRS signal intensity for reliable imaging. Ongoing developments aim to enhance SRS detection sensitivity and thereby extend its applicability to trace-level molecules. In designing label-free Raman or SRS experiments, researchers should carefully consider the unique Raman bands of the targeted analytes, the strength of the Raman cross-section, and the expected molecular concentration to assess the feasibility of achieving meaningful results.

Label-free SRS imaging offers a unique combination of chemical specificity, quantitative capability, and biocompatibility that distinguishes it from fluorescence-based or label-dependent imaging methods. By directly probing the intrinsic molecular vibrations of native biomolecules, SRS enables visualization of various cellular constituents without the need for exogenous dyes or genetic modification in cells and organisms. This label-free nature minimizes perturbation to live cells and organisms, allowing repeated and longitudinal imaging in intact biological systems and in vivo in humans. SRS provides a linear, concentration-dependent signal that permits quantitative assessment of molecular composition and distribution with high spatial and temporal resolution. These advantages make label-free SRS an exceptionally powerful modality for noninvasive, chemically specific imaging across biological, pharmacological, and clinical applications.

3.5 SRS imaging with Raman probes

Beyond label-free SRS imaging, Raman tags and probes have been used to enhance the detection sensitivity and chemical specificity of SRS imaging (Wei et al., 2016; Li and Yue, 2025). Common bonds used for Raman tags include the oxygen-deuterium bonds (O-D), carbon-deuterium bonds (C-D), carbon-carbon triple bonds (C≡C, alkyne), carbon-nitrogen triple bonds (C≡N, nitrile). Most Raman tags and probes exhibit Raman peaks in the “silent region” (1800-2,800 cm^-1) of biomolecules, enabling cellular and tissue background-free imaging of the targeted molecules with enhanced sensitivity. In addition, the use of Raman tags allows tracing of a specific molecule through the metabolic pathways, providing a unique tool to study metabolism at a single-cell level in health and disease. However, the incorporation of Raman-tagged molecules into living systems may introduce perturbations, such as additional cellular stress or altered metabolic kinetics. Further studies are therefore needed to systematically evaluate and minimize the potential biological impacts of Raman tags to ensure accurate interpretation of dynamic processes in live cells and in vivo.

In this section, we covered Raman-based vibrational microscopy and spectroscopy from the perspectives of system design and data analysis, including Raman, CARS, and SRS imaging platforms, the structure of hyperspectral datasets, and the contrast mechanisms used in SRS imaging. The differences among Raman, CARS, and SRS imaging approaches should be considered in terms of several practical parameters, including imaging speed, quantitative capability, nonresonant background contributions, imaging depth in biological tissues, spectral coverage, and typical application scenarios. In practice, users should select the most appropriate modality based on the specific biomedical question and the imaging capabilities required. Over a dozen previously published review papers have provided comprehensive discussions of the fundamental principles, instrumentation, and diverse bioimaging applications of SRS microscopy (Min et al., 2011; Chung and Potma, 2013; Alfonso-Garcia et al., 2014; Cheng and Xie, 2015; Winterhalder and Zumbusch, 2015; Zhang et al., 2015; Liao and Cheng, 2016; Prince et al., 2017; Li et al., 2021b; Brzozowski et al., 2022; Manifold and Fu, 2022; Tsikritsis et al., 2022; Genchi et al., 2023; Tian et al., 2023; Ranjan and Sirleto, 2024; Cheng et al., 2025; Gao et al., 2025; Min et al., 2025). Readers seeking deeper theoretical background or broader historical context are referred to these reviews. In the following sections, this review focuses on technological innovations that have enabled and reshaped SRS microscopy and highlights recent progress in its biological and biomedical applications.

4 Technological advances of SRS microscopy

Over the past 2 decades, continuous technological innovation has greatly expanded the performance and versatility of SRS microscopy. Improvements in laser sources, detection strategies, system architecture, and probe design have increased imaging sensitivity, spectral coverage, and operational stability, transforming SRS into a robust platform for chemical and biological imaging. Recent progress spans several tightly connected directions: advances in CARS and SRS instrumentation, the maturation of ultrafast laser sources tailored for CRS microscopy, the emergence of CRS endoscopy for in vivo applications, and a growing suite of hybrid and quantum-enhanced SRS approaches that collectively point toward the next generation of coherent Raman imaging.

4.1 Recent advances in SRS microscopy systems

SRS microscopy has evolved from a narrowly tuned single-color nonlinear imaging modality into a versatile, highly engineered broadband platform driven by advances in laser technology, detection schemes, spectral acquisition, and computational approaches. Early implementations relied on dual-beam solid-state near-infrared (NIR) lasers and high frequency modulation with lock-in and balanced detection to reach shot noise limited sensitivity (Freudiger et al., 2008; Freudiger et al., 2014). Building on this foundation, frequency modulation schemes have further improved sensitivity while suppressing nonresonant background compared to the traditional amplitude modulation approach (Ganikhanov et al., 2006; Hill et al., 2019; Xiong et al., 2020; Grassi et al., 2023; Wallmeier et al., 2023). In parallel, hyperspectral SRS imaging has advanced, combining rapid wavelength tuning and broadband spectral coverage to acquire high-content vibrational datasets from materials and biological samples (Hellerer et al., 2004; Rocha-Mendoza et al., 2008; Fu et al., 2012; Ozeki et al., 2012; Fu et al., 2013; Zhang et al., 2013; Fu and Xie, 2014; Liao et al., 2015; Rehault et al., 2015; Figueroa et al., 2018; Lin et al., 2021; Ragupathy et al., 2021; Tan et al., 2023; Xu F. X. et al., 2023; Clark et al., 2024; Yu et al., 2024; Tipping et al., 2025).

Another major line of innovation has focused on contrast enhancement and spatial resolution. Electronic pre-resonance SRS (epr-SRS) detunes pump and Stokes frequencies close to an electronic transition of tailored Raman probes, which boosts the effective nonlinear susceptibility and dramatically increases signal without sacrificing spectral specificity (Wei et al., 2017). Visible and near-resonance SRS push excitation into the visible to exploit shorter wavelengths and near resonant cross-section enhancement, thereby improving spatial resolution toward the 100 nm scale at the cost of increased scattering and stricter control of photodamage (Bi et al., 2018; Zhuge et al., 2021; Choorakuttil et al., 2023; Pruccoli et al., 2023; Colazo et al., 2025). In parallel, there has been a push toward compact, robust, and portable hardware (Légaré et al., 2006; Lombardini et al., 2018; Pshenay-Severin et al., 2021). Synchronously pumped OPOs have been complemented or replaced by turnkey picosecond and femtosecond fiber laser platforms and fiber based optical parametric devices (Freudiger et al., 2014; Brinkmann et al., 2019).

At the frontier of sensitivity, quantum-enhanced (QE)-SRS approaches use squeezed light or entangled photons to surpass the classical shot noise limit and improve SNR and imaging speed at a fixed laser dose, while remaining compatible with bright beams and practical microscopy architectures. Rapid development in this field is ongoing (Dorfman et al., 2014; de Andrade et al., 2020). On the data side, SRS microscopy is now closely integrated with computation (Lin and Cheng, 2023; Zhou et al., 2025). A range of denoising and reconstruction methods, from basic spectral and spatial filters to modern machine learning and deep learning frameworks, is used to improve image quality, especially at low dose or with undersampling (Manifold et al., 2019; Abdolghader et al., 2021; Li et al., 2021c; Ranjan et al., 2022). U-Net models, self-supervised denoisers, and deep spectral unmixing and segmentation tools help accelerate hyperspectral imaging and extract quantitative chemical maps with higher accuracy (Hollon et al., 2020; Zhang et al., 2021; Krishnan Nambudiri et al., 2024; Kondepudi et al., 2025).

A growing class of hybrid SRS technologies couples vibrational excitation to distinct readout mechanisms and imaging geometries, extending the capabilities for bioimaging applications. Stimulated Raman photothermal microscopy utilizes SRS as an energy deposition process and detects the resulting refractive index change, greatly enhancing effective modulation depth and vibrational sensitivity (Zhu et al., 2023; Ge et al., 2025). Stimulated Raman photoacoustic imaging uses vibrationally selective excitation to generate ultrasonic waves for deeper, molecularly specific imaging in scattering tissue (Yakovlev et al., 2010). Flow-based platforms with multicolor and multiplex SRS achieve label-free, high-throughput single-cell chemical profiling (Zhang et al., 2017a; Suzuki et al., 2019). Depth-encoding schemes integrate SRS with projection tomography, spectroscopic optical coherence tomography (OCT), or time-of-flight Bessel light bullets to enable rapid volumetric and deep-tissue chemical imaging without conventional z-scanning (Chen et al., 2017; Robles et al., 2017; Lin et al., 2024). Other approaches use SRS excitation but fluorescence detection, as in stimulated Raman excited fluorescence, which encodes vibrational resonances in a fluorescent readout while preserving single-molecule-level sensitivity (Xiong et al., 2019). Finally, photoswitchable vibrational probes bring light control into SRS, with photochromic diarylethenes and photoswitchable polyynes providing switchable or frequency-shifted Raman bands for optically gated and multiplexed SRS imaging (Ao et al., 2021; Shou and Ozeki, 2021; Yang et al., 2024). In the following parts of this section, we focus on three technological directions shaping current SRS development: the adoption of fiber-laser sources for compact and stable SRS systems, the advancement of CRS endoscopy for in vivo imaging, and emerging quantum-level experiments that probe the fundamental limits of SRS noise and sensitivity.

