A FULLY automated deep-learning radiomic biomarker derived from serial CT imaging has demonstrated 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 Serial CT Imaging
Reliable early biomarkers of long-term outcomes remain a major unmet need in immunotherapy-treated NSCLC. Conventional imaging-based measures, such as Response Evaluation Criteria in Solid Tumors (RECIST) and tumour volume change (TVC), often fail to capture complex response patterns associated with ICIs, including pseudoprogression and stable disease with survival benefit. The current study sought to address these limitations by developing and validating a deep-learning model using serial CT scans obtained before treatment and at 12 weeks.
The investigators analysed data from 1,830 adults with advanced NSCLC treated with ICIs across routine clinical practice (RCP) datasets and a multinational Phase I clinical trial (GARNET). A serial CT response score (Serial CTRS) was developed using an RCP discovery cohort and validated across 10 US and European institutional datasets, as well as independently validated in the GARNET trial.
Deep Learning Enables Early Survival Risk Stratification
Serial CTRS showed a significant independent association with overall survival in multivariable analyses adjusting for age, sex, tumour histology, programmed death-ligand 1 expression, and tumour volume. In the RCP test cohort, each 10 percentage-point increase in predicted 12-month survival probability was associated with a 26% reduction in mortality risk. Prognostic performance was even stronger in the clinical trial validation cohort.
Importantly, Serial CTRS consistently outperformed RECIST and TVC in discriminating between high- and low-risk survival groups. Risk stratification using the deep-learning model remained robust across key subgroups, including patients classified as having stable disease by RECIST, a population in whom clinical decision-making is often challenging.
The model required no manual tumour measurements, offering a fully automated approach using routine imaging already obtained in clinical practice. This contrasts with RECIST and volumetric assessments, which are resource intensive and subject to inter-observer variability.
The authors conclude that Serial CTRS provides superior prognostic information compared with existing imaging metrics and may support earlier, more informed treatment decisions for patients receiving immunotherapy. In addition to clinical use, the biomarker could improve patient stratification and endpoint assessment in immuno-oncology trials.
These findings highlight the growing role of artificial intelligence–driven imaging biomarkers in personalised cancer care and suggest a path towards more precise response assessment in advanced NSCLC.
Reference
Sako C et al. Deep-learning serial CT prediction of survival in immunotherapy-treated non–small cell lung cancer. JAMA Netw Open. 2026;9;(1):e2555759.
