Prediction of lymphovascular invasion in breast carcinoma using MRI-based radiomics: A systematic review and meta-analysis.

Abstract

e12560 Background: Lymphovascular invasion (LVI) correlates with poor breast cancer prognosis, indicating metastatic progression status. Detection of LVI prior to surgery is clinically challenging. This meta-analysis evaluates the diagnostic value of MRI radiomics in detecting LVI in breast cancer patients. Methods: Studies were retrieved from 4 databases: PubMed, Scopus, Embase, and Web of Science. Eligible studies were those with radiomics models derived from machine learning algorithms, and studies using clinical-only models were excluded. Data regarding LVI status, feature selection, machine learning models, and diagnostic performance were extracted from externally validated testing cohorts. The pooled effect size was analyzed using the inverse-variance method, and a funnel plot was generated to assess publication bias. HSROC curves were generated using Bayesian estimation. This project was prospectively registered with PROSPERO (CRD420251186235). Results: 23 studies comprising 2,167 cases were included in the meta-analysis. The pooled area under the curve (AUC) for detecting LVI was 0.76 (95% CI 0.71 - 0.80) on random-effects model. Heterogeneity was noted among the studies (I 2 = 33%). Conclusions: Integration of MRI radiomics serves as a valuable clinical tool in determining LVI in breast cancer patients, aiding clinicians in earlier decision-making. It demonstrates potential as a noninvasive adjunct to enhance preoperative assessment and guide better treatment plans.