Abstract
Background:
Human epidermal growth factor receptor 2 (HER2) expression in breast cancer (BC) determines the options for targeted therapy. Multiparametric MRI (mpMRI) has the potential for noninvasive HER2 status prediction but remains underexplored.
Objectives:
To develop and validate an mpMRI-based nomogram incorporating clinicopathological factors for predicting HER2 status in BC patients.
Methods:
In this retrospective analysis, 313 BC patients were classified as HER2-overexpression, HER2-low, or HER2-zero on the basis of immunohistochemistry and fluorescence in situ hybridization. The patients were divided into training (n=232) and validation (n=81) datasets. Clinicopathological factors and mpMRI parameters were analyzed. Logistic regression identified independent predictors that were used to construct and validate the nomogram. Discrimination was evaluated by the area under the receiver operating characteristic curve (AUC).
Results:
CA125, Ki-67, the minimum apparent diffusion coefficient (ADC-min), and early-phase maximum enhancement (ME) differed significantly among the HER2 subgroups. The nomogram integrating these factors achieved AUCs of 0.762 (95% CI: 0.686–0.838) and 0.738 (95% CI: 0.594–0.882) in differentiating HER2-over/HER2-low from HER2-zero in the training and validation datasets, respectively. Differentiation between the HER2-over and HER2-low subtypes exhibited AUCs of 0.719 and 0.772, respectively.
Conclusions:
Our nomogram, which combines mpMRI and clinicopathological variables, effectively predicts HER2 expression in BC patients, providing a promising noninvasive clinical tool to guide targeted therapy selection.
Introduction
Breast cancer (BC) remains the most prevalent malignancy and a leading cause of cancer-related mortality among women worldwide (, ). The expression status of human epidermal growth factor receptor 2 (HER2) is a critical determinant of tumor biology, prognosis, and, crucially, therapeutic selection (). Although traditionally classified as HER2-positive [immunohistochemistry (IHC) 3+ or IHC 2+ with fluorescence in situ hybridization (FISH) amplification] or HER2-negative (IHC 0, 1+, or IHC 2+ without FISH amplification), the landscape has evolved significantly. While HER2-positive patients (10–15%) benefit from trastuzumab (–), the advent of novel antibody–drug conjugates (ADCs), particularly trastuzumab deruxtecan (T-DXd), has established HER2-low-expression (IHC 1+ or IHC 2+ without FISH amplification) as a distinct therapeutic entity with proven survival benefits (–). Consequently, the latest ASCO/CAP guidelines formally recognize HER2-low BC as an indication for T-DXd (, ).
Current HER2 status assessment relies on IHC and FISH performed on biopsy or surgical samples. However, this approach faces significant limitations: tumor heterogeneity may lead to sampling error, the acquired tissue is limited, and HER2 status can dynamically change during treatment (). Therefore, there is an urgent clinical need for reliable, noninvasive methods capable of comprehensively evaluating HER2 expression longitudinally.
Multiparametric magnetic resonance imaging (mpMRI), which refers to the combined use of morphological and functional sequences (e.g., T2WI, DWI, DCE-MRI), plays a pivotal role in BC diagnosis and characterization (, , ) in clinical practice. It offers the potential to noninvasively capture tumor heterogeneity and provide biomarkers reflecting the underlying pathophysiology. Although studies utilizing mpMRI features, sometimes combined with artificial intelligence (AI), have attempted to predict HER2 status, they have often focused on the traditional binary distinction (positive vs. negative) or solely on differentiating HER2-low from HER2-zero (, , ). Furthermore, the widespread clinical translation of complex AI models is often hindered by concerns regarding overfitting, a lack of reproducibility across diverse datasets, and limited interpretability (“black box” nature) (–). Crucially, the evolving therapeutic landscape necessitates models that can simultaneously differentiate HER2-zero, HER2-low, and HER2-overexpression subtypes to optimally guide targeted therapy selection (e.g., identifying candidates for T-DXd vs. other HER2-targeted agents vs. non-HER2-targeting regimens).
Consequently, the primary objective of this study was to develop and validate a clinically practical, noninvasive tool based on readily available mpMRI parameters and clinicopathological factors for predicting HER2 expression status in BC. Specifically, we first aimed to differentiate HER2-nonzero (combining HER2-overexpression and HER2-low) BC from HER2-zero BC, thus identifying patients who are potentially eligible for HER2-targeted therapies (including T-DXd). Second, we aimed to differentiate HER2-over from HER2-low BC within the nonzero group, thus refining therapeutic stratification (e.g., traditional HER2-targeted therapy vs. ADCs therapy). We hypothesized that a multivariate model combining key mpMRI biomarkers (such as diffusion and enhancement characteristics) and relevant clinicopathological factors (e.g., serum markers and the proliferation index) could provide a robust and interpretable predictive tool implemented as a nomogram for clinical use.
Materials and methods
Study population
This retrospective study was approved by the institutional review board of Meizhou People’s Hospital (2022-C-32), with a waiver of informed consent due to the use of anonymized data. This study was conducted in accordance with the Declaration of Helsinki and reported following the STARD guidelines. We initially screened 361 female BC patients who underwent preoperative mpMRI at our institution between December 2016 and October 2023. The inclusion criteria were as follows: (1) pathologically confirmed invasive BC, (2) pretreatment mpMRI, and (3) HER2 status assessed by IHC and FISH per the American Society of Clinical Oncology (ASCO) and College of American Pathologists (CAP) 2018 guidelines. The exclusion criteria were as follows: (1) incomplete clinical and pathological data (n=3), (2) incomplete MRI data or poor MR image quality, such as motion artifacts (n=39), (3) occult BC (n=1), and (4) recurrent BC (n=2). The patient enrollment process is presented in Figure 1. Finally, 313 patients were included and randomly divided into a training dataset (n=232) and a validation dataset (n=81) at a 3:1 ratio.
