1118 Deep learning serial CT imaging biomarker for predicting overall survival: a real-world validation in multiple indications of advanced solid tumors treated with immune checkpoint inhibitors

This study validated Serial CT Response Score (Serial CTRS), a fully automated deep learning biomarker, across 243 patients with advanced cancers treated with immune checkpoint inhibitors, including metastatic renal cell carcinoma, metastatic melanoma, and extensive-stage small cell lung cancer. By analyzing baseline and early-treatment CT scans without requiring manual lesion annotations, Serial CTRS achieved a concordance index of 0.77 across the combined cohort and demonstrated consistent prognostic accuracy across all three cancer types, with 12-month overall survival AUROCs ranging from 0.78 to 0.82. Patients stratified into the highest Serial CTRS group exhibited dramatically longer overall survival compared to the lowest group, with a hazard ratio of 10.54.

These findings represent a meaningful step forward in the clinical utility of AI-driven imaging biomarkers for patients receiving immunotherapy. Current standard prognostic tools such as RECIST-based tumor size assessments frequently fall short in capturing the full complexity of treatment response, leaving clinicians with significant uncertainty in decision-making. By generalizing effectively from non-small cell lung cancer to multiple other cancer types, Serial CTRS demonstrates the potential to serve as a broadly applicable, objective prognostic tool that can support personalized treatment decisions across oncology practices and clinical trials.