Histopathology based AI model predicts anti-angiogenic therapy response in renal cancer clinical trial

Area of Science:

• Oncology
• Computational Pathology
• Biomarker Discovery

Background:

• Anti-angiogenic (AA) therapy is crucial for metastatic clear cell renal cell carcinoma (ccRCC), but patient response varies due to a lack of predictive biomarkers.
• Current markers like CD31 are insufficient, and the RNA-based Angioscore is expensive, slow, and struggles with tumor heterogeneity.

Purpose of the Study:

• To develop an interpretable deep learning (DL) model for predicting angiogenesis directly from histopathology slides.
• To create a cost-effective and efficient method for assessing angiogenesis and predicting AA therapy response in ccRCC.

Main Methods:

• Developed a deep learning model (H&E DL Angio) to predict the Angioscore from standard H&E stained histopathology slides.
• Validated the model across multiple cohorts, comparing its performance against CD31 and the original Angioscore.

Main Results:

• H&E DL Angio demonstrated strong correlations with the Angioscore (Spearman's rho = 0.77 and 0.73).
• Angiogenesis inversely correlated with tumor grade and stage, and was associated with driver mutation status.
• DL Angio effectively predicted AA response in real-world and clinical trial cohorts, outperforming CD31 and closely matching Angioscore predictive power (c-index 0.66 vs 0.67).

Conclusions:

• Interpretable deep learning on histopathology slides offers a powerful, cost-effective alternative for angiogenesis assessment in ccRCC.
• H&E DL Angio can guide anti-angiogenic therapy selection, improving patient outcomes by overcoming limitations of current biomarkers.