According to global cancer statistics in 2022, breast cancer is the most common cancer among women worldwide [1]. Human epidermal growth factor receptor 2 (HER2) correlates with breast cancer prognosis and critically influences therapeutic decision-making [2,3]. The detection of HER2 status is typically conducted using two techniques, immunohistochemistry (IHC) and in situ hybridization (ISH). Numerous studies have demonstrated that patients with HER2-positive (IHC 3+ or IHC 2+ with ISH amplification) breast cancer have benefited clinically through the application of anti-HER2 targeting agents [[4], [5], [6]]. In contrast, patients with HER2-negative (IHC 0, 1+, or 2+ without ISH amplification) did not respond to HER2-targeted therapy [7]. However, in recent years, the research findings of Modi et al. have prompted a reexamination of the conventional binary classification system for HER2 status. The novel antibody-drug conjugate trastuzumab-deruxtecan (T-DXd) has demonstrated remarkable improvements in clinical outcomes for patients with metastatic breast cancer who exhibit HER2 IHC 1+ or 2+ without ISH amplification (i.e., HER2-low) [8]. Statistically, such patients make up a large segment of the breast cancer population, ranging from 45 % to 55 % [5,9]. Since T-DXd was approved to treat HER2-low breast cancer, the distinction between HER2-low and HER2-zero has become clinically significant. HER2-low, as a unique biomarker, provides an important basis for screening patient populations suitable for T-DXd treatment [10]. Therefore, the accurate determination of HER2 status is a key component in the treatment management of breast cancer patients.

The 2023 guideline update reiterates the recommendation for HER2 testing in primary invasive breast cancer [11]. However, the HER2 heterogeneity present in breast cancer poses a great challenge to accurately evaluate HER2 status as it may lead to inconsistencies between HER2 IHC and ISH assays [12,13]. Lambein et al. found that pathologists had low concordance (only 26 %) in the low-scoring range (0 and 1+) HER2 IHC assays [14]. This suggests that when biopsying tumors with HER2 heterogeneity, semi-quantitative HER2 assay results may not truly reflect the overall biological status of the tumor [15]. Therefore, the development of sensitive and reproducible quantitative analysis methods capable of characterizing intratumoral heterogeneity (ITH) is important for improving the accuracy and reliability of HER2 status identification.

As the most sensitive imaging technique for diagnosing breast cancer, magnetic resonance imaging (MRI) excels in identifying aggressive disease [16]. Radiomics and deep learning (DL) have been shown to perform well in assessing the accuracy of HER2 status [17], but they typically treat the tumor as a homogeneous whole, failing to capture heterogeneity within the tumor. Habitat analysis has the potential to non-invasively identify ITH by accurately segmenting tumors into subregions that are representative of biodiversity [18]. Habitat analysis has been successfully applied to predict invasiveness in lung adenocarcinoma and pathologic complete response in neoadjuvant chemotherapy for breast cancer [19,20].

To date, no studies have combined the habitat radiomics model with the DL model for predicting HER2 status in breast cancer. We developed an MRI-based habitat model to quantify ITH and compared its predictive ability for HER2 status with radiomics. In addition, we integrated clinical characteristics, habitat model prediction score, and DL model prediction score to establish a combined model for accurate prediction of HER2 expression status.