Figure 1
MRI examination
All the scans used a 3.0T MR system (Magnetom Skyra, Siemens) with a 16-channel breast coil. The sequences included diffusion-weighted imaging (DWI) with single-shot echo planar imaging with the following scanning parameters: TR, 4200 msec; TE, 62 msec; b values of 0 s/mm² and 800 s/mm²; matrix, 86×220; FOV, 149 mm×340 mm; and slice thickness and spacing of 4.0 mm and 0.8 mm, respectively. Apparent diffusion coefficient (ADC) maps were automatically generated by the scanner’s integrated software using a monoexponential model. Dynamic contrast-enhanced MRI (DCE-MRI) was performed using time-resolved imaging with the stochastic trajectory technique using the following scanning parameters: TR, 6400 msec; TE, 3.3 msec; matrix, 288×384; field of view (FOV), 288 mm×384 mm; and slice thickness and spacing of 2.0 mm and 0.4 mm, respectively. Gadopentetate dimeglumine was administered as a contrast agent using a high-pressure injector at a dose of 0.2 ml/kg via intravenous injection at a rate of 3.0 ml/s. A total of 34 phases were acquired, with the first phase initiated 17.7 seconds after contrast injection, followed by a single-phase scan time of 8.7 seconds. After the contrast agent was injected, 20 ml of saline was injected at the same flow rate to flush the tubing.
MRI analysis
All MRI data were processed using a Siemens syngo.via. The MRI examination images were randomly assigned to two radiologists with 5 years of experience who were blinded to the pathology results. MR image features were independently analyzed according to the imaging standardized reporting system BI-RADS. Consensus was reached by discussing any disagreements. In cases where consensus could not be reached, a third senior radiologist with more than 15 years of experience was consulted to make the final determination.
The maximum and minimum diameters of the largest section of the tumor were recorded on the DCE-MRI images. When multiple ipsilateral breast masses were present, the largest mass was evaluated on the second postcontrast images (approximately 90 seconds) after the contrast injection. The shape and margin were determined by combining T2-weighted imaging (T2WI) and DCE-MRI images. The lesion morphology was determined by combining DCE-T1WI (fat suppressed), DWI, and T2WI. The MR Tissue 4D software platform was employed to perform motion correction on qualitative model data and subsequently generated parametric maps of semiquantitative parameters derived from the DCE-MRI analysis. When evaluating the time–signal intensity curve (TIC), the early-phase maximum enhancement (ME) was classified as rapid (>100%), medium (50–100%), or slow (<50%), and the delayed-phase patterns were classified as persistent (>10% signal increase), plateau (between 10% increase and 10% decrease) and washout (>10% signal decrease). Regions of interest (ROIs) with a minimum area of 0.10 cm2 were manually drawn on the three consecutive maximum sections with the greatest enhancement areas of the tumors, avoiding visible blood vessels, obvious bleeding, and necrotic and cystic areas. The average ADC (ADC-avg), minimum ADC (ADC-min), and maximum ADC (ADC-max) values were recorded from the darkest tumor area. The ADC ratio refers to the ratio of the apparent diffusion coefficient between tumor and normal tissue. Background parenchymal enhancement (BPE) was graded as minimal, mild, moderate, or marked on the basis of the first postcontrast subtraction images. The ROI measurement strategy is detailed in Figure 2.
Figure 2
Clinicopathological characteristics
The clinical characteristics, including patient age and CEA, CA15-3, and CA125 levels, were obtained from the electronic medical records system. HER2 status was determined by IHC and FISH according to the ASCO/CAP testing guidelines. HER2 status was classified as follows: HER2-over-expression (IHC 2+ with FISH gene amplification or IHC 3+); HER2-low-expression (IHC score of 1+ or IHC score of 2+ without FISH gene amplification); and HER2-zero expression (IHC score of 0). The status of the estrogen receptor (ER) or progesterone receptor (PR) was defined as positive if nuclear staining was present in more than 1% of the tumor cells. Ki-67 was assessed using IHC, and a cutoff value of 20% was used. The reference cutoff values were 5 ng/ml, 25 U/ml, and 35 U/ml, respectively.
Statistical analysis
R version 4.4.1 and SPSS (version 25.0) were used for the statistical analysis. Data distribution was measured using the Kolmogorov–Smirnov test. Continuous variables were analyzed using a one-way analysis of variance or the Student’ s t test according to the distribution normality. Categorical variables were assessed using the chi-square test or Fisher’s exact test. Interreader agreement was classified as good if the intraclass correlation coefficient (ICC) was > 0.75. Variables significantly associated with HER2 expression status (P < 0.05) in the univariate analysis were included in the multivariate logistic regression model. The area under the receiver operating characteristic curve (AUC) was calculated for significant features, and the performance of the combined model was assessed. The optimal threshold for identifying HER2 status was determined using the Youden index. Decision curve analysis (DCA) was used to evaluate the clinical benefit of the predictive models. The calibration curve was used to evaluate the goodness-of-fit of the optimal model. All the statistical tests were two-sided, with a significance level of P < 0.05.
Results
Clinicopathologic characteristics of the participants
A total of 313 women with BC were included in this study, with ages ranging from 28 to 74 years old. The clinicopathological characteristics of the patients in both the training and validation cohorts are detailed in Table A.1, and statistically significant features are presented in Table 1. There were statistically significant differences in ER and PR status and the N stage among the HER2 subgroups in the training and validation datasets. In the training dataset, there were statistically significant differences in CA125, PR, and Ki-67 between the HER2-zero and non-HER2-zero groups (P < 0.05). There were statistically significant differences in ER, PR, and Ki-67 statuses and AJCC, N, and M stages between the HER2-over and HER2-low patients (P < 0.05). In the validation dataset, only the N stage was significantly different between the HER2-zero and non-HER2-zero groups (P < 0.05). There were statistically significant differences in CA125, ER and PR levels between HER2-over and HER2-low patients (P < 0.05).
