Radiologic, Pathologic, and Deep Learning Predictors of Response to Immune Checkpoint Blockade in Renal Cell Carcinoma Patients Undergoing Post-Treatment Nephrectomy.

Background: Response assessment of primary kidney tumors in the consolidation cytoreductive and neoadjuvant settings offers a unique opportunity to inform postoperative adaptive treatment strategies. Yet, systematic analyses evaluating radiology, pathology, and machine learning are lacking. Methods: We retrospectively identified consecutive renal cell carcinoma (RCC) patients with locoregionally advanced or metastatic RCC who received at least one cycle of ICI-containing doublet therapy prior to nephrectomy at the UTSW Kidney Cancer Program (2017-2024). Radiologic and pathologic features were centrally reviewed and correlated with clinical outcomes: freedom from start of next systemic therapy (FFNT) in cytoreductive patients and metastasis-free survival (MFS) in neoadjuvant patients. Pathologic response to ICI results in tumor cell death and fibrosis creating hypocellular areas and increased immune infiltrate, features we utilized to build Deep learning (DL) models. We leveraged DL models to validate pathologist-assessed regression objectively and quantitate immune infiltrate. Results: Among 99 patients (cytoreductive nephrectomy {CN}, n=66; neoadjuvant nephrectomy {NaN}, n=33), radiologic tumor shrinkage ≥30% ( Conclusions: This study provides, for the first time, an integrated, quantitative framework of post-ICI response in RCC. Our data suggests that immune-mediated pathologic regression changes differ from coagulative necrosis, an indicator of poor prognosis. Our findings provide a blueprint for complementary role of radiology and pathology evaluation of post ICI-nephrectomy specimens that if validated prospectively, could guide adaptive approaches and clinical trial design for ICI-based therapies in kidney cancer and beyond.