Advanced AI for missing modality learning and uncertainty‑aware medical image analysis

Modern medical imaging increasingly relies on artificial intelligence to support detection, diagnosis, and prognostic decision‑making. As an emerging field in image analysis, missing modality learning and uncertainty‑aware modeling aim to bridge the gap between ideal multimodal AI pipelines and real‑world clinical imaging, where data are often incomplete or noisy. In practical clinical scenarios, complete multimodal datasets are rare: modalities such as MRI, CT, ultrasound, and PET may be unavailable due to acquisition protocols, resource limits, or patient‑specific constraints. Simultaneously, available images often exhibit significant uncertainty originating from noise, domain shifts, and limited annotations. These intertwined challenges compromise the reliability and safety of AI systems intended for deployment in hospitals and research centers.Recent research has achieved remarkable progress in deep learning–based segmentation, detection, and multimodal fusion; however, most models implicitly assume full and clean multimodal inputs. The resulting gap between idealized training conditions and clinical reality often leads to unreliable predictions, poor calibration, and brittle generalization across centers. Addressing missing modalities while quantifying prediction uncertainty constitutes a vital step toward trustworthy, deployable medical AI systems that can operate robustly under heterogeneous, incomplete data conditions.This Research Topic aims to bring together the medical imaging, computer vision, and AI communities to advance learning frameworks that integrate missing‑modality reasoning and uncertainty estimation. The goal is to promote AI pipelines that remain stable under incomplete and uncertain inputs, and that communicate their confidence transparently to clinicians. Contributions may combine data‑driven and physics‑based modeling, probabilistic learning, or clinically guided interpretability to ensure reliable real‑world performance and decision support. To gather further insights in this field, we welcome articles addressing, but not limited to, the following themes:—Missing‑modality‑aware learning for segmentation, lesion detection, classification, and prognosis—Deep learning for modality imputation, completion, and cross‑modal synthesis from partial observations—Uncertainty quantification and calibration under incomplete or noisy multimodal inputs—Robust multimodal fusion and representation learning when modalities are absent or corrupted—Domain adaptation and generalization strategies for incomplete multimodal pipelines—Clinically interpretable AI systems conveying uncertainty and reliability in decision‑support scenarios—Benchmark datasets, evaluation metrics, and validation protocols under missing‑modality settings Topic editors: Tongxue Zhou, Maël Balluet, Zhiqin Zhu, Bangli Liu Source: https://www.frontiersin.org/research-topics/82495/advanced-ai-for-missing-modality-learning-and-uncertainty-aware-medical-image-analysis