The field of computational pathology is increasingly reliant on the integration of diverse data modalities to advance beyond traditional diagnostic paradigms. This integration is critical for deciphering complex disease mechanisms and enabling personalized medicine, yet the rapidly evolving landscape of fusion techniques necessitates a consolidated overview. This mini review addresses the pressing need to synthesize current methodologies, applications, and challenges in multimodal data fusion within pathology. We explore the pivotal role of multimodal deep learning in pan-cancer integration for tasks such as cancer origin identification, survival prediction, and biomarker discovery. The review further examines the field of pathogenomics, which fuses pathology, genomic, and clinical data, and details computational frameworks for inferring molecular profiles directly from histology images. Additionally, we cover the integration of multiomics data to characterize the tumor microenvironment, survey emerging computational frameworks and foundation models, and present key clinical studies validating these integrative approaches. Finally, the review discusses significant challenges, including data heterogeneity and standardization, that must be overcome for clinical translation. Future progress in this domain will depend on the development of explainable and federated artificial intelligence models, innovative architectures to handle data complexity, and robust interdisciplinary collaborations. By addressing these frontiers, data fusion is poised to fundamentally transform pathology into a more precise and predictive discipline, ultimately culminating in the realization of unified disease atlases for research and clinical care.

View PDF