4.2 Advances in laser sources for SRS microscopy

While dual-beam solid-state ultrafast laser sources remain the workhorse for SRS microscopy, there is a sustained push toward lower-cost, more portable, more rapidly tunable, and lower-noise laser systems. Within the solid-state class, Steinle et al. introduced a simple and stable synchronization-free all-solid-state laser system. In this system, an 8-W, 450-fs ytterbium-doped potassium gadolinium tungstate (Yb:KGW) oscillator at 41 MHz was used to directly pump a single-pass magnesium-oxide-doped periodically poled lithium niobate (MgO:PPLN) optical parametric amplifier (OPA) that was continuous-wave seeded. The continuous-wave seeding suppressed superfluorescence, eliminated pump-seed timing jittering, and minimized spectral drift. Frequency-doubling the OPA output in PPLN crystal with SHG spectral compression yielded narrow-band ∼1.5 ps pump pulses which were tunable from 760 to 820 nm, while a filtered 1,032 nm branch from the oscillator provided a ∼1.7 ps?Stokes beam. Due to the favorable low-noise properties of the bulk oscillator, this all-solid-state system avoids the excess high-frequency noise that is intrinsic in fiber-based systems (Steinle et al., 2016).

An early study by Gambetta et al. reported a compact fiber-format laser source for SRS microscopy that was built from a single erbium (Er)-fiber oscillator. The femtosecond pulses from the oscillator were converted into synchronized, narrowband pump and Stokes pulse trains using spectral compression in separate PPLN crystals. Although the setup was not a fully all-fiber system, this work demonstrated a practical design that replaces complex dual-beam laser systems with spectral shaping of a single fiber oscillator for two-color and multicolor SRS imaging (Gambetta et al., 2010).

Hybrid laser systems that integrate a solid-state laser head with fiber-based subsystems have pushed practical tunability and spectral coverage for hyperspectral SRS imaging. A seminal demonstration of such solid-fiber hybrid systems achieved video-rate (∼30 frames per second) spectral imaging by synchronizing a Ti:sapphire laser with a Yb-fiber amplifier and performing frame-by-frame wavelength stepping via a galvanometer-based tunable band-pass in a 4-f filter configuration. This work illustrates how fiber-based filtering grafted onto a solid-state backbone can deliver both tuning speed and chemical specificity (Ozeki et al., 2012). Building on this philosophy, a multicolor SRS platform was demonstrated using a fast wavelength-tunable Yb-fiber oscillator electronically synchronized to a picosecond Ti:sapphire laser through a two-photon cross-correlator. This system provided ∼32 nm of Stokes tuning (∼300 cm^-1) with ∼0.3 nm linewidth, enabling rapid Stokes stepping imaging with a fixed pump beam and achieving video-rate SRS imaging (Ozeki et al., 2019). Another study integrated a parabolic Yb-fiber amplifier with a femtosecond dual-beam solid-state laser. The 1040-nm output of the femtosecond laser served as the seed, injected into a Yb-doped fiber parabolic amplifier pumped by a counter-propagating 976-nm high-power diode laser. Within about 4 m of the gain fiber, the seed evolved self-similarly into a linearly up-chirped parabolic pulse with a broadly expanded bandwidth, increasing from ∼6 nm to ∼50–70 nm, depending on the pump powers. This approach expanded the single-setting coverage to over 600 cm^-1 with a ∼10 cm^-1 spectral resolution, enabling full-fingerprint hyperspectral imaging at microsecond pixel dwell times (Figueroa et al., 2018). Takahashi et al. introduced a hybrid system that synchronized a fiber-OPO (FOPO) and a fiber laser to deliver a broadband Stokes beam. By pairing this with a 3.6-ps, 790-nm pump from a Ti:sapphire laser, hyperspectral SRS imaging with 2000 cm^-1 coverage was achieved, expanding the accessible spectral window (Takahashi et al., 2025).

All-fiber laser sources aim for ultimate compactness, portability, and alignment-free operation. Freudiger and co-workers developed a practical all-fiber SRS system that used a single Er-doped fiber oscillator to seed two fiber amplifiers. The oscillator output was split into two optically synchronized amplifier arms. One arm was spectrally filtered and frequency-doubled in a PPLN crystal to generate a narrowband pump beam near 790 nm. The other arm was sent through a short section of highly nonlinear fiber to shift its spectrum toward 1 µm and then amplified in a Yb-doped fiber to produce a narrowband Stokes beam tunable from 1,015 to 1,045 nm. This tuning range covered Raman shifts around 2,800-3,100 cm^-1, which are well suited for imaging lipids and proteins in the CH high-wavenumber band but did not extend into the fingerprint region. A major technical improvement in this work was the use of a high-speed voltage-subtraction balanced detector, which effectively removed the excess high-frequency noise inherent in fiber amplifiers. The all-fiber picosecond laser system achieved shot-noise-limited sensitivity at video-rate imaging speeds, providing a compact, stable, and alignment-free SRS light source optimized for CH-band imaging without the need for bulky OPOs (Freudiger et al., 2014).

A different all-fiber strategy uses a FOPO pumped by and optically synchronized to a mode-locked 7-ps Yb-fiber master oscillator. The key is dispersive matching with a chirped fiber Bragg grating, which equalizes group delay, so the FOPO ring cavity stays in step with the Yb-fiber oscillator across a wide tuning range without a mechanical delay line. This source delivers two synchronized picosecond pulse trains, with a widely tunable output (∼780–980 nm) from the FOPO used as the pump and a companion tunable branch (∼1,025–1,055 nm) from the Yb-fiber chain used as the Stokes. By electronically sweeping the FOPO filter, the wavelength difference between the two arms maps directly to the desired Raman shifts, enabling millisecond switching among the high-wavenumber, silent, and fingerprint regions. Together, fast electronic tuning, all-fiber optical synchronization, and a fully spliced layout make this architecture a compelling platform for portable SRS imaging (Brinkmann et al., 2019). Ni et al. characterized the relative intensity noise (RIN) of a FOPO-based all-fiber laser system and demonstrated multiwindow SRS imaging of CH and CD bonds in biological samples with improved sensitivity using auto-balanced detection (Ni et al., 2021).

Future laser development for SRS microscopy will likely focus on closing the gap between laboratory performance and field-ready usability. The convergence of solid-state precision and fiber-based robustness is enabling sources that are compact, electronically tunable, and spectrally agile. Continued advances in low-noise fiber amplifiers, fast electro-optic tuning elements, and self-synchronizing architectures are expected to expand spectral coverage from the fingerprint to the CH-stretch region within fully sealed fiber systems. Integration of all-fiber sources with miniaturized detection and scanning modules may lead to turnkey SRS microscopes suitable for clinical and intraoperative use. Ultimately, the trend toward electronically controlled, alignment-free, and maintenance-light designs points to a new generation of portable, hyperspectral SRS instruments bridging research and clinical translation.

4.3 Advances and challenges in CRS endoscopy

Coherent Raman scattering (CRS) endoscopy must deliver two synchronized ultrafast laser beams for excitation and collect weak nonlinear signals through optical fibers. The first challenge is parasitic nonlinear light generation in solid-core silica fibers during fiber delivery of the laser beams. When the pump and Stokes beams co-propagate in the fibers, multiple nonlinear processes, including four-wave mixing (FWM), cross-phase modulation (XPM), and SRS, can create background signals that contaminate CRS signals from the sample. Early probe designs characterized this background and motivated separate delivery and collection paths with distal filtering to keep fiber-generated light out of the field of view (Balu et al., 2010). In one study, a custom double-clad fiber (DCF) with a large-mode-area core and a micro-fabricated distal filter achieved >10 dB suppression of fiber FWM in a compact scanning probe head (Deladurantaye et al., 2014). A second route is to suppress the nonlinearity at the source by confining excitation beams to air. In another study, double-clad hollow-core photonic-crystal fiber (PCF) was used to co-deliver the pump and Stokes beams in an air core while routing the return signal in a high-NA outer cladding, which eliminated in-fiber CARS and SRS signal generation and enabled a practical endoscope scheme (Brustlein et al., 2011).

Lombardini and co-workers built a flexible fiber CARS endoscope using a Kagome-lattice hollow-core double-clad fiber to deliver the femtosecond pump and Stokes beams in the core without in-fiber background. The CARS signal was back-collected through the outer silica cladding. Imaging was achieved by resonant spiral scanning of the fiber head (Lee et al., 2010). As shown in Figures 6A,B, this probe exhibited a 4.2-mm outer diameter and a 71-mm rigid section and achieved a ∼320-µm field of view at about 0.8 frames per second for CARS and SHG imaging (Lombardini et al., 2018). The high-quality imaging performance of this CARS endoscope was demonstrated by clear visualization of lipid and collagen content in fresh human colon tissues (Figures 6C,D). In their later work, an extended CARS probe design using a double-clad antiresonant hollow-core fiber functionalized with a gradient-index (GRIN) micro-objective lens was demonstrated, yielding a 200-µm field of view with multimodal nonlinear contrasts along with CARS (Septier et al., 2022).