Table 1
| Training dataset (n=232) | Validation dataset (n =81) | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| HER2-over (n=100) | HER2-low (n=83) | HER2-zero (n=49) | P | P 1 | P 2 | HER2-over (n=32) | HER2-low (n=31) | HER2-zero (n=18) | P | P 1 | P 2 | |
| CA125 | <0.001 | 0.935 | <0.001 | 0.130 | 0.047 | 0.819 | ||||||
| Negative | 90 (90.0%) | 75 (90.4%) | 29 (59.2%) | 29 (90.6%) | 22 (71.0%) | 15 (83.3%) | ||||||
| Positive | 10 (10.0%) | 8 (9.6%) | 20 (40.8%) | 3 (9.4%) | 9 (29.0%) | 3 (16.7%) | ||||||
| ER | 0.002 | <0.001 | 0.482 | 0.021 | 0.005 | 0.854 | ||||||
| Negative | 51 (51.0%) | 21 (25.3%) | 22 (44.9%) | 17 (53.1%) | 6 (19.4%) | 7 (38.9%) | ||||||
| Positive | 49 (49.0%) | 62 (74.7%) | 27 (55.1%) | 15 (46.9%) | 25 (80.6%) | 11 (61.1%) | ||||||
| PR | <0.001 | <0.001 | 0.038 | 0.023 | 0.006 | 0.717 | ||||||
| Negative | 61 (61.0%) | 28 (33.7%) | 32 (65.3%) | 18 (56.2%) | 7 (22.6%) | 8 (44.4%) | ||||||
| Positive | 39 (39.0%) | 55 (66.3%) | 17 (34.7%) | 14 (43.8%) | 24 (77.4%) | 10 (55.6%) | ||||||
| Ki-67 | <0.001 | 0.003 | 0.010 | 0.437 | 0.822 | 0.205 | ||||||
| Negative | 14 (14.0%) | 27 (32.5%) | 3 (6.1%) | 8 (25.0%) | 7 (22.6%) | 7 (38.9%) | ||||||
| Positive | 86 (86.0%) | 56 (67.5%) | 46 (93.9%) | 24 (75.0%) | 24 (77.4%) | 11 (61.1%) | ||||||
| N stage | <0.001 | <0.001 | 0.655 | 0.008 | 0.084 | 0.015 | ||||||
| 0 or 1 | 26 (26.0%) | 41 (49.4%) | 19 (38.8%) | 12 (37.5%) | 19 (61.3%) | 8 (44.4%) | ||||||
| 2 | 23 (23.0%) | 26 (31.3%) | 10 (20.4%) | 3 (9.4%) | 4 (12.9%) | 7 (38.9%) | ||||||
| 3 | 51 (51.0%) | 16 (19.3%) | 20 (40.8%) | 17 (53.1%) | 8 (25.8%) | 3 (16.7%) | ||||||
| M stage | 0.038 | 0.010 | 0.881 | 0.813 | 0.063 | 0.672 | ||||||
| 0 | 83 (83.0%) | 79 (95.2%) | 43 (87.8%) | 30 (93.8%) | 28 (90.3%) | 16 (88.9%) | ||||||
| 1 | 17 (17.0%) | 4 (4.8%) | 6 (12.2%) | 2 (6.2%) | 3 (9.7%) | 2 (11.1%) | ||||||
| AJCC Stage | 0.035 | 0.006 | 0.866 | 0.186 | 0.056 | 0.871 | ||||||
| 1 or 2 | 9 (9.0%) | 19 (22.9%) | 9 (18.4%) | 3 (9.4%) | 10 (32.3%) | 3 (16.7%) | ||||||
| 3 | 74 (74.0%) | 59 (71.1%) | 34 (69.4%) | 27 (84.4%) | 18 (58.1%) | 13 (72.2%) | ||||||
| 4 | 17 (17.0%) | 5 (6.0%) | 6 (12.2%) | 2 (6.3%) | 3 (9.7%) | 2 (11.1%) | ||||||
Clinicopathologic characteristics of patients in the HER2-zero, HER2-low, and HER2-over groups across the training and validation datasets.
HER2: human epidermal growth factor receptor 2, ER: estrogen receptor, PR: progesterone receptor. P, HER2 expression among the three groups. P1, HER2-overexpression group vs. HER2-low-expression group. P2, HER2-low and HER2-overexpression groups vs. HER2-zero- expression group.
Comparison of imaging features among groups
Interobserver agreement was good for tumor diameter (max/min), ADC-avg, ADC-min, and ADC-max (Table A.2). The comparisons of imaging features among the HER2-zero, HER2-low, and HER2-over groups in the training and validation datasets are listed in Table 2. The ADC-avg, ADC-min, and ADC-max values differed significantly among the HER2 subgroups (P < 0.05) in both datasets. There were statistically significant differences in the ADC-avg and ADC-min values between the HER2-zero and non-HER2-zero groups (P < 0.05) in the training and validation datasets. In the HER2-over group, the ADC-avg, ADC-min and ADC-max values were significantly higher than those in the HER2-low group (P < 0.05). Furthermore, all ADC values were higher in the HER2-low group (P < 0.05).
