WSI plus foundation AI

Pathology labs are turning digitized biopsy slides into one data source for triage, prognosis, and clues about which molecular tests to order. (medscape.com) A whole-slide image is a high-resolution scan of an entire glass pathology slide, so a pathologist can review tissue on a screen instead of only through a microscope. The College of American Pathologists says the Food and Drug Administration has approved select whole-slide imaging systems for primary diagnosis, and labs still have to validate that digital reads match microscope reads before clinical use. (cap.org) Foundation models are large artificial intelligence systems trained on very large image sets so they can learn patterns once and then be adapted to many later tasks. In pathology, that means one model can be tuned for cancer detection, subtype classification, prognosis, or prediction of molecular features from the same slide. (jcp.bmj.com) Medscape reported this week that oncology pathology is moving from narrow image-detection tools toward workflow systems that can flag urgent slides, estimate prognosis, and help decide whether the same biopsy should go to additional molecular testing. The article said the shift depends on three pieces arriving together: validated digital slides, stronger artificial intelligence backbones, and a clearer regulatory path. (medscape.com) The technical change is scale. Providence researchers reported a whole-slide foundation model called Prov-GigaPath trained on 1.3 billion image tiles from 171,189 whole slides drawn from more than 30,000 patients across 31 tissue types. (pmc.ncbi.nlm.nih.gov) Other groups are pairing slide images with text and molecular data so the model learns both what tissue looks like and what that tissue means clinically. A Nature Medicine paper published in late 2025 described TITAN, a multimodal whole-slide foundation model pretrained on 335,645 whole-slide images with linked pathology reports and 423,122 synthetic captions. (nature.com) That matters in oncology because tissue is limited. Medscape said the same biopsy can now support triage, prognosis, molecular testing, and even treatment selection without additional tissue handling, which is a practical issue when small samples have to cover diagnosis plus biomarker workups. (medscape.com) Researchers are also pushing these models to infer molecular features directly from slide morphology, the visible structure of cells and tissue. A 2026 review said whole-slide imaging plus artificial intelligence is being used for risk stratification, prognostic prediction, and precision-medicine workflows, while a BMJ Journal of Clinical Pathology review said foundation models are being studied for biomarker identification and cancer subtyping. (mdpi.com) (jcp.bmj.com) The caution is that most of these systems are not replacing pathologists with autonomous reads. The College of American Pathologists’ guidance is built around laboratory validation, and Medscape framed the newer tools as decision support inside a broader diagnostic workflow rather than stand-alone slide readers. (cap.org) (medscape.com) The near-term story is less about one algorithm calling cancer by itself and more about one scanned slide doing more jobs in the lab. If that model holds up in validation and regulation, the biopsy that starts diagnosis could also help decide what gets tested next. (medscape.com)