The second challenge in CRS endoscopy is dispersion and timing of the two-color ultrafast pulses delivered in fibers. Long delivery paths introduce group-velocity walk-off, chirping, and modal effects that reduce temporal overlap at focus and weaken the nonlinear response. Pulse preserving delivery in hollow-core double-clad fibers, together with compact micro-optics at the tip, has been shown to maintain sub-micron resolution and clean contrast in a flexible CRS endoscope (Lombardini et al., 2018). Designs that combine custom endomicroscopy objectives with specialty fibers further address synchronization and aberration correction in a very small distal head (Pshenay-Severin et al., 2021).

The third challenge is the restricted epi-detection geometry. Phase-matched CARS signal is strongly emitted in the forward direction and is then back-scattered by heterogeneity of the tissue samples, so endoscopy designs must collect the back-scattered photons efficiently. Early fiber probes documented this constraint and adopted large-NA multimode collection paths to recover the useful signal (Balu et al., 2010). Handheld and rigid probe systems for potential intraoperative use reinforced the need for strong epi-collection with custom miniature objectives (Chen et al., 2015; Zirak et al., 2018).

These physics constraints cascade into scanning and packaging tradeoffs for CRS endoscopy designs. Distal scanners such as resonant fiber tips and miniature micro-electro-mechanical systems (MEMS) enable rapid imaging but add moving parts and wiring at the tip. This was shown both in a scanning-fiber SRS endoscope and in a compact multimodal intraoperative probe (Saar et al., 2011b; Chen et al., 2015). Wavefront-shaped multimode-fiber (MMF) probes can push the probe diameter down to a single 125-µm GRIN fiber, and Trägårdh et al. showed that holographically controlled MMFs can generate scannable diffraction-limited foci for label-free CARS imaging. However, this approach requires full transmission-matrix calibration and is highly bend-sensitive, so robust operation is limited to essentially static, well-controlled geometries (Trägårdh et al., 2019). More recently, Bae et al. adopted multicore imaging fiber bundles with proximal scanning, trading the extreme thinness of MMFs for a somewhat larger but mechanically robust all-optical probe head that preserves the spatial map without powered components at the distal tip. Although this design inherently yields pixelated images, straightforward depixelation and normalization workflows recover high-quality CARS, TPEF, and SHG tissue contrast. Together, these improvements outline a clearer path toward clinically practical nonlinear endoscopy (Bae et al., 2023).

Furthermore, clinical constraints set the practical envelope for CRS endoscopy. The outer diameter, usable rigid length, sterilizability, and the need to avoid high voltage and moving parts at the tip drive optical and mechanical design options. Compared to fiber probes, rigid CARS endoscopes illustrate the opposite end of the design space, where mechanical stability and imaging performance are prioritized over the probe diameter. Zirak et al. developed a slender 2.2-mm-diameter, 187-mm-long rigid CARS endoscope that delivers ∼750-nm spatial resolution and a usable field of view of ∼250 µm using custom GRIN optics and galvanometer scanning, all packaged in a shaft size compatible with standard rigid surgical endoscopes (Zirak et al., 2018). Hirose et al. pursued a larger 12-mm-diameter, 270-mm-long rigid CARS endoscope tailored for robot-assisted pelvic surgery, achieving a substantially larger field of view (∼650 µm) with low distortion and uniform CARS response suitable for intraoperative nerve visualization (Hirose et al., 2018). These systems demonstrate that rigid architectures can support high-NA focusing, high resolution, and stable epi-detection, provided that the required outer diameter aligns with the surgical application. In contrast, ultracompact fiber-scanning platforms extend CRS endoscopy into smaller, more flexible form factors by integrating scanning and collection directly at the distal fiber tip. Saar et al. introduced the miniature spiral-scanning fiber endoscope that delivered CARS and SRS contrast through a piezo-driven fiber cantilever and a sub-2-mm GRIN objective, enabling rapid imaging of tissues (Saar et al., 2011b). More recently, Pshenay-Severin et al. advanced this approach with a custom double-core double-clad fiber and a 2.4-mm-diameter resonant fiber-scanner head that provides background-free ps-pulse delivery, high-NA focusing, and robust multimodal CARS/SHG/TPEF imaging at sub-micron resolution in a compact rigid section suitable for clinical integration (Pshenay-Severin et al., 2021).

For SRS endoscopy, there is an additional detection burden that the CARS modality does not have. The SRS signal to be detected is a tiny intensity modulation on either the pump beam or the Stokes beam that must be measured with high fidelity in the presence of laser amplitude noise and fiber-induced artifacts. This makes any parasitic modulation from XPM, FWM, or SRS in the delivery path especially harmful. A background-free solution was shown by transporting the pump and Stokes beams temporally separated through the fiber and re-synchronizing the pulses at the tip with a highly dispersive element. This handheld implementation recovered SRS signals that were about two hundred times weaker than the background induced by the fiber (Liao et al., 2018). The use of double-clad hollow-core fibers is a highly promising approach to minimize nonlinear signal generation in the fibers (Brustlein et al., 2011). Endoscopic SRS also benefits from optimized epi-collection and custom objectives matched to a scanning fiber tip, as demonstrated in a fiber-scanner endoscope that produced chemically selective images at several frames per second (Saar et al., 2011b). In summary, to implement CRS endoscopy including both CARS and SRS, the major obstacles are fiber-generated background, dispersion and synchronization in two-color pulse delivery, and efficient epi-collection within a tight mechanical envelope. More efforts are expected to transform CRS endoscopy for real clinical use.

4.4 Quantum-based approaches in Raman and SRS spectroscopy and microscopy

4.4.1 Historical developments of quantum Raman and CRS spectroscopy

Quantum theory of Raman scattering has developed over a century, beginning with the matrix-mechanical treatment of light scattering by Kramers and Heisenberg (Kramers and Heisenberg, 1925). Dirac’s quantization of the electromagnetic field then led to the Kramers–Heisenberg–Dirac (KHD) expression, which remains the standard starting point for Raman and nonlinear Raman cross sections (Dirac, 1927). Heitler’s systematic nonrelativistic quantum electrodynamics (QED) formalism turned these concepts into a practical computational scheme for Rayleigh, Raman, and related scattering processes (Heitler, 1936). Building on this foundation, Louisell, Gordon, and co-workers developed the quantum theory of parametric amplification and frequency conversion, providing the operator framework and noise concepts that later carried directly into coherent Raman gain and fluctuation analysis (Louisell et al., 1961; Glauber, 1963).

In the 1960s, quantum optics supplied the tools needed to treat coherence and noise in ways that could be applied to Raman processes. Glauber introduced the modern theory of optical coherence, relating photon statistics and correlation functions to quantized fields and explaining how spontaneous Stokes emission acquires thermal statistics that Raman gain can amplify (Glauber, 1963). Lax formulated the quantum Langevin approach for open systems, justifying the introduction of operator noise sources into Maxwell–Bloch equations and ensuring consistency with microscopic commutators and fluctuation–dissipation relations (Lax, 1966). Caves then derived the quantum limits for linear amplification and detection, showing that any phase-insensitive amplifier must add at least one-half quantum of noise and establishing quadrature noise bounds that still set performance targets for Raman and SRS detection chains (Caves, 1982).

Fully quantized theories of Raman and coherent Raman processes emerged in parallel with these developments. Walls constructed a quantum theory of the Raman effect in which Stokes and anti-Stokes fields and molecular levels were treated without semiclassical linearization, demonstrating that Raman gain is initiated by vacuum fluctuations and that exact solutions are possible for few-level systems (Walls, 1970; 1971). Bjarnason and colleagues connected coherent Raman line shapes in gases and liquids to microscopic correlation functions, an early step toward the response-function description of nonlinear Raman spectroscopy (Bjarnason et al., 1979). Mukamel then formulated two-photon and four-wave-mixing (FWM) signals, including coherent Raman scattering (CRS), in terms of multi-point dipole correlation functions, unifying microscopic quantum dynamics with macroscopic third-order response functions that underpin modern treatments of Raman, CARS, and SRS (Mukamel, 1983).

Through the 1980s, a series of theoretical and experimental studies established SRS as a concrete realization of vacuum-noise-seeded amplification with nontrivial spatial and temporal coherence. Raymer and Mostowski developed the first space-dependent operator Maxwell–Bloch model of SRS, showing explicitly how the Stokes field is initiated by vacuum fluctuations, how it is amplified along the medium, and why its intensity is insensitive to pump phase noise (Raymer and Mostowski, 1981). Walmsley and Raymer then measured single-shot Stokes pulse energies and observed the predicted negative-exponential statistics characteristic of linear gain driven by quantum vacuum fluctuations, providing direct experimental evidence for quantum initiation of SRS (Walmsley and Raymer, 1983). Fabricius and co-workers independently confirmed these predictions in transient SRS by showing that large shot-to-shot fluctuations occur only when pump pulses are shorter than the molecular dephasing time, again matching the quantum-noise picture (Fabricius et al., 1984).

