Biopsy image-based deep learning for predicting pathologic response to neoadjuvant chemotherapy in patients with...

Area of Science:

• Oncology
• Artificial Intelligence
• Medical Imaging

Background:

• Neoadjuvant chemotherapy (NAC) is crucial for resectable non-small cell lung cancer (NSCLC), but predicting individual patient response remains challenging.
• Variability in NAC response limits its clinical effectiveness, necessitating improved predictive tools.

Purpose of the Study:

• To develop and validate DeepDrRVT, a weakly supervised deep learning model for predicting NAC response in NSCLC patients using pretreatment biopsy images.
• To assess the model's performance in predicting complete and major pathologic response and its correlation with survival outcomes.

Main Methods:

• A weakly supervised deep learning approach integrating self-supervised feature extraction and attention-based deep multiple instance learning was employed.
• The model, DeepDrRVT, was trained and validated on pretreatment biopsy images from NSCLC patients undergoing NAC.

Main Results:

• DeepDrRVT achieved high predictive performance for complete and major pathologic response across training, internal, and external validation cohorts (AUCs ranging from 0.831 to 0.968).
• The model's digital assessment of residual viable tumor showed significant correlation with pathologists' visual assessment and was associated with longer disease-free survival (DFS).
• DeepDrRVT was identified as an independent prognostic factor for DFS, even after adjusting for clinicopathologic variables.

Conclusions:

• DeepDrRVT demonstrates significant potential as an accurate and generalizable tool for predicting NAC response in NSCLC.
• The model can assist clinicians in making more informed treatment decisions prior to NAC initiation, potentially improving patient outcomes.
• DeepDrRVT offers a reliable and accessible method for enhancing NAC treatment strategies in NSCLC.