Fig. 1

Workflow of this study. a Selection of the region of interest on a digital H&E image. The selected region of interest was used to extract pathomics features by CellProfiler software. The region of interest was subjected to multiphoton imaging. Then, collagen features were extracted from the multiphoton image by MATLAB 2018b. LASSO regression was used to select the most predictive parameters to construct the PCS. b The PCS-nomogram was developed based on the PCS and four clinicopathological predictors to predict LNM in the training cohort. c The PCS-nomogram was verified in the internal and external validation cohorts. Scale bars: 1000 μm and 200 μm. PCS pathomics-collagen signature, LASSO least absolute shrinkage and selection operator, VELIPI venous emboli and/or lymphatic invasion and/or perineural invasion, LNM lymph node metastasis, AUROC area under the receiver operating characteristic curve