Spatial and temporal coherence in SRS were clarified by successive 3D Maxwell–Bloch treatments and coherent-mode analysis. Mostowski and Sobolewska extended quantum SRS theory to three dimensions, showing how Fresnel number and dephasing determine the number of amplified spatial modes and the evolution from large exponential intensity fluctuations in single-mode operation to narrower statistics in multimode regimes (Mostowski and Sobolewska, 1984). Raymer and co-workers introduced the concept of coherence modes that decompose the Stokes field into statistically independent spatiotemporal modes and related their weighting to the observed distribution of Stokes pulse energies (Raymer et al., 1985). Strauss and colleagues analyzed SRS for arbitrary pump profiles and described how transverse coherence areas grow during amplification, while Raymer and co-workers later developed a temporal coherent-mode description that links the complexity of single-shot Stokes pulse shapes to the excitation of higher-order temporal modes (Strauss et al., 1988; Raymer et al., 1989).

Taken together, the work from 1925 to 1990 established the core language for quantum Raman and CRS spectroscopy, explained how vacuum noise seeds Raman gain, how spatial and temporal coherence emerge in SRS, and how amplifier dynamics and detection-noise limits shape measurable signals (Raymer and Walmsley, 1990). Modern quantum treatments of Raman, CARS, and SRS are direct descendants of these foundations.

4.4.2 Recent developments of quantum SRS spectroscopy and microscopy

Over the past decade, quantum ideas have moved from theory to practice in Raman and SRS spectroscopy and microscopy. The goal of quantum-enhanced Raman/SRS is clear: beat shot-noise limits at low photon budgets, stay within photodamage ceilings, and gain new selectivity in time and frequency. Three thrusts now define the field. First, squeezed and correlated light lowers measurement noise at fixed power, so weak vibrational signals can be revealed without raising dose. The main questions are how much squeezing survives the optics and sample, how to run at the powers, bandwidths, and pixel rates used in modern microscopes, and how to compare gains fairly in SNR per photon. Second, entangled-photon excitation and interferometric readout open access to molecular response functions with joint time-frequency resolution that classical pulses cannot reach. Here the focus is on pathway selectivity, ultrafast dynamics, background control, and compatibility with condensed-phase samples. Third, QED modeling turns relative rules of thumb into absolute predictions by linking detected counts to molecular properties and instrument settings. Together, these directions shift the emphasis from foundational theory to applied sensitivity, speed, and spectral specificity in both spectrometers and microscopes.

4.4.2.1 QE-SRS with entangled photons

The recent use of entangled photons in nonlinear Raman spectroscopy begins with the theoretical framework of interferometric femtosecond SRS, which showed that time-frequency-entangled biphotons combined with coincidence detection can sharpen resolution and isolate specific matter pathways. In this scheme, one photon of a broadband entangled pair interacts with the sample together with a narrowband classical field to drive the Raman process, while its partner is sent to a reference arm. The photon-coincidence readout then separates the Raman gain and loss components and suppresses background signals, creating a detection window that classical probes with the same bandwidth cannot provide. This work introduced the key idea that quantum correlations can relax the usual trade-offs between temporal and spectral resolution in SRS and that interferometric pathway selection offers a powerful design tool for advanced spectroscopy (Dorfman et al., 2014). Zhang and co-workers developed a time-frequency-resolved coherent Raman framework that uses entangled photons to probe ultrafast electronic coherences in condensed-phase systems. This study demonstrated that quantum entanglement relaxes the usual trade-off between temporal and spectral resolution and connected the measured signal directly to well-defined coherence pathways in realistic vibronic models (Zhang et al., 2022). Fan et al. proposed an ultrafast SRS scheme that embeds a Hong-Ou-Mandel (HOM) interferometer into the measurement and uses entangled photons to define a new effective time-frequency scale. In their theory, the entangled probes select molecular correlation functions that are difficult to access with classical light and provide direct sensitivity to exciton populations and coherences in condensed-phase systems. They showed that coincidence-based gating and the interferometer phase offer extra control over pathway selectivity and background suppression, pointing to practical routes toward quantum-enhanced SRS measurements at low dose. This work turns the earlier entangled-photon concepts into a concrete interferometric SRS layout for ultrafast experiments and clarifies where entanglement can improve both sensitivity and selectivity beyond shot-noise-limited classical approaches (Fan et al., 2024).

4.4.2.2 QE-SRS with squeezed light

Andrade et al. reported the proof-of-principle quantum-enhanced continuous-wave SRS, showing that squeezed light can lift Raman sensitivity beyond the shot-noise-limit. They injected a bright amplitude-squeezed probe light into a continuous-wave SRS spectrometer and measured C-H stretch bands in polymers, observing about 3.60 dB improvement in SNR at fixed optical power. The result established squeezed-light illumination as a practical path to higher Raman sensitivity without increasing sample dose (de Andrade et al., 2020). Casacio and co-workers incorporated quantum-correlated illumination into an SRS microscope and achieved sub-shot-noise detection of molecular vibrations in live cells. The correlated photon pairs improved SNR by 35% at the same power and increased concentration sensitivity by 14%, revealing intracellular structures not visible under classical illumination. This work demonstrates that quantum correlations can provide practical performance gains in real-time Raman microscopy (Casacio et al., 2021).

The previous demonstrations of QE-SRS used weak optical power. Xu and colleagues demonstrated a quantum-enhanced balanced detection (QE-BD) scheme using high-power optical pulses (Xu et al., 2022a). The same group then adapted this approach to microscopy in a high-power regime. Their QE-BD-SRS microscope operated above 30 mW, achieved a 2.89 dB noise reduction relative to classical balanced detection. This work confirmed that QE-BD-SRS can work in the high-power regime, and helped pave the way for achieving higher sensitivity beyond the state-of-the-art SRS microscope using high laser powers (Xu et al., 2022b). Later, the same group introduced a dual-polarization architecture to extend QE-SRS to simultaneous parallel and orthogonal channels, enabling ultrasensitive measurements of vibrational-mode symmetry while remaining compatible with higher optical powers (Xu Z. C. et al., 2023). Min and Gao developed a first-principles QED theory of SRS that provides a closed-form, molecule-specific SRS cross section and an explicit link between SRS and spontaneous Raman cross sections through a single vacuum-field factor. This framework enables absolute predictions of SRS signal and SNR, quantifies the SRS enhancement factor, and offers a practical, instrument-independent baseline for modeling SRS performance in microscopy (Min and Gao, 2024).

Recent work shows a credible path to quantum advantages in real spectroscopy and microscopy instruments (Silberberg, 2009; Dorfman et al., 2016; Gao and Min, 2025; Xu et al., 2025). Squeezed light now improves spectroscopy and live-cell imaging at constant dose. Entangled-photon schemes promise new contrast mechanisms and femtosecond-scale selectivity. Modern QED treatments provide an absolute yardstick for signal and noise. The next steps are practical: preserving quantum correlations through lossy optics, pushing to high power and wide bandwidth, simplifying readouts, and standardizing benchmarks in sensitivity per photon, dose, and information gained. If these engineering goals are met, quantum Raman and SRS can become routine tools for low-dose chemical microscopy and ultrafast molecular spectroscopy.

5 Biomedical applications of SRS microscopy

This section reviews the major biomedical applications of SRS microscopy and highlights its evolution into a versatile platform for chemical imaging in living systems. At the single-cell level, SRS provides label-free metabolic contrast and enables quantitative visualization of lipid, protein, glucose, and other biomolecular processes with high spatial and temporal resolution. At the tissue level, SRS delivers diagnostic chemical contrast, and stimulated Raman histology (SRH) offers rapid, stain-free virtual pathology for intraoperative assessment, most notably in brain tumor surgery. Beyond native biomolecules, vibrational Raman tags enable multiplexed imaging of drugs, organelles, and specific metabolic pathways. SRS also extends to non-biological materials, supporting detection and characterization of microplastic and nanoplastic particles and other synthetic particulates. Applications in microbiology and plant biology further demonstrate how SRS reveals metabolic states and chemical heterogeneity in bacteria, fungi, and plant tissues without labels or destructive processing.

