RP32 - Computational Histopathology Across Tumour-Intrinsic and Microenvironmental Biomarkers in Melanoma
Recent advances in digital pathology, computer vision, and foundation models have enabled AI systems that extract clinically relevant information directly from routine histopathology images. This provides a unique opportunity to improve prediction of melanoma patient outcome, including recurrence, survival, and treatment response, by identifying image-based biomarkers that capture different aspects of tumour biology. In the first funding period, we developed AI models for clinically relevant biomarkers, including BRAF mutation status and mitotic activity from whole-slide images (RP18). Building on this, we will develop a comprehensive panel of specialised AI models that together provide a richer characterization of melanoma biology and prognosis. Rather than a single end-to-end model, each biomarker will be modelled independently, allowing rigorous evaluation, validation, and interpretation of its individual clinical value [1], [2].
The project will expand beyond tumour proliferation and molecular alterations to investigate additional tumour-intrinsic biomarkers, including tumour grade, necrosis, and lymphovascular invasion. In parallel, we will develop models characterising the tumour microenvironment, including tumour-infiltrating lymphocytes (TILs), macrophage abundance, and other immune-related features increasingly recognised as important determinants of prognosis and treatment response. Together, these models will provide a multidimensional digital representation of the tumour and its microenvironment. We hypothesise that combining validated digital biomarkers will enable more accurate prediction of clinically relevant endpoints, such as recurrence, melanoma-specific survival, and treatment response, than any single biomarker alone. The project will investigate strategies for integrating specialised model outputs into clinically meaningful patient-level risk scores and prognostic profiles, while preserving the interpretability of each biomarker. By maintaining a modular framework in which each AI model is independently developed, validated, and understood, we aim to increase transparency, facilitate clinical verification, and strengthen clinician trust and acceptance. Ultimately, this work will establish a computational pathology framework for precision melanoma medicine, offering explainable, biomarker-based decision support (RP36).
[1] Vorontsov, E., Bozkurt, A., Casson, A. et al. A foundation model for clinical-grade computational pathology and rare cancers detection. Nat Med 30, 2924–2935 (2024). https://doi.org/10.1038/s41591-024-03141-0
[2] Echle, A., Rindtorff, N.T., Brinker, T.J. et al. Deep learning in cancer pathology: a new generation of clinical biomarkers. Br J Cancer 124, 686–696 (2021). https://doi.org/10.1038/s41416-020-01122-x