Machine Learning Models Identify Key Histological Features of Renal Cell Carcinoma Subtypes

Renal Cell Carcinoma (RCC) is a heterogeneous disease that can be classified into subtypes for diagnosis based on assessment of tumor histology, and multiple molecular biomarkers and mutations .^{1, 2}

Clear cell (cc) RCC is the predominant subtype comprising 80% of all cases, with papillary and chromophobe carcinoma accounting for 80% of all other RCCs, but overall, 16 different RCC subtypes have been identified.¹ Treatment selection and prognosis varies by subtype, and treatment response has been associated with the cell and tissue composition of the tumor microenvironment (TME), including tumorspecific mutations.¹ For example, PBRM1 is commonly mutated gene in RCC that may contribute to disease prognosis and response to immunotherapy, although its role is currently unclear.^3-6 Exhaustive classification and quantification of the TME by machine learning (ML) models has the potential to reveal associations between tumor histology and mutations or molecular biomarkers.⁷

Here, ML models quantified histologic features of the TME directly from RCC hematoxylin and eosin (H&E)-stained whole slide images (WSI). The potential for model outputs to predict clinically-relevant biomarkers was investigated.

¹ Escudier et al. Ann. Oncol. 2019;30(5):706-720
² Cimadamore et al. Transl. Androl. Urol. 2021 Mar;10(3):1506-1520
³ Liu et al. Nat Commun. 2020;11(1):2135.
⁴ Braun et al. JAMA Oncol. 2019;5(11):1631- 1633.
⁵ Dizman et al. J Immunother Cancer. 2020;8(2):e000953
⁶ Carneiro et al. Kidney Cancer 2021 5(2):79-92
⁷ Diao et al. Nat Commun. 2021;12(1):1613