5.1 Label-free SRS imaging of lipids in cells for metabolic studies

SRS microscopy provides intrinsic chemical contrast for fatty acids and neutral lipids because the high-wavenumber C-H vibrations (2,800–3,100 cm^-1) are strong and chemically informative. As illustrated in Figure 7A, the SRS spectrum of oleic acid (OA) shows a dominant CH₂ peak at 2,854 cm^-1, and a carbon-to-carbon double bond (C=C) peak at 3,015 cm^-1. Figure 7B shows two-color images of U87 glioma cells acquired at these bands after 24 h of OA treatment at 0, 50, and 100 µM. The images reveal bright lipid droplets at 2,854 cm^-1 together with a measurable rise in the 3,015 cm^-1 signal, which marks incorporation of unsaturated acyl chains. Using these paired bands, one can directly observe and quantify rapid fatty acid uptake by cancer cells. The intensity ratio of the 3,015 cm^-1 to 2,854 cm^-1 bands serves as a reliable proxy for lipid unsaturation, establishing a straightforward, label-free method to assess lipid mass and composition in living cells (Yuan et al., 2021). This optical view connects directly to lipid droplet (LD) biology: lipid droplets are dynamic organelles that sequester neutral lipids such as triacylglycerols (TAG) and cholesteryl esters (CE), exchange fatty acids with membranes, and interface with pathways for energy production, membrane synthesis, redox balance, and signaling. The number, size, and composition of LDs are highly responsive to substrate supply and enzymatic control, and these responses are rewired in disease, including cancer, to support proliferation, stress tolerance, and therapy resistance. By coupling chemical specificity with high spatial and temporal resolution, SRS links fatty-acid uptake to lipid-droplet biogenesis and remodeling in situ, providing the foundation for mechanistic studies of lipid metabolism in health and disease.

Fu and co-workers used hyperspectral SRS to characterize neutral lipids in living cells by separating cholesteryl esters and triacylglycerols in the CH region and mapping their distributions within lipid droplets in cells and in vivo. This study established hyperspectral SRS as a quantitative, label-free approach for probing lipid storage and remodeling in physiology and disease (Fu et al., 2014a). Zhang et al. used label-free SRS microscopy to quantify lipid metabolism by measuring the motion of individual lipid droplets, showing that dynamic droplet behavior provides a sensitive indicator of metabolic state and reveals shifts in lipogenesis under different cellular conditions (Zhang et al., 2017b). Yuan et al. used label-free SRS microscopy to examine how exogenous fatty acids influence lipotoxicity in glioblastoma cells. They showed that uptake of palmitic acid, oleic acid, and eicosapentaenoic acid drives substantial lipid droplet accumulation, and that depleting these droplets sensitizes cells to fatty acid-induced stress, markedly reducing survival. This work links fatty acid handling, neutral-lipid storage capacity, and cell fate in glioblastoma (Yuan et al., 2021; Yuan and Lu, 2022). Figueroa et al. used a ratiometric SRS approach that compares macromolecular signals to the water band to quantify intracellular dry mass density in living cells. This method provides an optical measure of cell density independent of size or morphology and revealed tightly regulated density across cell types, rapid changes during osmotic stress, and spatial variation within 3D tumor spheroids (Figueroa et al., 2022).

Label-free SRS imaging of the C-H band can read out lipid composition in live cells and convert these measurements into metabolic fingerprints. Hislop et al. combined hyperspectral SRS with k-means cluster analysis to segment lipid droplets and compare their composition across malignant and non-malignant prostate cells. Their work demonstrated SRS as a practical readout for drug-metabolism interactions in living cells (Hislop et al., 2022). Rensonnet et al. advanced compositional profiling by combining hyperspectral SRS with spectral phasor analysis across both the high-wavenumber and fingerprint regions. This study provided a convenient label-free readout of lipid unsaturation and esterification in cells (Rensonnet et al., 2024). Szkalisity et al. developed a quantitative approach to separate cholesteryl esters and triacylglycerols within individual lipid droplets using hyperspectral SRS and reference-based spectral decomposition (Szkalisity et al., 2025). Beyond lipid metabolism, label-free SRS microscopy has been applied to other key aspects of cell physiology. Direct imaging of nucleic acids was demonstrated in metaphase cells with condensed chromosomes (Zhang et al., 2012). Hyperspectral SRS combined with spectral phasor analysis has enabled label-free spatiotemporal mapping of mitosis (Hislop et al., 2023). Ratiometric analysis of protein and lipid SRS signals has also allowed quantitative classification of apoptosis and necrosis in live CHO cells, with single-cell chemotypic and phenotypic signatures that generalize across stressors and cell types (Mahapatra et al., 2025).

5.2 SRS imaging with Raman tags for metabolic studies

Yamakoshi and colleagues established the concept of bioorthogonal alkyne-based (C≡C) Raman tagging. They showed that a simple alkyne group produces a sharp, background-free Raman peak in the cellular silent region, creating a clean optical channel for visualizing small molecules with minimal interference from endogenous signals. They found that aromatic alkynes and conjugated diynes generate stronger Raman intensities while maintaining spectral isolation, providing clear design rules for high-performance tags. This study laid the groundwork for the use of Raman tags in metabolic imaging (Yamakoshi et al., 2012). Wei and colleagues brought alkyne Raman tags into the SRS microscopy platform and established a broadly applicable method for imaging biomolecules in live cells and animals. Using a panel of alkyne-labeled precursors, they visualized de novo synthesis of DNA, RNA, proteins, phospholipids, and triglycerides, and showed that metabolic dynamics such as RNA turnover and lipid droplet formation can be followed in real time. This study defined the practical framework for SRS imaging of bioorthogonal Raman tags for metabolic studies (Wei et al., 2014).

Applications of SRS imaging with various types of Raman tags focused on using specific tagged precursors to read out central metabolic pathways in living systems. Glucose analogs labeled with small alkynes allowed direct visualization of glucose uptake in cells and tissues and revealed heterogeneous energy demands in tumors and brain tissue (Hu et al., 2015). Phenyl-diyne cholesterol and highly deuterated cholesterol provided robust probes for cholesterol storage and trafficking in cells, C. elegans, and disease models of lysosomal cholesterol accumulation (Lee et al., 2015; Alfonso-Garcia et al., 2016). In parallel, deuterated amino acids established a general platform to map protein synthesis, degradation, and pulse-chase turnover in cultured cells and whole animals (Wei et al., 2015). These early studies showed that small vibrational tags can convert otherwise invisible fluxes of glucose, cholesterol, and proteins into spatially and temporally resolved SRS images in intact biological systems.

SRS imaging with deuterium (²H or D) tags broadened into organelle-resolved metabolism, neurodegeneration, and microbial and nanoparticle pharmacology. A fluorescence and SRS dual-mode probe enabled simultaneous fluorescence and SRS readout of mitochondria and allowed noninvasive quantification of local probe concentration in the mitochondrial matrix (Li et al., 2017). Deuterium oxide (D₂O) probing with SRS (DO-SRS) generalized isotope labeling to a universal, minimally perturbing reporter that could separate lipid, protein, and DNA synthesis in live animals and reveal divergent metabolic dynamics across tissues and tumors (Shi et al., 2018). Tagged antibiotics and mitochondria-targeting cations were followed in real time to measure their penetration, binding, and pharmacokinetics in biofilms and live cells (Bae et al., 2019; 2020). Lysosome-targeted bisarylbutadiyne probes and quantitative SRS readouts enabled photostable tracking of lysosomes and clarified how lysosomotropic compounds reshape lysosomal lipid metabolism (Ding et al., 2019; Yuan et al., 2023). In live animals, vibrationally tagged fatty acids supported quantitative analysis of lipid synthesis and lipolysis in C. elegans (Li et al., 2019) and, with deuterated glucose, revealed subcellular glycogen enrichment and its link to stress resistance in melanoma cells (Lee et al., 2020). Meanwhile, deuterated glutamine labeling made it possible to image native polyQ aggregates without fluorescent fusion tags, quantify their composition, and monitor structural changes under proteostasis stress (Miao and Wei, 2020). Alkyne- or deuterium-modified polymers and nanoparticles extended these strategies beyond endogenous metabolism to drug carriers, allowing direct visualization of poly(lactic-co-glycolic acid) (PLGA) nanoparticles and their localization in microglia ex vivo (Vanden-Hehir et al., 2019).

More recent work has pushed bioorthogonal SRS imaging toward complex in vivo models and translational questions in pharmacology and cancer biology. Deuterated amino acids like methionine have been imaged in fly tissues to expose previously unappreciated cell-to-cell heterogeneity in amino acid uptake at subcellular resolution (Spratt et al., 2023). Solid lipid nanoparticles and other lipid-based carriers have been visualized at near single-particle sensitivity using deuterated tags, opening a route to detailed in vivo pharmacology of clinically relevant nanocarriers (Lang et al., 2023). Bisarylbutadiyne (BADY) tagging probes have been used to track nuclear and lysosomal distributions of drug analogs and to reveal how lysosomal sequestration drives lipid accumulation and potential resistance mechanisms in cancer cells (Yuan et al., 2023; Kawaguchi et al., 2024). In thyroid cancer models, deuterated tags and SRS quantification of triglyceride dynamics helped identify diacylglycerol O-acyltransferase 2 (DGAT2) as a target that links lipid storage to iodine uptake, suggesting a strategy to resensitize papillary thyroid carcinoma to radioiodine therapy (Zeng et al., 2025). Together, SRS imaging with Raman tags has been used for studying metabolism, drug action, and nanomedicine across a wide range of biological systems. For readers seeking a more comprehensive overview of bioorthogonal Raman tag chemistry, imaging strategies, and applications, several recent review articles provide extended discussions that complement this brief summary (Wei et al., 2016; Tipping et al., 2017; Hu et al., 2019; Li and Yue, 2025).

