The spatial distribution of tumor cells is an independent prognostic marker in breast cancer

The biological behavior of tumor cells is closely related to tumor invasion, but the relationship between the spatial distribution of tumor cells at the invasion boundary and prognosis is still unknown. In this study, we based Multiphoton Microscopy (MPM) to image the Tumor-Associated Collagen Signatures (TACS), a biomarker known to be associated with Breast Cancer (BC) prognosis and captured 180μm × 180μm regions of interest (ROI) on its colocalization of hematoxylin and eosin (H&E) images. The deep neural network was used to extract the 51 microscopic features of the spatial distribution of tumor cells from H&E images. The least absolute shrinkage and selection operator (LASSO) regression was used to select the most robust features to build a prognostic score. Our results showed that the spatial distribution of tumor cells was a poor prognostic factor in univariate analysis, and was proved to be an additional histological variable with an independent influence on disease-free survival (DFS) in the multivariate analysis in patients with BC. The area under the receiver operating characteristic curve (AUC) of the prognostic score was 0.784 in the training cohort and 0.717 in the validation cohort. The study suggested that the spatial distribution of tumor cells was an independent prognostic marker for DFS in BC and can provide prognostic information for BC.