Performance of MRI-based deep learning models in differentiation of triple negative breast cancer from other breast cancer subtypes: A systematic review and meta-analysis

Abstract

Background and aim: Triple negative breast cancer (TNBC) is an aggressive subtype of breast cancer with limited targeted therapies. Deep learning (DL) applied to magnetic resonance imaging (MRI) offers a promising noninvasive alternative to biopsy. This systematic review and meta-analysis aimed to synthesize current evidence on the diagnostic performance of MRI-based DL models for identifying TNBC. Material and methods: Comprehensive searches of PubMed, Scopus, and Web of Science were conducted up to December 5, 2025. Eligible studies evaluated histologically confirmed breast cancer using MRI and DL-based models for distinguishing TNBC from non-TNBC. Study quality was assessed using the METRICS. Pooled diagnostic estimates were computed using a bivariate random-effects model in Stata version 18. Sensitivity analyses and publication bias tests were performed. Certainty of evidence was evaluated by GRADE. Results: Nine studies comprising 2985 patients met inclusion criteria. Pooled estimates in the validation cohorts demonstrated an AUC of 0.85 (95 CI: 0.81-0.88), sensitivity of 0.83 (95 CI: 0.76-0.89), specificity of 0.87 (95 CI: 0.82-0.91), positive likelihood ratio of 6.47 (95 CI: 3.98-10.54), negative likelihood ratio of 0.24 (95 CI: 0.18-0.33), and diagnostic odds ratio of 26.53 (95 CI: 13.86-50.81). Heterogeneity was moderate (I2=30.2), and no publication bias was detected. Sensitivity analysis revealed no influential individual study. Evidence certainty was rated as low. Conclusion: DL models applied to breast MRI showed potential for noninvasive TNBC identification with high diagnostic accuracy. However, limited external validation and variability in methodological quality highlight the need for standardized, multicenter studies before clinical implementation can be considered.