Fully automated deep learning radioactive biomarkers derived from serial CT imaging showed strong prognostic value for overall survival in patients with advanced non-small cell lung cancer (NSCLC) treated with immune checkpoint inhibitors (ICIs), according to a large prognostic study.
Deep learning model developed using continuous CT imaging
Reliable early biomarkers of long-term outcome remain a major unmet need in NSCLC treated with immunotherapy. Traditional image-based measurements, such as Response Evaluation Criteria in Solid Tumors (RECIST) and Tumor Volume Change (TVC), often fail to capture the complex response patterns associated with ICIs, such as pseudoprogression and stable disease with survival benefits. In the current study, we sought to address these limitations by developing and validating a deep learning model using serial CT scans obtained before and 12 weeks after treatment.
Researchers analyzed data from 1,830 adults with advanced NSCLC treated with ICIs in the Routine Clinical Practice (RCP) dataset and the Multinational Phase I Clinical Trial (GARNET). The Serial CT Response Score (Serial CTRS) was developed using the RCP Discovery Cohort and validated across 10 center datasets in the United States and Europe, as well as independently validated in the GARNET trial.
Deep learning enables early survival risk stratification
Continuous CTRS showed a significant independent association with overall survival in multivariate analysis adjusting for age, sex, tumor histology, programmed death ligand 1 expression, and tumor volume. In the RCP test cohort, each 10 percentage point increase in predicted 12-month survival probability was associated with a 26% reduction in the risk of death. Prognostic results were even stronger in the clinical trial validation cohort.
Importantly, continuous CTRS consistently outperforms RECIST and TVC in distinguishing between high- and low-risk survival groups. Risk stratification using deep learning models remained robust across key subgroups, including patients classified as having stable disease by RECIST, a population in which clinical decision-making is often difficult.
This model does not require manual tumor measurements and provides a fully automated approach using routine imaging already obtained in clinical practice. This is in contrast to RECIST and volumetric assessment, which are resource intensive and subject to interobserver variability.
The authors conclude that Serial CTRS provides superior prognostic information compared to existing imaging metrics and may support earlier, more informed treatment decisions in patients receiving immunotherapy. In addition to clinical use, biomarkers have the potential to improve patient stratification and endpoint assessment in immuno-oncology trials.
These findings highlight the growing role of artificial intelligence-driven imaging biomarkers in personalized cancer therapy and suggest a path toward more accurate response assessment in advanced NSCLC.
reference
Sako C et al. Deep learning sequential CT prediction of survival in non-small cell lung cancer treated with immunotherapy. JAMA net open. 2026;9;(1):e2555759.