Table 2
| Training dataset (n=232) | Validation dataset (n=81) | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| HER2-over (n=100) | HER2-low (n=83) | HER2-zero (n=49) | P | P 1 | P 2 | HER2-over (n=32) | HER2-low (n=31) | HER2-zero (n=18) | P | P 1 | P 2 | |
| Max diameter | 4.81 ± 1.99 | 4.43 ± 1.76 | 4.63 ± 2.01 | 0.396 | 0.174 | 0.977 | 4.64 ± 1.82 | 4.31 ± 1.91 | 4.18 ± 1.83 | 0.656 | 0.489 | 0.545 |
| Min diameter | 2.58 ± 1.21 | 2.70 ± 1.21 | 3.01 ± 1.27 | 0.126 | 0.513 | 0.053 | 2.31 ± 1.19 | 2.33 ± 1.16 | 2.32 ± 1.06 | 0.997 | 0.943 | 0.986 |
| Number of lesions | 0.017 | 0.004 | 0.781 | 0.234 | 0.919 | 0.089 | ||||||
| Single | 49 (49.0%) | 58 (69.9%) | 27 (56.2%) | 18 (56.3%) | 17 (54.8%) | 14 (77.8%) | ||||||
| Multiple | 51 (51.0%) | 25 (30.1%) | 22 (43.8%) | 14 (43.8%) | 14 (45.2%) | 4 (22.2%) | ||||||
| Morphology | 0.001 | 0.020 | 0.007 | 0.409 | 0.649 | 0.211 | ||||||
| Mass | 61 (61.0%) | 64 (77.1%) | 43 (87.8%) | 21 (65.6%) | 22 (71.0%) | 15 (83.3%) | ||||||
| NME | 39 (39.0%) | 19 (22.9%) | 6 (12.2%) | 11 (34.3%) | 9 (29.0%) | 3 (16.7%) | ||||||
| Enhancement pattern | 0.264 | 0.195 | 0.462 | NA | NA | NA | ||||||
| Homogeneous | 2 (2.0%) | 0 (0.0%) | 0 (0.0%) | 0 (0.0%) | 0 (0.0%) | 0 (0.0%) | ||||||
| Nonhomogeneous | 98 (98.0) | 83 (100.0%) | 49 (100.0%) | 32 (100.0%) | 31 (100.0%) | 18 (100.0%) | ||||||
| Shape | 0.878 | 0.816 | 0.653 | 0.793 | 0.974 | 0.496 | ||||||
| Regular | 3 (3.0%) | 3 (3.6%) | 1 (2.0%) | 2 (6.2%) | 2 (6.5%) | 2 (11.1%) | ||||||
| Irregular | 97 (97.0%) | 80 (96.4%) | 48 (98.0%) | 30 (93.8%) | 29 (93.5%) | 16 (88.9%) | ||||||
| BPE | 0.117 | 0.049 | 0.529 | 0.582 | 0.237 | 0.917 | ||||||
| 1 | 19 (19.0%) | 21 (25.3%) | 7 (14.3%) | 7 (21.9%) | 3 (9.7%) | 4 (22.2%) | ||||||
| 2 | 52 (52.0%) | 28 (33.7%) | 22 (44.9%) | 13 (40.6%) | 14 (45.2%) | 7 (38.9%) | ||||||
| 3 | 20 (20.0%) | 18 (21.7%) | 14 (28.6%) | 8 (35.0%) | 5 (16.1%) | 4 (22.2%) | ||||||
| 4 | 9 (9.0%) | 16 (19.3%) | 6 (12.2%) | 4 (12.5%) | 9 (29.0%) | 3 (16.7%) | ||||||
| TIC | 0.634 | 0.798 | 0.333 | 0.768 | 0.539 | 0.622 | ||||||
| Persistent | 2 (2.0%) | 3 (3.6%) | 3 (6.1%) | 0 (0.0%) | 1 (3.2%) | 1 (5.6%) | ||||||
| Plateau | 35 (35.0%) | 29 (34.9%) | 20 (40.8%) | 9 (28.1%) | 7 (22.6%) | 4 (22.2%) | ||||||
| Washout | 63 (63.0%) | 51 (61.4%) | 26 (53.1%) | 23 (71.9%) | 23 (74.2%) | 13 (72.2%) | ||||||
| Early-phase ME | 0.016 | 0.041 | 0.026 | 0.378 | 0.157 | 0.637 | ||||||
| Slow | 0 (0.0%) | 0 (0.0%) | 0 (0.0%) | 0 (0.0%) | 0 (0.0%) | 0 (0.0%) | ||||||
| Medium | 4 (4.0%) | 10 (12.0%) | 9 (18.4%) | 2 (6.3%) | 0 (0.0%) | 1 (5.6%) | ||||||
| Rapid | 96 (96.0%) | 73 (88.0%) | 40 (81.6%) | 30 (93.8%) | 31 (100.0%) | 17 (94.4%) | ||||||
| ADC-avg | 0.876 ± 0.144 | 0.801 ± 0.182 | 0.784 ± 0.154 | 0.001 | 0.002 | 0.028 | 0.844 ± 0.140 | 0.747 ± 0.108 | 0.712 ± 0.132 | 0.001 | 0.003 | 0.021 |
| ADC-min | 0.801 ± 0.132 | 0.731 ± 0.165 | 0.708 ± 0.151 | <0.00 | 0.002 | 0.013 | 0.787 ± 0.135 | 0.694 ± 0.100 | 0.646 ± 0.114 | <0.001 | 0.002 | 0.005 |
| ADC-max | 0.947 ± 0.179 | 0.872 ± 0.209 | 0.870 ± 0.194 | 0.014 | 0.010 | 0.172 | 0.909 ± 0.160 | 0.802 ± 0.133 | 0.772 ± 0.156 | 0.003 | 0.005 | 0.045 |
Imaging characteristics of patients in the HER2-zero, HER2-low, and HER2-over groups across the training and validation datasets.
ME: maximum degree of tumor enhancement, NA: not applicable. P, HER2 expression among the three groups. P1, HER2-overexpression group vs. HER2-low-expression group. P2, HER2-low and HER2-overexpression groups vs. HER2-zero-expression group.
