Deep Learning–Driven Medical Imaging for Pediatric Oncology: From Diagnosis to Clinical Decision Support

Pediatric oncology is rapidly advancing through precision therapeutics, biomarker-informed trials, and adaptive treatment strategies designed to improve cure rates while reducing toxicity and long-term side effects. In parallel, deep learning has progressed from proof-of-concept image classification to clinically oriented methods that can support risk stratification, therapy planning, and longitudinal monitoring using routinely acquired medical imaging. Yet translation into pediatric practice remains limited by small and heterogeneous cohorts, multi-site variability, and persistent concerns about generalizability, bias and fairness, interpretability, privacy, and insufficient external and real-world clinical validation. These gaps are especially consequential when AI outputs are used to guide decisions around treatment intensity, targeted therapies, advanced radiotherapy approaches, and response-adapted care.This Research Topic focuses on trustworthy, clinically meaningful deep learning–based medical imaging across all pediatric cancers. The aim is to consolidate advances that deliver actionable imaging-based support for diagnosis and staging, risk stratification, response assessment, toxicity prediction, and surveillance — with clear links to pediatric oncology treatment strategies and biomarker-driven clinical development. We welcome original research, systematic reviews, and clinically grounded methods papers that use real-world pediatric imaging data, employ robust validation strategies (including external and multi-center validation where feasible), and incorporate explainable and interpretable AI and uncertainty reporting to support safe and transparent clinical adoption.In scope are imaging studies that evaluate clinically meaningful endpoints and demonstrate readiness for pediatric oncology practice, including performance across sites, scanners, and patient subgroups. We also welcome imaging-anchored multimodal approaches that integrate imaging features with clinical variables, pathology, or molecular profiling to improve biological interpretability and clinical utility (e.g., radiogenomics, imaging-based biomarker discovery). Submissions should clearly articulate how the proposed models could support pediatric oncology decision-making, such as treatment escalation or de-escalation, response-adapted therapy, radiotherapy planning, and survivorship-oriented care.Topics of interest include, but are not limited to:- Deep learning–based pediatric tumor detection, segmentation, and characterization directly tied to clinical decisions (staging, treatment planning, response evaluation)- AI-assisted pathology diagnosis, staging, and risk stratification using CT, MRI, PET, X-ray, ultrasound, and clinically relevant multi-parametric imaging- Longitudinal imaging for treatment response prediction, early response assessment, and relapse surveillance- Imaging-based outcome and toxicity prediction aligned with treatment optimization and late-effects reduction- Radiogenomics and imaging-anchored multimodal frameworks linking imaging phenotypes to molecular, pathological, and clinical variables for biologically grounded risk stratification- Pediatric imaging AI, including explainability, calibration and uncertainty estimation, bias and fairness assessment, privacy-preserving or federated learning, multi-site harmonization, and rigorous external and real-world validation- Clinical validation and trial integration, including imaging AI workflow evaluation and studies designed to support biomarker-driven pediatric trials and clinical adoptionWe do not consider isolated case reports, nor studies focused on imaging methods or biomarkers without a clear pediatric oncology foundation and direct relevance to clinical management, treatment strategy, or patient outcomes.Please note: manuscripts consisting solely of bioinformatics, computational analysis, or predictions based on public databases — without accompanying validation in an independent clinical or patient cohort, or biological validation in vitro or in vivo — are not suitable for publication in Frontiers in Oncology. Authors with such manuscripts are encouraged to consider submission via Frontiers in Artificial Intelligence or Frontiers in Big Data, both of which are participating journals in this Research Topic. Topic editors: Isidro Machado, Francisco Giner, Suhaila Abuowaida, Hamza Abu Owida Source: https://www.frontiersin.org/research-topics/78411/deep-learning-driven-medical-imaging-for-pediatric-oncology-from-diagnosis-to-clinical-decision-support