Deep Learning Model for Predicting Immunotherapy Response in Advanced Non-Small Cell Lung Cancer.

Area of Science:

• Oncology
• Artificial Intelligence
• Biomarker Discovery

Background:

• Limited response rates to immune checkpoint inhibitors (ICIs) in advanced non-small cell lung cancer (NSCLC) necessitate personalized treatment strategies.
• Identifying patients likely to benefit from ICI therapy is crucial for optimizing clinical outcomes in advanced NSCLC.

Purpose of the Study:

• To develop and validate a deep learning (DL) model for predicting ICI response in advanced NSCLC patients.
• To compare the DL model's predictive performance against established biomarkers like PD-L1, TMB, and TILs.
• To assess the DL model's association with progression-free and overall survival in advanced NSCLC.

Main Methods:

• A multicenter cohort study developed a DL model using whole slide images from advanced NSCLC patients treated with ICI monotherapy.
• The model was independently validated in a separate cohort.
• Performance was evaluated by comparing its predictive power for objective response rate (ORR) against PD-L1, TMB, and TILs.

Main Results:

• The DL model demonstrated significant predictive capability for ICI response, with an AUC of 0.66 in the validation cohort.
• The DL model was an independent predictor of progression-free and overall survival.
• The DL model outperformed TMB and TILs and was comparable to PD-L1, with improved specificity; combining DL with PD-L1 further enhanced prediction.

Conclusions:

• A deep learning model derived from histopathology images shows strong, independent association with ICI response in advanced NSCLC.
• This DL model has the potential to refine treatment selection and improve precision medicine for advanced NSCLC patients.
• Clinical implementation of this DL tool could enhance patient identification for beneficial ICI therapy.