SRS imaging has begun to inform aging and senescence biology, often through lipid and protein metabolism studies. SRS imaging with deuterated and alkyne-tagged fatty acids enabled quantitative measurements of lipid synthesis and lipolysis in living animals such as C. elegans. This approach provides a direct way to track lipid turnover in a whole organism (Li et al., 2019). Using Drosophila models, DO-SRS imaging quantified lipid turnover in lipid droplets at different ages. The same study examined diet composition and insulin signaling to connect lipid metabolic dynamics with aging and lifespan regulation (Li et al., 2022). In the ovary, DO-SRS imaging showed that de novo protein synthesis and lipogenesis decrease sharply in old flies. The decline was most evident in the germline stem cell niche. Unsaturated lipids accumulated in nurse cells and oocytes during aging (Li et al., 2021a). In the brain, SRS imaging revealed reduced lipid signals and lower lipid turnover with age. Dietary restriction or reduced insulin signaling increased lipid turnover in old flies and shifted carbon utilization toward acetate for lipogenesis (Li et al., 2023). CRS imaging tools have also been applied to study cellular senescence, which is closely linked to tissue aging and cancer biology. Multimodal nonlinear optical imaging combined with SRS captured metabolic remodeling during therapy-induced senescence in cancer cells. These imaging results enabled automated classification of senescent cells (Sorrentino et al., 2023). Multiplex CARS microspectroscopy provided a complementary marker by detecting nucleolar protein aggregation through amide I spectral shifts toward β-sheet rich structures (Ishibashi et al., 2024).

5.3 Stimulated Raman histology (SRH) for cancer diagnosis

5.3.1 SRH for brain tumor diagnosis and surgical guidance

The concept of stimulated Raman pathology originated from early work showing that two-color SRS imaging, using lipid and protein contrasts, can produce pathology-like images with hematoxylin-and-eosin (H&E) staining characteristics (Freudiger et al., 2012). Ji and colleagues then demonstrated its diagnostic potential in brain tumors by showing that this two-color lipid/protein SRS approach rapidly distinguishes tumor from non-neoplastic brain in a xenograft glioblastoma model and in fresh human specimens (Ji et al., 2013). Subsequent work expanded these characterizations in human brain tissue specimens. Lu and co-workers revealed substantial pathological heterogeneity of brain tumors accessible to SRS imaging. This work profiled a broad range of brain tumors and correlated thousands of SRS fields with matched H&E slides, showing that SRS captures key diagnostic hallmarks and reveals additional features such as lipid droplets in glioma cells and disrupted myelinated fibers in infiltrated white matter (Lu et al., 2016). Orringer and colleagues translated the technology to the operating room using a portable fiber-laser microscope and introduced the term stimulated Raman histology (SRH) to describe virtual H&E-like images generated from two-color SRS images (Orringer et al., 2017). SRH was then applied to pediatric brain tumors, producing fresh-tissue images with preserved cytologic and architectural detail and enabling rapid intraoperative diagnosis that aligned closely with conventional histology (Hollon et al., 2018). Additional studies in skull-base tumors demonstrated similar diagnostic performance (Shin et al., 2019). Clinical workflow evaluations showed that SRH can be used intraoperatively across distinct regions of a tumor cavity, including cyst walls, solid nodules, and infiltrative edges. SRH findings aligned well with MRI-identified targets, as illustrated in Figures 8A,B (Neidert et al., 2022).

SRH has become increasingly quantitative through advances in computation and artificial intelligence (AI). A major step was the development of deep convolutional neural networks (CNNs) trained on large SRH datasets, which achieved near real-time intraoperative diagnoses that were noninferior to standard pathology in a prospective multicenter trial (Hollon et al., 2020). Further studies demonstrated the value of SRH at tumor margins, where quantitative analysis detected microscopic residual disease in infiltrative zones (Pekmezci et al., 2021). AI tools were also developed for automatic cell counting (Zhang et al., 2021) and automated classification of SRH fields into tumor, non-tumor, and low-quality categories (Reinecke et al., 2022). More recent work has extended SRH to large-scale AI systems and disease-specific decision support. Foundation models such as FastGlioma, pretrained on millions of SRH patches, generate continuous infiltration scores across whole-slide images and have shown fast and accurate detection of infiltrative glioma in multicenter testing (Kondepudi et al., 2025). Other systems focus on challenging differential diagnoses. RapidLymphoma applies self-supervised learning to distinguish primary CNS lymphoma from other lesions with accuracy comparable to or better than frozen sections (Reinecke et al., 2025), while transformer-based classifiers separate CNS lymphoma from glioblastoma on intraoperative SRH images, providing a practical triage tool when neuropathology resources are limited (Scheffler et al., 2025).

Despite the successful translation of SRH toward clinical use, it is important to note the limitation of the current label-free, two-color SRH. Because SRH relies primarily on lipid and protein contrasts, it does not selectively image DNA with nucleic-acid-specific contrast. The signals seen within cell nuclei in SRH images therefore arise mainly from proteins, and the resulting depiction of nuclear features can be less pronounced than the hematoxylin contrast in conventional H&E stained images. Future technical development aims at providing positive nucleic acid contrast, whether labeled or label-free, in order to overcome the strong lipid and protein background and meaningfully broaden SRH’s diagnostic utility. Taken together, advances ranging from early single-site demonstrations to multicenter trials, telepathology deployment, virtual staining, and AI foundation models demonstrate that SRH is a highly promising translational platform for brain tumor diagnosis and neurosurgical guidance, as well as for other tumor types. Further insights into SRH are available in a few recent review papers (Cicerone and Camp, 2017; Hollon and Orringer, 2021; Lee et al., 2021; Ma et al., 2024).

5.3.2 SRH for prostate cancer and breast cancer diagnosis

SRS microscopy and SRH have been evaluated as rapid, label-free approaches for prostate cancer diagnosis and intraoperative guidance. Early studies showed that SRH produces high-quality images of fresh prostate biopsy cores and that trained pathologists can identify prostate cancer with accuracy above 95% compared with H&E staining (Mannas et al., 2023). Deep-learning models trained on SRH images further demonstrated automated Gleason grading, achieving accuracies above 85% on core-needle biopsies and showing good agreement with expert pathologists (Ao et al., 2023). These results indicate that SRH captures the glandular and stromal patterns needed for clinical grading. SRH has also been explored for real-time margin assessment during prostate surgery. In partial gland ablation, SRH allowed detection of tumor near the treatment margin and informed adjustments to ablation plans, with accuracy near 98% for both human readers and AI classifiers (Mannas et al., 2025a). During radical prostatectomy, SRH interpretation identified positive surgical margins intraoperatively, suggesting potential to support nerve-sparing procedures while reducing residual tumor risk (Mannas et al., 2025b). Together, these studies show that SRH can aid prostate biopsy evaluation, automate Gleason grading, and guide surgical decision-making, although further optimization and clinical integration are still needed.

In breast cancer, SRS microscopy has been used for microcalcification analysis and rapid digital pathology. Hyperspectral SRS imaging of fresh breast biopsies revealed that the combined chemical and morphological signatures of calcium deposits can differentiate benign from malignant lesions with high precision, exceeding traditional mammography and bulk Raman spectroscopy (Yang et al., 2021). Subsequent work applied SRH to whole core-needle biopsies and paired the images with a weakly supervised multi-instance learning model, achieving roughly 95% accuracy in distinguishing benign from malignant tissue and highlighting intralesional heterogeneity through activation maps (Yang et al., 2023).

Recent advances in high-content SRH have expanded beyond morphology to include quantitative maps of saturated and unsaturated lipids, protein, extracellular matrix, and water in breast tissue. Using hyperspectral SRS with sparse unmixing and fast sampling on a compact fiber-laser platform, this approach can distinguish ducts, stroma, adipose tissue, necrosis, and vessels within the same acquisition (Ni et al., 2024). Collectively, these results demonstrate that SRS-based histology offers rapid evaluation of breast biopsies, chemically informed microcalcification assessment, and rich compositional information that may support future biomarker discovery. Continued technical refinement and validation will be required for breast SRH to reach the level of clinical maturity already demonstrated in brain tumor imaging.

5.3.3 SRS imaging of gastrointestinal and hepatobiliary tumor tissues

SRS and SRH have also been applied to gastrointestinal and hepatobiliary cancers to accelerate diagnosis in settings where biopsy and cytology are central but often slow. In gastric cancer, single-shot femtosecond SRS combined with U-Net-based reconstruction generated SRH images of fresh gastroscopic biopsies in about 1 minute, revealing histoarchitectural features consistent with standard pathology and enabling a convolutional neural network to classify cancer with accuracy above 96%. The same platform performed semantic segmentation of intratumor heterogeneity and assessed margins in endoscopic submucosal dissection specimens, demonstrating near real time histologic feedback during endoscopy (Liu et al., 2022). Complementary cytology-oriented approaches have paired single-color SRS imaging with virtual staining, where a CellGAN model produced standardized H&E-like images of peritoneal lavage specimens and achieved diagnostic performance comparable to conventional cytology while reducing staining variability and infrastructure needs (Fang et al., 2024). In hepatobiliary disease, transient SRS spectroscopy has enabled bile-based testing for cholangiocarcinoma by acquiring stable C-H stretching spectra from microliter volumes of patient bile and distinguishing malignant from benign conditions with high accuracy (Guo et al., 2025). Together, these studies show that SRS can support fast gastric biopsy evaluation, virtual cytology for peritoneal lavage, and minimally invasive bile spectroscopy for cholangiocarcinoma detection within workflows compatible with endoscopic and intraoperative care.

