Multiparametric MRI and artificial intelligence for non-invasive HER2 assessment in breast cancer: a comprehensive review
Source: PubMed Central Open Access, NCBI / U.S. National Library of Medicine
Objective To systematically evaluate the utility of multiparametric MRI (mpMRI) combined with radiomics and artificial intelligence (AI) for non-invasive assessment of human epidermal growth factor receptor 2 (HER2) status in breast cancer. Methods Following the PRISMA 2020 guidelines, we searched PubMed, Web of Science, and Embase for studies published between January 2015 and December 2025 (last search: 27 June 2026). After duplicate removal using Zotero, two reviewers independently screened the records. Thirty-four studies were included and assessed using PROBAST, with the 16 radiomics studies additionally evaluated using the RQS. We qualitatively synthesized evidence on conventional MRI, diffusion imaging (DWI/ADC, IVIM, DKI, and DSI), DCE-MRI, metabolic MRI, and radiomics/AI approaches without performing a meta-analysis. Results mpMRI sequences provide complementary insights into HER2 status. Advanced diffusion techniques (DKI: *p* = 0.003; DSI: AUC range, 0.700–0.721) and DCE-MRI radiomics provide predictive value by capturing microstructural and dynamic tumor characteristics. Multiparametric radiomics incorporating peritumoral features achieved AUCs of 0.84–0.923 for three-level HER2 classification, while radiomics and deep learning distinguished HER2-low from HER2-zero tumors with AUCs up to 0.892. Machine learning and deep learning models demonstrated moderate cross-center generalizability (external AUC, 0.66–0.84). However, variability in reference standards, retros
Abstract
Objective To systematically evaluate the utility of multiparametric MRI (mpMRI) combined with radiomics and artificial intelligence (AI) for non-invasive assessment of human epidermal growth factor receptor 2 (HER2) status in breast cancer. Methods Following the PRISMA 2020 guidelines, we searched PubMed, Web of Science, and Embase for studies published between January 2015 and December 2025 (last search: 27 June 2026). After duplicate removal using Zotero, two reviewers independently screened the records. Thirty-four studies were included and assessed using PROBAST, with the 16 radiomics studies additionally evaluated using the RQS. We qualitatively synthesized evidence on conventional MRI, diffusion imaging (DWI/ADC, IVIM, DKI, and DSI), DCE-MRI, metabolic MRI, and radiomics/AI approaches without performing a meta-analysis. Results mpMRI sequences provide complementary insights into HER2 status. Advanced diffusion techniques (DKI: *p* = 0.003; DSI: AUC range, 0.700–0.721) and DCE-MRI radiomics provide predictive value by capturing microstructural and dynamic tumor characteristics. Multiparametric radiomics incorporating peritumoral features achieved AUCs of 0.84–0.923 for three-level HER2 classification, while radiomics and deep learning distinguished HER2-low from HER2-zero tumors with AUCs up to 0.892. Machine learning and deep learning models demonstrated moderate cross-center generalizability (external AUC, 0.66–0.84). However, variability in reference standards, retrospective study designs, and limited external validation remain major limitations. Conclusion mpMRI combined with radiomics and AI demonstrates promising value for HER2 assessment, with internal testing often AUCs exceeding 0.90 but less consistent on external validation. It complements rather than replaces biopsy by capturing intratumoral heterogeneity and supporting treatment decisions. Future priorities include prospective multicenter validation, methodological standardization, and radiopathogenomic integration.