AI-Based Model Predicts Kidney Cancer Therapy Response

Each year, nearly 435,000 individuals are diagnosed with clear cell renal cell carcinoma (ccRCC), making it the most prevalent subtype of kidney cancer. When the disease spreads, anti-angiogenic therapies are commonly employed as a treatment. These drugs work by inhibiting the formation of new blood vessels in tumors, limiting access to the molecules that fuel tumor growth. While anti-angiogenic drugs are widely prescribed, fewer than 50% of patients experience benefits from them, leading many to face unnecessary toxicity and financial strain. Currently, there are no clinically available biomarkers to accurately predict which patients will respond to anti-angiogenic drugs. A previous clinical trial suggested that Angioscore, a test that evaluates the expression of six blood vessel-associated genes, could show promise. However, this genetic test is expensive, difficult to standardize across clinics, and causes delays in treatment. It also examines only a small portion of the tumor, and since ccRCC is highly heterogeneous, gene expression can vary in different regions of the cancer.

Researchers at UT Southwestern Medical Center (Dallas, TX, USA) have now developed an artificial intelligence (AI)-based model that can accurately predict which kidney cancer patients are likely to benefit from anti-angiogenic therapy, a treatment class effective only in certain cases. Their findings, published in Nature Communications, may offer a viable way to use AI in guiding treatment decisions for this and other cancer types. The researchers created this predictive model using AI to analyze histopathological slides – thinly sliced tumor tissue sections that are stained to highlight cellular characteristics. These slides are routinely included in a patient’s diagnostic workup, and their images are increasingly available in electronic health records.

The AI algorithm, which is based on deep learning, was trained using two datasets: one that matched ccRCC histopathological slides with their corresponding Angioscore results, and another that matched slides with a test the researchers developed to assess blood vessels in the tumor sections. Unlike many deep learning algorithms that provide results without insight into their reasoning, this approach is designed to be visually interpretable. Rather than outputting a single number and directly predicting a response, the algorithm generates a visualization of the predicted blood vessels, which closely correlates with the RNA-based Angioscore. Patients with more blood vessels are more likely to respond to therapy, and this method enables users to understand how the model arrived at its conclusions.

When the researchers tested this approach on slides from over 200 patients, who were not included in the training data – including samples from the clinical trial that demonstrated the potential of Angioscore – the model predicted which patients were most likely to respond to anti-angiogenic therapies almost as accurately as Angioscore. The algorithm predicted that a responder would have a higher score than a non-responder 73% of the time, compared to 75% with Angioscore. The researchers believe that AI analysis of histopathological slides could ultimately help guide diagnostic, prognostic, and therapeutic decisions across various conditions. They also plan to develop a similar algorithm to predict which ccRCC patients will respond to immunotherapy, another treatment class that is effective for only some patients.

“There’s a real unmet need in the clinic to predict who will respond to certain therapies. Our work demonstrates that histopathological slides, a readily available resource, can be mined to produce state-of-the-art biomarkers that provide insight on which treatments might benefit which patients,” said Satwik Rajaram, Ph.D.