5.3.4 SRS imaging of head-and-neck, oral cavity, and lung tumor tissues

SRS and SRH have also been applied to head-and-neck, oral cavity, and lung cancers, where rapid intraoperative assessment is often limited by access to frozen-section pathology. In head-and-neck carcinoma, SRS imaging distinguished neoplastic from non-neoplastic tissue in fresh surgical specimens with high sensitivity and specificity, producing histology-like images without sectioning or staining and highlighting its potential for margin evaluation (Hoesli et al., 2017). SRH has been tested in oral squamous cell carcinoma, where fresh-tissue SRH images showed strong agreement with frozen-section H&E for identifying tumor and normal tissue and allowed sub-classification of non-neoplastic structures such as mucosa, muscle, and salivary gland tissue, suggesting utility for guiding resections and providing immediate digital review (Steybe et al., 2023). In the lung, SRH combined with convolutional neural networks (CNNs) supported rapid discrimination of malignant from normal lung tissue, achieving a high diagnostic accuracy on resection specimens, although performance was more limited for distinguishing primary lung cancers from metastases or for separating major histologic subtypes (Schröder et al., 2025). Collectively, these studies indicate that SRS and SRH can provide fast, label-free evaluation of tumors in the head-and-neck region, oral cavity, and lung, particularly for tumor-versus-normal tasks, while further work is needed to improve subtype classification and strengthen clinical integration.

5.4 SRS imaging of biological tissues for metabolic and physiological studies

5.4.1 Lipid metabolism in cancer tissues

SRS microscopy enables direct visualization of lipid composition in intact tissues, providing insights into metabolic alterations that accompany cancer progression. Early hyperspectral SRS study quantified cholesterol distributions in healthy and diseased tissues without labeling (Wang et al., 2013). Extended work performed lipid profiling to human cancers. In hepatocellular carcinoma, hyperspectral SRS revealed marked accumulation of saturated lipids in tumor regions, supported by elevated glyceryl tripalmitate levels, which suggested a saturated triglyceride signature of malignant transformation (Yan et al., 2018). In papillary thyroid carcinoma, multicolor SRS combined with confocal Raman spectroscopy showed strong triglyceride enrichment and higher lipid unsaturation in metastatic lymph nodes compared with primary tumors, correlating with increased expression of lipid uptake and transport genes (Zeng et al., 2024). These studies demonstrate that SRS-based lipid profiling can reveal tumor-specific metabolic programs and support biomarker discovery related to malignancy and metastatic potential.

5.4.2 Adipogenesis, retinoids, and developmental metabolism

SRS has been broadly applied to whole-organism models to investigate how metabolism changes during development and environmental stress. In zebrafish larvae, label-free SRS imaging visualized adipocyte formation under different diets and chemical exposures and revealed that both high-fat feeding and tributyltin markedly increase adipogenesis and alter lipid-regulatory gene expression (den Broeder et al., 2017). In C. elegans, hyperspectral SRS uncovered cytoplasmic retinoid stores and showed that retinoid levels correlate with fat accumulation, help preserve lipids during starvation, promote unsaturation under high-glucose stress, and ultimately enhance survival, pointing to a previously unrecognized role for retinoids in metabolic homeostasis (Chen et al., 2018). More recently, hyper-3D SRS enabled 3D mapping of lipid droplets in oocytes and early mouse embryos. This work revealed pronounced heterogeneity in droplet size and unsaturation even at the two-cell stage and identified abnormal profiles in a progeria model, suggesting that lipid organization may serve as a biomarker for oocyte and embryo quality (Huang et al., 2025). These studies highlight SRS as a tool for noninvasive tracking of developmental metabolism and organismal responses to environmental stress.

5.4.3 Non-neoplastic diseases and tissue microenvironments

SRS microscopy has also advanced understanding of tissue physiology and non-cancer disease. In one study, SRS imaging provided quantitative maps of protein, lipid, and water mass at single-cell resolution and revealed distinct compositional shifts in tissue models of Alzheimer’s pathology (Oh et al., 2022). In the skin, SRS visualized key steps in keratinocyte differentiation and identified structural defects associated with impaired barrier formation (Egawa et al., 2019). SRS imaging of the stratum corneum visualized water uptake and retention in human skin tissue and revealed heterogeneous hydration patterns along with chemical-treatment induced changes in epidermal barrier function (Mizuguchi et al., 2023; Ligorio et al., 2025). In joint disease, SRS imaging of synovial organoids captured monosodium urate crystal deposition and differential inflammatory responses across tissue layers, providing insight into gout pathophysiology (Chen et al., 2024). SRS has also detected silicone deposition in tissues surrounding breast implants and quantified silicone loss during histological preparation (Schmidt et al., 2025). Multimodal integration with SHG and TPF has created optical biopsy approaches for diabetic kidney disease, enabling simultaneous assessment of fibrosis, lipid metabolism, redox status, and tissue architecture (Fung et al., 2025). Collectively, these studies demonstrate the broad applicability of SRS for probing tissue composition and microenvironmental changes across diverse non-neoplastic conditions.

5.5 SRS imaging for drug delivery studies

SRS microscopy has been used to visualize the delivery, uptake, and intracellular behavior of therapeutic small molecules with label-free chemical specificity and subcellular resolution. Early work in living cells showed that hyperspectral SRS can directly image clinically important tyrosine kinase inhibitors and quantify their lysosomal sequestration, revealing large concentration gradients and drug-drug interactions that alter intracellular accumulation (Fu et al., 2014b). Further studies extended these capabilities to additional drug molecules. SRS imaging of the anticancer agent ponatinib captured its uptake at nanomolar levels and revealed lysosomal trapping during the development of drug resistance in leukemia models (Sepp et al., 2020). Real-time perfusion SRS imaging visualized the cellular dynamics of the alkyne-containing discoidin domain receptor 1 (DDR1) inhibitor 7RH, revealing heterogeneous uptake, repeated dosing behavior, and combined treatment responses in live cancer cells (Tipping et al., 2022). Three-band (C-D, C-H, and fingerprint regions) SRS imaging has also been used to compare drug uptake and response in 2D versus 3D culture systems, showing that drug tolerance in spheroids arises mainly from microenvironmental cues and restricted penetration rather than reduced intracellular accumulation (Xu et al., 2024).

SRS imaging has also been applied to drug delivery in tissues, where penetration barriers, heterogeneity, and formulation effects are critical. Initial work in skin demonstrated 3D tracking of topical formulations, distinguishing follicular and intercellular transport routes and visualizing drug precipitation during transdermal delivery (Saar et al., 2011a). Phase-modulated SRS enabled depth-resolved quantification of caffeine in reconstructed epidermis and revealed how formulation type modulates penetration profiles (Iguchi et al., 2022). Deep-learning-assisted SRS has been used to evaluate pharmacokinetics in skin tissues, where convolutional neural networks (CNNs) helped extract local drug flux, accumulation, and partition ratios across layers (Feizpour et al., 2021). An improved dual-modality SRS detection scheme reduced variance in forward-detected signals and improved the reliability of bioequivalence measurements for topical products, pointing toward regulatory-grade assessments of formulation performance (Tu et al., 2025).

Because fluorescent tagging often requires bulky chromophores that alter a drug molecule’s physicochemical properties and pharmacokinetics, SRS imaging of native vibrational bonds provides a crucial label-free alternative that preserves true dynamics and distribution. Overall, SRS has progressed from visualizing intracellular drug accumulation to quantifying tissue-level transport and pharmacokinetics, establishing a versatile platform for drug delivery studies in both cellular and tissue environments (Tipping et al., 2016; Zeng et al., 2021).

5.6 SRS imaging of fungi, bacteria, and plant samples

SRS microscopy has been applied broadly to microorganisms, enabling label-free analysis of metabolism, drug response, and engineered material formation. In microalgae, video-rate SRS revealed metabolic heterogeneity in live Euglena gracilis and showed that storage metabolites vary strongly across single cells under different culture conditions (Wakisaka et al., 2016). In fungal pathogens, fingerprint SRS combined with spectral unmixing identified abnormal ergosteryl ester accumulation as a metabolic hallmark of azole-resistant Candida albicans and showed that inhibiting sterol esterification resensitizes the fungus to fluconazole (Zhang et al., 2023). In bacteria-based living materials, SRS monitored phosphatase-catalyzed mineralization in engineered films, quantifying the evolving organic and inorganic components and linking spatially heterogeneous mineralization to mechanical performance (Qian C. et al., 2024). These studies highlight how SRS can probe microbial metabolism, drug resistance, and functional biomaterial formation without labels or destructive sample preparation.

