Fig. 1. Schematic illustration of PS_GC construction.

a Selection of a representative H&E tile. The H&E-stained images of all included 480 patients are used, and 10 regions of interest with a field of view of 1000 × 1000 pixels per image containing the greatest number of tumour cells are randomly selected for analysis. One pixel is equal to 0.504 μm. Scale bars: 1500 and 50 μm, respectively. b Framework for constructing the PS_GC. The pathomics features are extracted from the H&E-stained image tiles, and then the potential predictors are selected using a LASSO-Cox regression model in the training cohort with 264 patients. The PS_GC is calculated via a linear combination of the selected features, and the PS_GC for the validation cohort with 216 patients is directly calculated from the formula obtained in the training cohort. PS_GC pathomics signature of gastric cancer, H&E haematoxylin and eosin, LASSO least absolute shrinkage and selection operator.