Issues
Generative intelligence in precision oncology: Priorities in informatics engineering, pathology and oncology
Generative artificial intelligence (GAI) applied to clinical diagnostics and research is reshaping the panorama of precision oncology. Combining hematoxylin-eosin-stained whole slide images with computational algorithms opens new avenues in digital pathology. GAI allows for extracting molecular, immunological, and prognostic information based on routinely processed histological sections and removes the need for additional molecular testing.
In oncology, GAI models excelled in cancer histotyping, malignancy ranking, molecular profiling, identification of prognostic and predictive biomarkers, and inference of immune gene signatures. The latest foundational models provide additional opportunities to develop generalizable, scalable tools that can be consistently leveraged in line with pathology missions.
However, several challenges must still be addressed to optimize GAI performance and encourage its clinical application. These include data quality, algorithm bias, generalizability across institutions, and validation through robust multicenter trials. This strategy is crucial for increasing clinical confidence, ensuring reproducibility, and facilitating the routine use of AI in precision oncology.
This review focuses on the operational application of computational pathology within the broader context of precision oncology. It addresses the most significant technical innovations in biomarker assessment and critically examines the priorities to enhance the reliability, scalability, and performance of AI-driven tools in precision oncology.
Impact statement
Generative artificial intelligence applied to digital histopathology offers novel strategies for biomarker identification and tumor classification, advancing precision oncology and diagnostic accuracy.