SRS microscopy has also become a versatile tool in plant sciences for visualizing cell wall chemistry, cuticular structure, nutrient flow, and agrochemical penetration. Early work demonstrated real-time monitoring of biomass processing, providing contrast for cellulose, hemicellulose, and lignin during cell wall deconstruction (Saar et al., 2010). SRS imaging of plant cuticles then showed reduced pigment-related background relative to CARS and enabled chemically specific mapping of wax architecture across development and species (Littlejohn et al., 2015). Hyperspectral SRS has been used to distinguish overlapping wall components; selective removal and spectroscopic analysis identified Raman bands specific to xylan, enabling 2D and 3D imaging of hemicellulose in situ (Zeng et al., 2016). Engineered poplar with modified lignin was also characterized using hyperspectral SRS and AFM, revealing spectral shifts associated with ester-linked lignin and linking these distributions to improved enzymatic digestibility (Shen et al., 2019). Recent work demonstrated multicolor SRS imaging that discriminates up to six biomolecular components in leaves, roots, and liverwort tissues (Iino et al., 2021).

SRS imaging combined with deuterium labeling has enabled direct visualization of water and small-molecule transport in plants. D₂O tracing in Arabidopsis thaliana showed incorporation of deuterium into newly synthesized biomolecules during germination, revealing metabolic heterogeneity along axial and radial tissue structures (Bi et al., 2024). Similar isotope-enabled SRS mapping of seed coats visualized biopolymers such as pectin, tannin, and suberin and showed how coat composition influences water permeability, and the same approach tracked the penetration of a deuterated insecticide (Wang et al., 2025). At the biomass pretreatment stage, SRS imaging of poplar cell walls revealed that the spatial pattern of lignin removal, not just its bulk level, governs cellulase accessibility and cellulose digestibility across chemical pretreatments (Zhang et al., 2025). Collectively, these microbial and plant studies demonstrate that SRS can map structural polymers, metabolites, water flux, and agrochemical transport in situ, offering a powerful label-free platform for studying microbial physiology, plant biology, and biomass processing.

5.7 SRS imaging of non-biological materials

5.7.1 SRS imaging of pharmaceutical solids and drug tablets

SRS microscopy has been used to analyze drug tablets, solid-state formulations, and pharmaceutical excipients with rapid, label-free chemical contrast. Early work showed that SRS can map active ingredients and excipients in intact tablets much faster than confocal Raman mapping and with better chemical specificity than CARS (Slipchenko et al., 2010). Hyperspectral SRS was used to detect microscale chemical reactions in tablets, including salt disproportionation at low drug loading, providing insight into transformations that affect drug stability and dissolution (Figueroa et al., 2019). SRS imaging at a few key Raman bands also enabled rapid discrimination of drug polymorphs and excipients across millimeter-scale areas (Sarri et al., 2019). More recently, an epi-detected sparse spectral sampling SRS method allowed fast evaluation of tablet aging by reconstructing chemical maps from a limited set of Raman frequencies, supporting real-time quality control during storage (Wei et al., 2024). Multimodal SRS combined with sum-frequency generation (SFG) has further resolved multiple crystalline and amorphous forms in pharmaceutical solids, including distinct forms of lactose at submicron resolution (Harju et al., 2025). Together, these studies show that SRS enables rapid characterization of tablet composition, polymorphs, and stability.

5.7.2 SRS imaging of electrolyte dynamics and lithium deposition

SRS has been applied to visualize electrolyte behavior and lithium deposition during electrochemical cycling. By imaging anions as proxies for lithium ions, SRS revealed how electrolyte concentration evolves near electrodes and identified three stages of lithium metal growth that reflect progressive ion depletion (Cheng et al., 2018). This study also showed that growing lithium structures sharpen ionic gradients, which promotes uneven deposition, and that introducing an artificial solid-electrolyte interphase stabilizes the local environment. These results demonstrate that SRS can capture operando chemical information that complements electrochemical and microscopy methods during battery operation.

5.7.3 SRS imaging of micro- and nanoplastic particles (MNPs)

SRS microscopy has become a valuable tool for detecting and characterizing microplastics and nanoplastics in environmental and biological samples. An early study demonstrated rapid identification of common polymers in microplastic particles from consumer products and estuarine sediments by imaging at a small set of diagnostic wavenumbers and generating polymer-specific maps, achieving screening speeds far beyond conventional Raman mapping (Zada et al., 2018). Recent work extended SRS to plastics embedded in tissues. In fish samples, SRS distinguished different polymer types and quantified detection sensitivity as a function of particle size, showing advantages in settings where fluorescence or spontaneous Raman are limited by background (Floess et al., 2024). High-speed SRS imaging in flow enabled detection of individual nanoplastics, identification of polymer type, and estimation of particle size down to 100 nm without optical trapping (Huber et al., 2024). An in vivo study in zebrafish larvae used hyperspectral SRS to map organ-specific accumulation of microplastics and nanoplastics and revealed stronger inflammatory and metabolic effects from 50 nm polystyrene beads compared with micrometer-sized beads (Xin et al., 2024). To improve throughput and sensitivity, Qian et al. developed a hyperspectral SRS platform with automated spectral matching for single-particle characterization. Applied to bottled water, it identified common polymer types, quantified submicron particle counts, and revealed substantial heterogeneity in particle size and composition (Qian N. et al., 2024). These studies establish SRS as a powerful approach for polymer identification, microplastic and nanoplastic detection in complex matrices, and assessment of their biological effects. Continued advances in hyperspectral speed and automated analysis will strengthen SRS as a practical tool for microplastic monitoring and environmental research.

6 Summary and future perspectives

Stimulated Raman scattering (SRS) microscopy has progressed from a specialized nonlinear-optical method into a broadly useful platform for fast, label-free chemical imaging across biology, medicine, and materials science. This review has outlined its theoretical foundations, advances in lasers and detection, emerging quantum-inspired approaches, and an expanding application range that includes metabolic imaging, bioorthogonal Raman tagging, stimulated Raman histology, drug delivery, environmental analysis, plant science, and materials characterization.

On the technology side, SRS instrumentation is steadily moving toward compact, alignment-free operation. Solid-state and all-fiber laser sources provide robust, electronically tunable excitation across the fingerprint and high-wavenumber regions, while low-noise modulation and balanced detection schemes approach shot-noise performance in realistic conditions. Parallel progress in pulse shaping, spectral focusing, and fast tunable sources supports video-rate hyperspectral imaging. Computation has become integral to the instrument, with real-time spectral unmixing, denoising, and deep-learning-based segmentation running on GPUs. These advances should enable chemically resolved images at acquisition speed, along with quantitative readouts such as concentration maps and metabolic flux metrics. Quantum-enhanced detection and QED-based performance models offer additional routes for improving sensitivity. SRS endoscopy represents another major frontier. Flexible coherent Raman probes have demonstrated in vivo chemical contrast through hollow-core and double-clad fibers, and improvements in fiber design, miniature optics, and background-reduction strategies will support applications in gastroenterology, neurosurgery, dermatology, and head-and-neck imaging. Integration with two-photon fluorescence, SHG, THG, OCT, and photoacoustics within shared probe geometries will further expand diagnostic capabilities.

On the application side, SRS is transforming studies of metabolism and molecular organization. Label-free imaging of intrinsic vibrational modes enables direct visualization of lipids, proteins, nucleic acids, and water in living systems, while bioorthogonal Raman tags extend this capability to specific pathways, drugs, and delivery vehicles. Deeper integration with genetic models, omics measurements, and microfluidics will expand SRS as a quantitative and longitudinal tool for cell and tissue metabolism. Toward clinical use, stimulated Raman histology (SRH) provides rapid, stain-free pathology that has shown strong performance across multiple organs, supported by deep-learning models for near real-time interpretation. Continued development of disease coverage, nuclear and molecular specificity, and cross-institutional AI validation will be important for broader adoption. In materials and environmental science, SRS is increasingly used to study battery electrolytes, catalytic interfaces, plant cell walls, seeds, and micro- and nanoplastics (MNPs), where chemically specific imaging reveals mechanisms of transport, reaction, and degradation under realistic conditions.

In conclusion, the broader impact of SRS will depend on both improved optical performance and thoughtful engineering that meets the needs of users outside specialized optics laboratories. Systems must be compact, robust, safe, and simple to operate, with seamless integration into clinical or industrial workflows. With continued progress in light sources, detection, computation, and probe design, SRS microscopy is poised to become a widely adopted method for quantitative, label-free imaging across biological and material systems.

Generative AI statement

The author(s) declared that generative AI was used in the creation of this manuscript. During the preparation of this review article, the author used ChatGPT 5.1 for limited tasks such as locating relevant research papers and refining English phrasing. All generated content was reviewed, edited, and verified by the author, who takes full responsibility for the final manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Liviu Movileanu, Syracuse University, United States

Reviewed by: Giuseppe Sancataldo, University of Palermo, Italy

Salvatore Sorrentino, Massachusetts Institute of Technology, United States