Mind the gap: challenges and future directions for content-based image retrieval in clinical radiology

Artificial intelligence (AI) has become increasingly integrated into radiology across multiple domains, including image acquisition, interpretation, workflow optimization, and clinical decision support. Among these applications, advanced image retrieval systems have the potential to assist clinicians in identifying visually and clinically similar cases, as well as relevant prior studies. This article provides a comprehensive overview of content-based image retrieval (CBIR) and introduces the emerging paradigm of trajectory retrieval, which focuses on the longitudinal evolution of diseases over time. We discuss the clinical applications of these systems, including diagnostic decision support, workflow optimization, educational uses, and research cohort building, highlighting their potential to enhance diagnostic reasoning and reduce uncertainty in complex or rare cases. Despite substantial research progress, adoption in real-world clinical settings remains limited due to challenges such as data heterogeneity, privacy constraints, the need for multimodal integration, and difficulties in capturing temporal dynamics. Trajectory-based approaches, combined with multimodal data integration and human-in-the-loop feedback mechanisms, offer a promising path toward overcoming these barriers by aligning retrieval systems more closely with the longitudinal and holistic nature of clinical decision-making. By addressing these challenges, AI-powered radiology image retrieval has the potential to transform workflows, support more precise and confident diagnoses, and ultimately improve patient care outcomes.