Figure 1.

The research design and process of this study. We collected resected tumor tissues from 121 stage-I NSCLC patients and sliced them into paraffin sections for IHC staining of ten immune checkpoints. All sections were captured whole-slide images under microscopy (Part 1). Next, we inputted all original IHC images into the EfficientUnet-b3 model to acquire tumor cell segmentation masks. These masks were then processed with HSV thresholds, filtering, connected components extraction, and sampling to classify four types of cells in IHC images and the quantitative and spatial analysis of immune checkpoints expression on TCs and TILs. Fifty-five features were extracted upon analysis and were imputed into univariable and multivariable Cox regressions to establish the IC-Score (Part 2). Meanwhile, we also inputted all original IHC images into the Resnet to extract prognostic features and then applied univariable and multivariable Cox regressions to establish the Res-Score (Part 3). Further, to evaluate the performance of the IC-Score, Res-Score, and their combination with clinical features, we performed the AUC and NRI analysis on the dataset pretreated with 5-fold cross-validation with 100× bootstrap (Part 4). Abbreviations: NSCLC, non-small cell lung cancer; IHC, immunohistochemistry; HSV, hue, saturation, and value; TC, tumor cell; TIL, tumor-infiltrating lymphocytes; IC-Score, immune checkpoints score; Res-Score, ResNet score; AUC, area under the receiver operating characteristic curve; NRI, net reclassification index.