Multimodal learning in gastrointestinal diseases

Abstract

Gastrointestinal diseases represent a significant global health challenge, with current diagnostic and management approaches constrained by the integration of heterogeneous multimodal data. This review synthesizes the development prospects of multimodal applications in gastroenterology. By summarizing recent literatures, we demonstrate how multimodal integration can facilitate gastrointestinal disease screening, enable more accurate staging, support treatment decision-making, and optimize clinical workflows. Feature-level fusion serves as the dominant technique in current implementations, while hybrid approaches combining multiple fusion levels are increasingly adopted to enhance flexibility in complex clinical scenarios. Despite these advances, retrospective performance does not guarantee clinical success. Persistent challenges, including data heterogeneity, modality incompleteness, and barriers to clinical translation, remain to be addressed. Overall, this review underscores the transformative potential of multimodal learning to advance precision gastroenterology through integrated diagnostic and therapeutic.