Lymphovascular Invasion Detection in Breast Cancer Using Deep Learning

Abstract Lymphovascular invasion (LVI) is a critical pathological feature in breast cancer, strongly associated with an increased risk of metastasis and poorer prognosis. However, manual detection of LVI is labor-intensive and prone to inter-observer variability. To address these challenges, this study explores the potential of Swin-Transformer, a state-of-the-art deep learning model, and GigaPath, a cutting-edge foundation model, for automating the detection of LVI in whole-slide images (WSIs) of breast cancer tissue. We trained the models on a dataset of 90 annotated Hematoxylin and Eosin (H&E)-stained breast cancer slides, achieving strong performance with a slide-level Area Under the Receiver Operating Characteristic (AUC) of 97%, a sensitivity of 79% with an average of 8 false positives (FPs) per slide using the best-performing model. The results underscore the potential of Swin-Transformer and GigaPath to enhance diagnostic accuracy and consistency in LVI detection. Competing Interest Statement The authors have declared no competing interest. Funding Statement Yes Author Declarations I confirm all relevant ethical guidelines have been followed, and any necessary IRB and/or ethics committee approvals have been obtained. Not Applicable The details of the IRB/oversight body that provided approval or exemption for the research described are given below: This study involved the retrospective analysis of fully anonymized H\&E-stained breast cancer slides obtained from the University Medical Center Utrecht (UMCU). According to the Medical Ethics Review Committee (METC) of UMC Utrecht, the use of anonymized, retrospective data does not require ethical approval or informed consent. No identifiable information was accessed by the authors during or after data collection. The slides were accessed for research purposes in April 2024. All experiments were carried out in accordance with the Declaration of Helsinki and the International Ethical Guidelines for Biomedical Research Involving Human Subjects by the Council for International Organizations of Medical Sciences (CIOMS). I confirm that all necessary patient/participant consent has been obtained and the appropriate institutional forms have been archived, and that any patient/participant/sample identifiers included were not known to anyone (e.g., hospital staff, patients or participants themselves) outside the research group so cannot be used to identify individuals. Not Applicable I understand that all clinical trials and any other prospective interventional studies must be registered with an ICMJE-approved registry, such as ClinicalTrials.gov. I confirm that any such study reported in the manuscript has been registered and the trial registration ID is provided (note: if posting a prospective study registered retrospectively, please provide a statement in the trial ID field explaining why the study was not registered in advance). Not Applicable I have followed all appropriate research reporting guidelines, such as any relevant EQUATOR Network research reporting checklist(s) and other pertinent material, if applicable. Not Applicable Data Availability Statement The dataset used in this study was obtained from UMCU and is not publicly available due to patient privacy regulations. It is available from the corresponding author upon reasonable request and with appropriate institutional approvals. All experiments were conducted using PyTorch on a high-performance computing system provided by UMCU. The source code for the framework is publicly available at https://github.com/tueimage/LVI-Detection.