Journal: The Journal of pathology
This Perspective article reviews how AI-based computational pathology tools are becoming true companion diagnostics rather than just “digital pathologists.”
Key points:
- Current AI tools in pathology mostly replicate human tasks (tumor detection, grading, classification) on H&E and IHC slides, but newer models go further, predicting biomarkers (e.g., MSI, IHC-based markers) and even prognosis directly from whole slide images.
- A major emerging use is as drug-linked companion diagnostics: automated, quantitative IHC assessment tied to prediction of treatment benefit.
- The TROPION-PanTumor01 trial is highlighted as a pivotal example: a supervised machine learning–derived quantitative continuous score for TROP2 IHC outperformed human scoring in stratifying non–small cell lung cancer patients for datopotamab deruxtecan.
- Similar quantitative AI approaches for HER2 and PD-L1 are starting to refine patient selection, revealing subgroups that may benefit from targeted or immunotherapies beyond conventional pathologist-read thresholds.
- Multiple comparable commercial tools are approaching or entering clinical practice, creating opportunities (greater consistency, finer stratification, potential expansion of treated populations) and challenges (validation, regulation, integration into workflows, interpretability, and training requirements).
- The article, promoted by European multidisciplinary societies focused on digital pathology and AI in cancer, is aimed at equipping practicing pathologists and oncologists with the conceptual background needed to evaluate, adopt, and critically use these computational companion diagnostics as they move from research into routine care.