Artificial Intelligence and Digital Pathology for Preoperative Lymphovascular Invasion and Metastatic Risk Prediction in Breast Cancer: Bridging the Biopsy-Resection Gap.

Authors

Abstract

Breast cancer is a leading global diagnosis where lymphovascular invasion (LVI) serves as a critical indicator of metastatic potential. Conventional assessment is often hindered by its reliance on postoperative findings, sampling errors, and subjective interobserver variability. This review evaluates how artificial intelligence (AI), digital pathology, and MRI radiomics provide earlier, quantitative estimations of LVI-related risk. The strongest direct evidence currently comes from MRI-based studies, where validation performance often reaches area under the receiver operating characteristic curve (AUC) values of 0.84-0.90. By contrast, digital pathology is especially mature for LVI-adjacent tasks such as lymph node metastasis detection and slide-based relapse-risk modelling, which together provide an important translational foundation for future LVI-specific tools. This review also addresses issues that remain underdeveloped in the literature, including the distinction between lymphatic and vascular invasion, AI-specific risk-of-bias and reporting frameworks, the emerging regulatory landscape for adjacent breast AI tools, and the gap between resection-based development datasets and biopsy-level preoperative use. Although most modern computational models are developed using extensive surgical resection specimens, their true clinical utility hinges on successful validation and performance within the highly restricted, fragmented tissue context of preoperative core needle biopsies. Overall, the field appears most promising when LVI prediction is framed not as an isolated binary task, but as one component of a broader metastatic-risk workflow that supports calibrated, multidisciplinary breast cancer decision-making.

Tags

Topics