data.zip

Programmed death-ligand 1 (PD-L1) is an important biomarker increasingly used as a predictive marker in breast cancer immunotherapy. Immunohistochemical quantification is the current standard assessment method; however, it presents challenges in terms of time, cost, and reliability. Hematoxylin and eosin (H&E) staining is a routine staining method in cancer pathology, known for being easily accessible and consistently reliable. Deep learning has demonstrated the potential to predict biomarkers in cancer histopathology. This study uses a weakly supervised, multiple instance learning(MIL) approach to predict PD-L1 expression from H&E-stained images using deep learning techniques. In the internal test set, the TransMIL method achieved an area under curve (AUC) of 0.833, and in an independent external test set, it achieved an AUC of 0.799. Additionally, since RNA sequencing results indicate a threshold that makes H&E pathology images separable, we also conducted predictions on the public TCGA-TNBC database, achieving an AUC of 0.721. This demonstrates that the Transformer-based TransMIL model can effectively capture highly heterogeneous features within a MIL framework, exhibiting strong cross-center generalization ability. Our study shows that appropriate deep learning techniques can enable effective PD-L1 prediction on datasets with limited data and across regions and centers. This not only demonstrates the significant potential of deep learning in pathological AI but also provides a valuable reference for the rational and efficient allocation of medical resources.