In the training dataset, a comparison of qualitative features among the three groups revealed significant differences in lesion number, morphology, and early-phase ME (P < 0.05). The morphology and early-phase ME significantly differed between the HER2-zero group and the non-HER2-zero group. Compared with patients with HER2-low BC, patients with HER2-overexpression BC more often presented with multiple lesions, NME, and rapid early-phase maximum enhancement.
Predictive efficiency of clinicopathological and radiological signatures
A univariable logistic regression analysis of the training dataset revealed that CA125 (P < 0.001), PR (P = 0.040), Ki-67 (P = 0.017), ADC-avg (P = 0.029), ADC-min (P = 0.014), the ADC ratio (P = 0.026), and early-phase ME (P = 0.031) were associated with being able to differentiate the HER2-low and HER2-over groups from the HER2-zero group (Table 3). Variables with P < 0.05 in the univariable analysis were subsequently entered into the multivariable logistic regression model. On the basis of the forward stepwise selection method used in the multivariate logistic regression analysis, CA125 (P < 0.001), Ki-67 (P = 0.032), ADC-min (P = 0.022), and early-phase ME (P = 0.015) were independent risk factors for differentiating the HER2-low and HER2-over groups from the HER2-zero group.
Table 3
| Variables | Univariate logistic regression | Multivariate logistic regression | ||
|---|---|---|---|---|
| OR (95% CI) | P | OR (95% CI) | P | |
| Lesion number, single vs. multiple | 0.931 (0.481,1.735) | 0.782 | NA | |
| Morphology, mass vs. NME | 0.456 (0.192,1.082) | 0.075 | NA | |
| Stage, 1 or 2 vs. others | 1.245 (0.544,2.849) | 0.603 | NA | |
| N stage, 0 or 1 vs. others | 1.097 (0.573,2.097) | 0.781 | NA | |
| ER, negative vs. positive | 1.256 (0.665,2.373) | 0.482 | NA | |
| PR, negative vs. positive | 1.988 (1.032,3.830) | 0.040 | NA | |
| Ki-67, negative vs. positive | 0.226 (0.067,0.764) | 0.017 | 0.245 (0.068,0.887) | 0.032 |
| CA125, negative vs. positive | 0.158 (0.075,0.335) | <0.001 | 0.162 (0.074,0.355) | <0.001 |
| ME | 2.716 (1.098,6.718) | 0.031 | 3.375 (1.263,9.020) | 0.015 |
| ADC-avg | 10.474 (1.273,86.196) | 0.029 | NA | |
| ADC-min | 16.567 (1.749,156.953) | 0.014 | 16.737 (1.491,187.907) | 0.022 |
| ADC-max | 3.340 (0.590,18.899) | 0.173 | NA | |
| ADC-ratio | 26.516 (1.310,536.704) | 0.033 | NA | |
Results of the univariate and multivariate logistic regression analyses.
Variables that emerged as significant (P < 0.05) in the univariable analyses were included in the multivariable analysis. NA: not applicable.
Nomogram establishment
For differentiating HER2-over and HER2-low BC from HER2-zero BC, the combined model was used to develop a nomogram that included CA125, Ki-67, ADC-min and early-phase ME; this model achieved an AUC of 0.762 with a sensitivity of 0.743 and a specificity of 0.653 in the training dataset. In the validation dataset, the combined model had an AUC of 0.738, with a sensitivity of 0.683 and a specificity of 0.778. For differentiating between HER2-over and HER2-low BC, the same combined model achieved an AUC of 0.719 with a sensitivity of 0.743 and a specificity of 0.653 in the training dataset and an improved AUC of 0.772 with a sensitivity of 0.656 and a specificity of 0.774 in the validation dataset. The developed nomogram provides a user-friendly tool for discriminating between the HER2-zero and non-HER2-zero groups and between the HER2-over and HER2-low groups on the basis of the combination of key clinicopathological and imaging variables, including CA125, Ki-67, ADC-min, and early-phase ME. The calibration curve of the combined model shows a good fit between the calibration prediction curve and the ideal curve, indicating that the combined model has high predictive performance. The DCA curve demonstrated that the model is clinically useful in decision-making. Figure 3 and Figure 4 present the predictive performance of the combined model in the training cohort, including the nomograms, ROC curves, calibration curves, and decision curve analyses for both classification tasks: Task 1 (HER2-zero vs. non-HER2-zero) and Task 2 (HER2-overexpression vs. HER2-low-expression). The ROC curves of the combined model in the validation cohort are presented in Supplementary Figure A.1.
Figure 3
Figure 4
Discussion
In this study, we developed and validated a clinically practical nomogram integrating readily available mpMRI parameters (ADC-min, early-phase ME) and clinicopathological factors (CA125, Ki-67) to noninvasively predict HER2 expression status in BC. Our key findings suggest that this combined model effectively differentiates 1) HER2-nonzero (HER2-overexpression and HER2-low) BC from HER2-zero BC (AUC ~0.76 training, ~0.74 validation) and 2) HER2-overexpression BC from HER2-low BC within the nonzero group (AUC~0.72 training, ~0.77 validation). This “two-step” stratification strategy may help identify patients who could potentially benefit from HER2-targeted therapies, including both traditional agents and novel ADCs such as T-DXd, thereby offering a valuable tool to guide personalized treatment selection.
Our results indicate that Ki-67 is a significant independent predictor of HER2 status, which is consistent with the findings of prior studies (–). We observed a trend toward an inverse relationship between the Ki-67 proliferation index and the HER2 expression level (lower Ki-67 levels are associated with HER2-over vs. HER2-low), supporting the notion that HER2-low tumors may represent a biologically distinct entity with potentially less aggressive proliferative activity than HER2-overexpression tumors (, ). This finding underscores the relevance of proliferative activity in HER2 subtyping.
