Deep Learning Analysis of Tumor-Infiltrating Lymphocyte, High Endothelial Venule, and Tertiary Lymphoid Structures in Breast Cancer Histopathology Whole Slide Imaging

Understanding the tumor immune microenvironment (TIME) is crucial for predicting prognosis and guiding immunotherapy in breast cancer. In this work, we present a deep learning pipeline to jointly analyze tumor-infiltrating lymphocytes (TILs), high endothelial venules (HEVs), and tertiary lymphoid structures (TLSs) from whole slide images (WSIs). Our method integrates multi-modal histopathology data with advanced segmentation and classification models. Specifically, we use HoverNet to segment tumor and immune cells, followed by ResNet-50-based classification to identify immune cell types and generate spatial heatmaps via Gaussian smoothing. For HEV detection, we apply nnU-Netv2 with color normalization using ICC profiles and morphological post-processing to improve accuracy. The resulting masks enable quantitative assessment of immune infiltration patterns and vascular architecture. This automated framework provides a scalable tool for comprehensive TIME analysis in digital pathology.