Purpose <p>Accurate differentiation of human epidermal growth factor receptor 2 (HER2)-zero, HER2-low, and HER2-positive breast cancer (BC) is crucial for guiding therapeutic decisions, particularly with the development of novel HER2-targeted treatments. This study aimed to evaluate whether intravoxel incoherent motion (IVIM) imaging histogram parameters could provide additional diagnostic value in distinguishing among these HER2 expression subtypes beyond conventional BI-RADS assessments.</p> Materials and methods <p>This retrospective single-center study included 181 breast cancer patients (30 HER2-zero, 107 HER2-low, and 44 HER2-positive). All patients underwent preoperative breast MRI on a 3.0-T scanner, including conventional sequences (T1-weighted, T2-weighted, dynamic contrast-enhanced and diffusion-weighted imaging) and intravoxel incoherent motion (IVIM) imaging. Histogram features derived from IVIM parameters—true diffusion coefficient (<i>D</i>), pseudo-diffusion coefficient (<i>D</i>*), and perfusion fraction (<i>f</i>)—were extracted and compared across the HER2 subgroups. Two diagnostic models were constructed: a conventional model based on BI-RADS features and a combined model incorporating both BI-RADS and IVIM histogram features. Statistical analyses included univariate logistic regression, Akaike information criterion (AIC), area under the receiver operating characteristic curve (AUC), integrated discrimination improvement (IDI), net reclassification index (NRI), and DeLong test.</p> Results <p>Conventional BI-RADS features alone could not distinguish between HER2-zero and HER2-low tumors. Although D_Entropy demonstrated modest discrimination (AUC = 0.569), this comparison was analyzed descriptively without constructing predictive models. Consequently, conventional and combined models were not constructed for this comparison. For HER2-low versus HER2-positive differentiation, the combined model (including D_Skewness and tumor size) achieved significantly higher performance (AUC = 0.727) than the conventional model (AUC = 0.587). Similarly, for HER2-zero versus HER2-positive, the combined model (D_Skewness, tumor size, and axillary lymph node [ALN] status) outperformed the conventional model (tumor size and ALN status), with AUCs of 0.730 versus 0.700. Improvements in diagnostic performance with the combined model were statistically confirmed by the DeLong test, NRI, and IDI metrics in the HER2-low versus HER2-positive and HER2-zero versus HER2-positive comparisons.</p> Conclusions <p>IVIM histogram parameters, particularly D_Skewness and D_Entropy, may enhance the ability to differentiate HER2 expression subtypes in breast cancer when integrated with conventional BI-RADS features. The combined model improves diagnostic performance, supporting its potential utility in refining HER2 status evaluation and optimizing personalized treatment strategies.</p>