The significant association between elevated serum CA125 levels and HER2 status is notable. Although traditionally linked to ovarian cancer, elevated CA125 in BC has been associated with more aggressive subtypes and poorer prognosis (–). Our results suggest a potential relationship between CA125 and HER2 pathway biology that warrants further mechanistic investigation. This highlights the potential utility of incorporating routine serum markers into HER2 status prediction models.
ADC-min has emerged as a robust independent imaging biomarker. We observed the highest ADC-min values in the HER2-over group and the lowest in the HER2-zero group, with the HER2-low group exhibiting intermediate values. This aligns with studies reporting higher ADC values in HER2-positive BC than in HER2-negative BC (, ), although discrepancies exist in the literature regarding HER2-low BC (–). ADC-min may reflect areas of highest cellularity or fibrosis within the heterogeneous tumor. The superior performance of ADC-min over ADC-avg or ADC-max in our multivariate analysis underscores its potential advantage in reflecting the most restricted diffusion component, which is potentially less confounded by overall tumor heterogeneity (). The finding that HER2-over tumors had higher ADC-min values than HER2-low tumors was somewhat unexpected based purely on aggressiveness but may be related to specific microenvironmental characteristics influenced by HER2 signaling.
Rapid early-phase ME was an independent predictor for distinguishing HER2 status groups (nonzero vs. zero; over vs. low). This may be explained by HER2 overexpression activates proangiogenic pathways (e.g., VEGF via PI3K-AKT/mTOR), leading to disordered, hyperpermeable neovasculature (). This likely contributes to the observed association between HER2 expression (particularly overexpression) and intense early enhancement. Although some prior studies reported no significant correlation (), our results support the functional relevance of DCE-MRI kinetics in characterizing HER2-driven vascular phenotypes.
The predictive performance of our nomogram (AUC values of 0.72–0.77 across tasks and datasets) moderate discriminatory ability. Although some recent studies utilizing complex radiomic or AI features have reported higher AUCs (e.g., 0.78–0.89) (, –), our model has potential advantages. (1) Clinical interpretability and practicality: the model is based on a limited number of variables routinely assessed in clinical practice (standard serum markers, IHC markers, basic DWI and DCE-MRI metrics). In contrast, more complex AI-based models often face challenges related to interpretability, reproducibility, and generalizability (, ). (2) Direct relevance to evolving treatment paradigms: unlike models focused solely on HER2-positive vs. HER2-negative (, , ) or HER2-low vs. HER2-zero (), our model explicitly addresses the critical three-tier stratification (zero/low/over) that is required by T-DXd therapy (–, ). The “two-step” approach efficiently identifies potential candidates for any HER2-targeted therapy (Step 1: Nonzero vs. Zero) and further refines therapy selection within eligible patients (Step 2: Over vs. Low).
Limitations
Our study has several limitations. First, its retrospective, single-center design may introduce selection bias, and external validation in larger, multicenter prospective cohorts is essential to confirm generalizability. Second, while we employed robust statistical methods (logistic regression) using clinically meaningful features, we did not explore deep learning radiomics. Although radiomics may extract additional information (, , ), concerns regarding feature stability, reproducibility, and clinical interpretability persist (, ). Our pragmatic approach prioritized clinically translatable variables. Third, the ADC measurement was based on ROIs placed on the most solid part of key slices of the tumor rather than whole-tumor segmentation. Future studies investigating volumetric ADC metrics might provide additional insights. Finally, the sample size, while sufficient for model development, limits subgroup analyses, and the validation cohort size was relatively modest.
Conclusion
In conclusion, we successfully developed and validated a practical nomogram integrating mpMRI parameters (ADC-min and early-phase ME) and clinicopathological factors (CA125 and Ki-67) for the noninvasive prediction of HER2 expression status in BC. This tool effectively stratifies patients into HER2-zero, HER2-low, and HER2-overexpression categories, directly informing eligibility for current and emerging HER2-targeted therapies. The model’s reliance on routinely available clinical and imaging data enhances its potential for widespread clinical adoption as a valuable adjunct to pathological assessment, aiding in personalized therapeutic decision-making. Future prospective multicenter studies are warranted to further validate and refine this approach.
Statements
Data availability statement
The original contributions presented in the study are included in the article/Supplementary Material. Further inquiries can be directed to the corresponding authors.
Ethics statement
The studies involving humans were approved by Meizhou People’s Hospital (2022-C-32). The studies were conducted in accordance with the local legislation and institutional requirements. Written informed consent for participation was not required from the participants or the participants’ legal guardians/next of kin in accordance with the national legislation and institutional requirements. Written informed consent was obtained from the individual(s) for the publication of any potentially identifiable images or data included in this article.
Author contributions
YC: Writing – original draft, Software, Formal analysis, Data curation. XFC: Funding acquisition, Resources, Formal analysis, Writing – original draft, Methodology. BY: Data curation, Validation, Writing – review & editing. XZ: Validation, Data curation, Writing – review & editing. HZ: Writing – review & editing, Visualization, Formal analysis. XHC: Formal analysis, Visualization, Writing – review & editing. XGC: Writing – review & editing, Conceptualization. ZD: Writing – review & editing, Conceptualization. ZY: Conceptualization, Supervision, Writing – review & editing.
Funding
The author(s) declared that financial support was received for this work and/or its publication. Meizhou People’s Hospital Research Cultivation Project (PY-C2022011).
Conflict of interest
The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Generative AI statement
The author(s) declared that generative AI was not used in the creation of this manuscript.
Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.
Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fonc.2026.1824808/full#supplementary-material
References
1
BhangdiaKMayMLKocarnikJMPritchettNCristAPenberthyLet al. Global, regional, and national burden of breast cancer among females, 1990–2023, with forecasts to 2050: a systematic analysis for the global burden of disease study 2023. Lancet Oncol. (2026) 27:302–26. doi: 10.1016/S1470-2045(25)00730-2
2
BrayFLaversanneMSungHFerlayJSiegelRLSoerjomataramIet al. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin. (2024) 74:229–63. doi: 10.3322/caac.21834
3
LoiblSGianniL. HER2-positive breast cancer. Lancet. (2017) 389:2415–29. doi: 10.1016/S0140-6736(16)32417-5
4
VialeGBasikMNiikuraNTokunagaEBruckerSPenault-LlorcaFet al. Retrospective study to estimate the prevalence and describe the clinicopathological characteristics, treatments received, and outcomes of HER2-low breast cancer. ESMO Open. (2023) 8:101615. doi: 10.1016/j.esmoop.2023.101615
5
MannRMChoNMoyL. Breast MRI: state of the art. Radiology. (2019) 292:520–36. doi: 10.1148/radiol.2019182947
6
ZhangHPengY. Current biological, pathological and clinical landscape of HER2-low breast cancer. Cancers. (2022) 15:126–40. doi: 10.3390/cancers15010126
7
ModiSJacotWYamashitaTSohnJVidalMTokunagaEet al. Trastuzumab deruxtecan in previously treated HER2-low advanced breast cancer. N Engl J Med. (2022) 387:9–20. doi: 10.1056/NEJMoa2203690
8
MartínMPandiellaAVargas-CastrillónEDíaz-RodríguezEIglesias-HernangómezTMartínez CanoCet al. Trastuzumab deruxtecan in breast cancer. Crit Rev Oncology/Hematology. (2024) 198:104355. doi: 10.1016/j.critrevonc.2024.104355
9
FanPXuK. Antibody-drug conjugates in breast cancer: Marching from HER2-overexpression into HER2-low. Biochim Biophys Acta (BBA) - Rev Cancer. (2023) 1878:188849. doi: 10.1016/j.bbcan.2022.188849
10
WolffACSomerfieldMRDowsettMHammondMEHHayesDFMcShaneLMet al. Human epidermal growth factor receptor 2 testing in breast cancer: ASCO–college of American pathologists guideline update. J Clin Oncol. (2023) 41:3867–72. doi: 10.1200/JCO.22.02864
11
AgostinettoEReditiMFimereliDDebienVPiccartMAftimosPet al. HER2-low breast cancer: molecular characteristics and prognosis. (2021) 13:2824. doi: 10.3390/cancers13112824
12
ZhouJTanHLiWLiuZWuYBaiYet al. Radiomics signatures based on multiparametric MRI for the preoperative prediction of the HER2 status of patients with breast cancer. Acad Radiol. (2021) 28:1352–60. doi: 10.1016/j.acra.2020.05.040
13
MoyLEliasKPatelVLeeJBabbJSTothHKet al. Is breast MRI helpful in the evaluation of inconclusive mammographic findings? Am J Roentgenol. (2009) 193:986–93. doi: 10.2214/AJR.08.1229
14
MarinoMAHelbichTBaltzerPPinker‐DomenigK. Multiparametric MRI of the breast: A review. J Magn Reson Imaging. (2018) 47:301–15. doi: 10.1002/jmri.25790
15
WangHSangLXuJHuangCHuangZ. Multiparametric MRI-based radiomic nomogram for predicting HER-2 2+ status of breast cancer. Heliyon. (2024) 10:e29875. doi: 10.1016/j.heliyon.2024.e29875
16
ZhanTDaiJLiY. Noninvasive identification of HER2-zero, -low, or -overexpressing breast cancers: Multiparametric MRI-based quantitative characterization in predicting HER2-low status of breast cancer. Eur J Radiol. (2024) 177:111573. doi: 10.1016/j.ejrad.2024.111573
17
ChoiYYuWNagarajanMBTengPGoldinJGRamanSSet al. Translating AI to clinical practice: overcoming data shift with explainability. RadioGraphics. (2023) 43:e220105. doi: 10.1148/rg.220105
18
MoassefiMRouzrokhPConteGMVahdatiSFuTTahmasebiAet al. Reproducibility of deep learning algorithms developed for medical imaging analysis: a systematic review. J Digital Imaging. (2023) 36:2306–12. doi: 10.1007/s10278-023-00870-5
19
SalahuddinZWoodruffHCChatterjeeALambinP. Transparency of deep neural networks for medical image analysis: a review of interpretability methods. Comput Biol Med. (2022) 140:105111. doi: 10.1016/j.compbiomed.2021.105111
20
PengYZhangXQiuYLiBYangZHuangJet al. Development and validation of MRI radiomics models to differentiate HER2-zero, -low, and -positive breast cancer. Am J Roentgenology. (2024) 222:e2330603. doi: 10.2214/AJR.23.30603
21
DenkertCSeitherFSchneeweissALinkTBlohmerJ-UJustMet al. Clinical and molecular characteristics of HER2-low-positive breast cancer: pooled analysis of individual patient data from four prospective, neoadjuvant clinical trials. Lancet Oncol. (2021) 22:1151–61. doi: 10.1016/S1470-2045(21)00301-6
22
ZhangGRenCLiCWangYChenBWenLet al. Distinct clinical and somatic mutational features of breast tumors with high-, low-, or non-expressing human epidermal growth factor receptor 2 status. BMC Med. (2022) 20:142. doi: 10.1186/s12916-022-02346-9
23
LiJLiuLFengZWangXHuangYDaiHet al. Tumor markers CA15-3, CA125, CEA and breast cancer survival by molecular subtype: a cohort study. Breast Cancer. (2020) 27:621–30. doi: 10.1007/s12282-020-01058-3
24
