Knowledge Distillation in Medical Imaging: A Comprehensive Survey of Trends, Challenges, and Opportunities

Knowledge Distillation (KD) has rapidly evolved into a cornerstone technique for model compression and efficient deployment of deep learning systems. In the domain of medical imaging, where model interpretability, computational constraints, and clinical trust are critical, KD enables the development of lightweight models that retain the diagnostic performance of their larger counterparts. Despite its growing relevance, the literature on KD in medical imaging remains fragmented, with diverse methodologies and applications scattered across tasks, modalities, and clinical settings. This paper presents one of the most comprehensive and technically rigorous overviews to date on KD in medical imaging. It systematically categorizes existing techniques according to three primary knowledge-transfer paradigms: response-based, feature-based, and relation-based distillation, while also examining complementary strategies such as hybrid, self-distillation, and cross-modal distillation. The survey evaluates their suitability for classification, segmentation, detection, and registration tasks across major medical imaging modalities. Beyond methodological taxonomy, the paper compiles an extensive analysis of benchmark datasets, trends in publication across leading conferences and journals, and the role of KD in clinical deployment pipelines. Unique to this work is a detailed treatment of regulatory, explainability, and workflow integration challenges, alongside a deep exploration of interdisciplinary perspectives and future research directions such as federated distillation, domain adaptation, and hybrid compression strategies. By consolidating and contextualizing a wide body of research, this paper aims to serve as both a foundational reference and a forward-looking roadmap for researchers, clinicians, and developers striving to build efficient, trustworthy, and clinically translatable AI models through KD.