Deep-Learning Serial CT Prediction of Survival in Immunotherapy-Treated Non-Small Cell Lung Cancer

Area of Science:

• Oncology
• Radiology
• Artificial Intelligence

Background:

• Reliable early biomarkers for overall survival (OS) are needed for advanced non-small cell lung cancer (NSCLC) patients receiving immune checkpoint inhibitors (ICI).
• Current imaging metrics like RECIST and tumor volume change (TVC) have limited predictive power for long-term outcomes.
• Advanced imaging biomarkers can improve clinical decision-making in NSCLC treatment.

Purpose of the Study:

• To develop and validate a fully automated deep-learning imaging biomarker using pretherapy and 12-week follow-up computed tomography (CT) scans.
• To assess the biomarker's predictive performance for overall survival (OS) in advanced NSCLC patients undergoing ICI therapy.

Main Methods:

• A prognostic study utilized retrospective routine clinical practice (RCP) and clinical trial data (2013-2023).
• A deep-learning model, Serial CT response score (Serial CTRS), was developed and validated on multiple datasets, including a multinational clinical trial.
• Cox proportional hazards regression and ROC-AUC analysis modeled the association between Serial CTRS and OS.

Main Results:

• The study included 1830 patients with advanced NSCLC receiving ICI therapy.
• Serial CTRS demonstrated a significant association with OS in multivariable analyses, outperforming RECIST and TVC in risk discrimination.
• The biomarker's predictive value was consistent across subgroups, including PD-L1 expression and RECIST criteria.

Conclusions:

• The fully automated Serial CTRS biomarker effectively predicts OS in advanced NSCLC patients treated with ICI.
• Serial CTRS offers superior prognostic accuracy compared to RECIST and TVC, using the same CT scans.
• This advanced imaging biomarker can enhance clinical trial design and patient management in advanced NSCLC.