BidardF-CHajageDBachelotTDelalogeSBrainECamponeMet al. Assessment of circulating tumor cells and serum markers for progression-free survival prediction in metastatic breast cancer: a prospective observational study. (2012) 14:R29. doi: 10.1186/bcr3114
25
ZhaoWLiXWangWChenBWangLZhangNet al. Association of preoperative serum levels of CEA and CA15-3 with molecular subtypes of breast cancer. Dis Markers. (2021) 2021:1–9. doi: 10.1155/2021/5529106
26
FangCCaoYLiuXZengX-TLiY. Serum CA125 is a predictive marker for breast cancer outcomes and correlates with molecular subtypes. Oncotarget. (2017) 8:63963–70. doi: 10.18632/oncotarget.19246
27
ParkSHChoiHHahnSY. Correlations between apparent diffusion coefficient values of invasive ductal carcinoma and pathologic factors on diffusion‐weighted MRI at 3.0 Tesla. J Magn Reson Imaging. (2015) 41:175–82. doi: 10.1002/jmri.24519
28
LeeHSKimSHKangBJBaekJESongBJ. Perfusion parameters in dynamic contrast-enhanced MRI and apparent diffusion coefficient value in diffusion-weighted MRI. Acad Radiol. (2016) 23:446–56. doi: 10.1016/j.acra.2015.12.011
29
MaoCHuLJiangWQiuYYangZLiuYet al. Discrimination between human epidermal growth factor receptor 2 (HER2)-low-expressing and HER2-overexpressing breast cancers: a comparative study of four MRI diffusion models. Eur Radiol. (2023) 34:2546–59. doi: 10.1007/s00330-023-10198-x
30
RoknsharifiSFishmanMDCAgarwalMDBrookAKharbandaVDialaniV. The role of diffusion weighted imaging as supplement to dynamic contrast enhanced breast MRI: Can it help predict Malignancy, histologic grade and recurrence? Acad Radiol. (2019) 26:923–9. doi: 10.1016/j.acra.2018.09.003
31
YoukJHSonEJChungJKimJ-AKimE. Triple-negative invasive breast cancer on dynamic contrast-enhanced and diffusion-weighted MR imaging: comparison with other breast cancer subtypes. Eur Radiol. (2012) 22:1724–34. doi: 10.1007/s00330-012-2425-2
32
CostantiniMBelliPRinaldiPBufiEGiardinaGFranceschiniGet al. Diffusion-weighted imaging in breast cancer: relationship between apparent diffusion coefficient and tumour aggressiveness. Clin Radiol. (2010) 65:1005–12. doi: 10.1016/j.crad.2010.07.008
33
AlameddineRSOtrockZKAwadaAShamseddineA. Crosstalk between HER2 signaling and angiogenesis in breast cancer: molecular basis, clinical applications and challenges. Curr Opin Oncol. (2013) 25:313–24. doi: 10.1097/CCO.0b013e32835ff362
34
LiuWYangYWangXLiCLiuCLiXet al. A comprehensive model outperformed the single radiomics model in noninvasively predicting the HER2 status in patients with breast cancer. Acad Radiol. (2025) 32:24–36. doi: 10.1016/j.acra.2024.07.048
35
ZhengSYangZDuGZhangYJiangCXuTet al. Discrimination between HER2-overexpressing, -low-expressing, and -zero-expressing statuses in breast cancer using multiparametric MRI-based radiomics. Eur Radiol. (2024) 34:6132–44. doi: 10.1007/s00330-024-10641-7
36
RamtohulTDjerroudiLLissavalidENhyCRedonLIkniLet al. Multiparametric MRI and radiomics for the prediction of HER2-zero, -low, and -positive breast cancers. Radiology. (2023) 308:e222646. doi: 10.1148/radiol.222646
37
ZhouJZhangYMiaoHYoonGYWangJLinYet al. Preoperative differentiation of HER2‐Zero and HER2‐Low from HER2‐Positive invasive ductal breast cancers using BI‐RADS MRI features and machine learning modeling. J Magn Reson Imaging. (2025) 61:928–41. doi: 10.1002/jmri.29447
38
EcheTSchwartzLHMokraneF-ZDercleL. Toward generalizability in the deployment of artificial intelligence in radiology: role of computation stress testing to overcome underspecification. Radiology: Artif Intell. (2021) 3:e210097. doi: 10.1148/ryai.2021210097
39
ZhangLCuiQ-XZhouL-QWangX-YZhangH-XZhuY-Met al. MRI-based vector radiomics for predicting breast cancer HER2 status and its changes after neoadjuvant therapy. Comput Med Imaging Graphics. (2024) 118:102443. doi: 10.1016/j.compmedimag.2024.102443
Summary
Keywords
breast cancer, clinicopathological factors, human epidermal growth factor receptor 2, multiparametric MRI, nomogram
Citation
Chen Y, Chen X, Yue B, Zhong X, Zhang H, Chen X, Chen X, Dai Z and Yang Z (2026) Multiparametric MRI-based nomogram integrating clinicopathological factors for predicting HER2 expression status in breast cancer. Front. Oncol. 16:1824808. doi: 10.3389/fonc.2026.1824808
Received
06 March 2026
Revised
04 June 2026
Accepted
08 June 2026
Published
18 June 2026
Volume
16 - 2026
Edited by
Zongwei Zhou, Johns Hopkins University, United States
Reviewed by
Ashish Verma, Banaras Hindu University, India
Minh Toan Ngo, University of Debrecen, Hungary
Updates
Copyright
© 2026 Chen, Chen, Yue, Zhong, Zhang, Chen, Chen, Dai and Yang.
This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
*Correspondence: Zhiqi Yang, y13643090854@163.com; Xiangguang Chen, cxg966504@163.com; Zhuozhi Dai, zzdai@stu.edu.cn
†These authors have contributed equally to this work
Disclaimer
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.