Abstract
Background:
HER2 heterogeneity in breast cancer requires precise stratification for treatment guidance. Differences in dual-energy computed tomography (DECT)-derived parameters among HER2-zero, HER2-low, and HER2-positive breast cancers remain unknown. We aimed to evaluate the performance of DECT-based quantitative parameters to distinguish HER2-zero, HER2-low, and HER2-positive expression status in breast cancer, and to compare the efficacy between DECT-predicted HER2 status and pathological HER2 status in predicting pathologic complete response (pCR).
Methods:
A total of 469 breast cancer patients who underwent chest DECT for staging (November 2020–May 2023) were prospectively recruited. DECT quantitative parameters, such as the normalized iodine concentration (NIC), the normalized effective atomic number (nZeff), the slope of the spectral Hounsfield unit curve (λHu), and polyenergetic images (PEI), were obtained from reconstructed images. Decision tree models were constructed to classify the HER2-zero, HER2-low, and HER2-positive status based on six DECT parameters. These models were tested for performance using the area under the receiver operating characteristic curve (AUC) in training (n = 281) and validation (n = 188) cohorts. Efficacy analysis was performed to compare the efficacy of the DECT-predicted HER2 status with pathological HER2 status in predicting pCR.
Results:
Statistical analysis revealed significant differences in arterial and venous phase NIC, ΔNIC, venous phase λHu, ΔλHu, and venous phase PEI (HU) among HER2 subtypes (all P < 0.001). The decision tree model achieved AUCs of 0.821 (HER2-zero), 0.848 (HER2-low), and 0.899 (HER2-positive) in training, with validation cohort AUCs of 0.825, 0.850, and 0.879, respectively. DECT-predicted HER2 combined with estrogen/progesterone receptor status demonstrated superior pCR prediction (AUC: 0.760) compared to pathological HER2-based models (AUC: 0.689; P = 0.010).
Conclusions:
DECT-based quantitative parameters enable noninvasive stratification of HER2-zero, HER2-low, and HER2-positive breast cancers. Additionally, HER2 predicted by DECT outperformed conventional pathological HER2 status in predicting pCR in patients undergoing neoadjuvant chemotherapy.
Keywords: breast imaging, HER2-low, multiparametric dual-energy computed tomography, neoadjuvant chemotherapy, pathologic complete response, prospective cohort study
Introduction
The American Society of Clinical Oncology (ASCO) guidelines define the classification of human epidermal growth factor receptor 2 (HER2) status into three categories: HER2-positive (immunohistochemistry [IHC] score of 3+ or 2+/HER2 with fluorescence in situ hybridization [FISH]-positive), HER2-low (IHC score of 1+ or IHC score of 2+ with FISH-negative), and HER2- zero (IHC score of 0)[1–3]. While HER2-positive breast cancer (15–20% of cases) has significantly benefited from targeted therapies[4,5], emerging research suggests that patients with HER2-low breast cancer – approximately half of breast cancer patients[6] – may now benefit from novel HER2-targeted antibody-drug conjugates, with recent clinical trials demonstrating significant improvements in both progression-free and overall survival[7,8]. Thus, identifying HER2-status may potentially optimize the treatment strategies for patients with breast cancer.
HIGHLIGHTS
DECT parameters differentiate HER2-zero, HER2-low, and HER2-positive in breast cancer.
Decision tree model with DECT parameters achieves high-accuracy classification.
DECT-predicted HER2 status outperforms pathological HER2 in predicting pCR.
Current clinical assessment of HER2 expression relies on invasive tissue sampling by IHC and FISH analyses, but these methods have several limitations. Biopsy or surgery is invasive; the heterogeneous nature of breast cancer tissue may lead to specimen bias; and insufficient tissue quantity or quality may produce inconclusive results[9]. Therefore, there is an urgent need for a noninvasive and comprehensive method to assess HER2 status. Previous studies have shown that imaging can capture various HER2 statuses, with high predictive performance [the area under the Receiver Operating Characteristic (ROC) curve (AUC) range: 0.76–0.87][9–11]. However, these studies primarily used radiomics or deep learning methods, which have poor interpretability and are limited to HER2 pairwise classification, with no studies to date addressing ternary classification of HER2 status. Thus, there is an urgent need to establish an interpretable image-based HER2 ternary classification model.
Routine imaging examinations for breast cancer include mammography, ultrasonography, and magnetic resonance imaging (MRI)[12]. These three methods mainly focus on the evaluation of breast lesions, and additional imaging examinations are required for lesions in other parts outside the breast. PET/CT is mainly used for staging of breast cancer not suitable for routine screening due to its ionizing radiation and high examination costs[13].
The Chinese Society of Clinical Oncology (CSCO)[14] guidelines recommend computed tomography (CT) for breast cancer to assess potential lung metastases or concurrent pulmonary diseases. The chest scan usually includes the breast area, which provides the possibility for breast cancer patients to “scan more than one.” Dual-energy computed tomography (DECT) uses two different X-ray energy spectra to separate tissues with similar attenuation profiles, thereby achieving material decomposition. It then provides multi-parametric images through post-processing software for spectral imaging, and it holds broad clinical applications in the field of tumor imaging.
