Computational pathology in 2026: whole-slide imaging at scale, AI integration, and the clinical adoption gap

Digital and computational pathology has moved from specialty academic center deployment to mainstream adoption faster than most imaging AI categories. The infrastructure constraint — whole-slide scanners, high-bandwidth storage, specialized workstations — was real five years ago and has been addressed by cost declines and cloud storage. The remaining barriers are integration (pathology AI that doesn’t integrate with the pathology LIS is not clinically usable), evidence (a cleared algorithm is not the same as one with proven clinical outcome benefit), and the subspecialty structure of pathology practice (a breast pathology AI is useful to breast subspecialists and useless to a generalist).

The multimodal models integrating pathology images with genomic data are the most significant technical development in the review. The ability to predict molecular subtype from hematoxylin and eosin (H&E) slide appearance — rather than requiring separate genomic sequencing — could substantially reduce turnaround time and cost for biomarker-driven treatment decisions in oncology. If a foundation model can predict whether a tumor will respond to PD-1 inhibition from the slide alone, that changes the treatment planning workflow for a large proportion of solid-tumor oncology.

The clinical adoption gap identified in the review is structural: pathology departments in community hospitals and non-academic settings have significantly lower adoption of digital pathology infrastructure than academic centers, partly because the financial case is less clear when volume is lower and subspecialty read-out patterns are different. The AI algorithms that are clearest in their value proposition — automated tumor grading, mitosis counting, lymph node assessment — are also the ones most likely to be adopted in the community setting, but infrastructure investment remains a prerequisite.