Researchers at the Dana-Farber Cancer Institute have trained artificial intelligence (AI) models to assess the clinical features of kidney cancer tumor samples to predict how tumors may respond to immune therapy.
Writing in Cell Reports Medicine, the Dana-Farber team described how they used their deep-learning model to evaluate pictures of tumor samples on pathology slides from clear cell renal cell carcinoma (ccRCC), a type of kidney cancer. The AI-based tool can identify and analyze features of tumors that can aid in predicting how a particular tumor may respond to treatment with immune checkpoint inhibitors (ICIs).
The AI model has thus far been trained to assess a tumor’s nuclear grade, which quantifies how far tumor cells deviate from normal cells; measure the microheterogeneity of the tumor (how much the tumor’s nuclear grade varies across the slide); and assess the levels of immune infiltration -- i.e., how deeply lymphocytes have penetrated the tumor to fight it. These tasks are time-consuming for pathologists to perform routinely.
Renal cell carcinoma is among the 10 most common cancers in the world; 75% to 80% of metastatic cases are the ccRCC subtype. There is currently no method to predict which ccRCC tumors will respond to treatment with ICIs.
Using their AI-based tool, the team assessed pathology slides of tumors from patients who were part of the CheckMate 025 randomized phase III clinical trial, which tested treatment with an ICI or an mTOR inhibitor in patients with ccRCC who had previously been treated with standard therapy.
The results showed that tumor microheterogeneity and immune infiltration were associated with improved overall survival among patients taking ICIs. Furthermore, tumors that responded to ICIs had both higher levels of tumor microheterogeneity and denser infiltration of lymphocytes in high-grade regions.
The Dana-Farber team next plans to assess the deep learning tool in an ongoing clinical trial using combination immunotherapy in patients with ccRCC.