This study introduces an AI-driven framework for predicting immunotherapy efficacy in non-small cell lung cancer (NSCLC) using digital pathology scans. Leveraging hematoxylin and eosin (H&E) slides, the proposed method bypasses conventional PD-L1 immunohistochemistry by extracting robust feature representations through a pre-trained foundation model. An attention-based multiple instance learning (ABMIL) framework aggregates tile-level information to predict PD-L1 expression categories and treatment response in a weakly supervised setting. An ensemble strategy of multiple models is employed to further enhance predictive performance and mitigate overfitting. The results demonstrate the potential of this non-invasive, cost-effective strategy for refined patient stratification and improved prognostic assessment in NSCLC immunotherapy. |
*** Title, author list and abstract as submitted during Camera-Ready version delivery. Small changes that may have occurred during processing by Springer may not appear in this window.