DECT is a well-established and extensively used tool in the realm of modern medical diagnostics[15]. In patients with breast cancer, a DECT is performed for staging and to exclude other pulmonary diseases. DECT also covers the breast tissue. In breast cancer, DECT-derived parameters can help distinguish between benign and malignant tumors, predict molecular subtypes, and assess prognostic histopathological factors[16–21]; however, its potential for HER2 status stratification remains uninvestigated. The aim of this study was twofold: (1) to evaluate whether multiparametric DECT can noninvasively differentiate HER2-zero, HER2-low, and HER2-positive breast cancer through ternary classification, and (2) to investigate the predictive performance of HER2 status in relation to pathological complete response (pCR) to neoadjuvant chemotherapy (NAC) in patients with breast cancer.
Materials and methods
Study sample
Approval for this prospective study was granted by the institutional review board, and written informed consent was secured from every participant. In accordance with the protocol, the study enrolled patients with suspected breast cancer who underwent dual-phase contrast-enhanced DECT chest scans to assess potential lung metastasis, pulmonary tuberculosis, or other inflammatory conditions between November 2020 and May 2023 at the outpatient clinic and inpatient department of Breast Surgery in our hospital. A consecutive enrollment method was adopted in this study: eligible patients were included after providing informed consent. During the recruitment period, case enrollment was delayed in 2021 due to the COVID-19 pandemic, so the recruitment period was extended to May 2023. No recruitment incentives were provided to the participants in this study. No adverse events occurred in this study.
The inclusion criteria for assessing HER2 status were: (1) an initial breast cancer diagnosis, (2) no previous breast chemotherapy or radiotherapy, (3) breast cancer with a shortest diameter exceeding 1 cm, visible on the contrast-enhanced DECT images and within the DECT field of view, and (4) no biopsy within the week preceding the DECT scan. Exclusion criteria for differentiating HER2 status were as follows: (1) incomplete pathological data or (2) poor image quality due to severe metal artifacts. The inclusion criteria for predicting pCR were as follows: the patients already enrolled who received NAC therapy. The comprehensive design of the study is depicted in Figure 1. The work has been reported in line with the STROCSS criteria[22].
Figure 1.
Flowchart of participant enrollment through the study. DECT, dual-energy computed tomography; HER2, human epidermal growth factor receptor 2.
DECT image acquisition
Images were acquired using a 2.5 generation DECT scanner (SOMATOM Drive, Siemens Healthineers, Forchheim, Germany) operating in dual-energy mode with two X-ray tubes set at distinct voltages: tube A at 100 kV and tube B at Sn 140 kV, a tin filter was used for the high-voltage tube, and automatic exposure control (CARE Dose 4D, Siemens Healthineers) was implemented for all scans. The scanning parameters were: collimation, 64 mm × 0.6 mm, rotation time of 0.28 seconds, pitch of 0.55, and reference tube current time product of 71 mAs for the 100-kVp tube and 60 mAs for the Sn140-kVp tube. The reconstruction of images involved a section thickness and increment both set at 1.5 mm.
Patients were positioned supine with arms elevated by the head and scanned in the craniocaudal direction. The scan covered the entire chest from the superior thoracic aperture to the inferior edge of the costophrenic angle, ensuring comprehensive imaging of the breast and axillary regions. For contrast-enhanced imaging, an iodinated nonionic contrast agent (Ioversol, 320 mg/ml iodine, HENGRUI Medicine, Jiangsu, China) was administered via the right or left ulnar vein using a dual-head injector. A dose of 1.5 ml/kg was administered at a flow rate of 2.5 ml/s, followed by a 30 ml saline flush at the same flow rate. The opposite ulnar vein to the suspected breast lesion was used for injection to minimize beam-hardening artifacts. Arterial phase imaging was initiated using a bolus-tracking technique, with a trigger threshold of 100 HU in the descending aorta, followed by a 10-second delay. Venous phase imaging commenced 25 seconds after completion of the arterial phase.
DECT quantitative parameter analysis
Multiparametric analysis was performed using syngo.via software (Dual Energy, syngo.via VB20A, Siemens Healthineers, Forchheim, Germany) on a dedicated workstation. Polyenergetic images were produced by combining 50% of the low-kV data with 50% of the high-kV data, using a blending factor of 0.5. The DECT quantitative parameters were assessed by two radiologists who have 8 years of experience in breast and chest imaging and another radiologist with 5 years of experience in post-reconstruction imaging, both of whom were unaware of the immunohistochemical results of invasive breast cancer. Circular regions of interest (ROI) with an area of 1 cm2 were placed on axial slices at the level of each breast lesion, avoiding areas of necrosis, calcification, or large vessels. For each lesion, ROIs were kept consistent on both arterial- and venous-phase DECT images with the copy-and-paste function in the workstation. The intraclass correlation coefficient (ICC) is calculated to evaluate the consistency between different radiologists. The inter-observer ICCs of all radiologists are >0.85, ensuring the reliability of parameter extraction.
The study involved 12 DECT quantitative parameters derived from both the venous and arterial phase scans. Both the normalized iodine concentration (NIC) and normalized effective atomic number (nZeff) were calculated by normalizing these values to the corresponding measurements in the aorta. The following calculation was used to determine the slope of the spectral Hounsfield unit curve (λHu in Hounsfield unit per kiloelectron-volt)[18]:
These parameters included the difference between the arterial and venous phases for iodine concentration (ΔNIC), effective atomic number (ΔnZeff), and λHu (ΔλHu). Three measurements of HU and standard deviation were taken from polyenergetic images (PEIs), and average values were recorded.
Criteria for HER2 expression in pathological sections
All patients underwent surgical resection or core needle biopsy of the primary tumor. Pathological evaluations were conducted in accordance with the 2018 guidelines issued by ASCO and the College of American Pathologists[1], with all assessments performed by a pathologist with over 15 years of experience. Tumors with equivocal scores of 2+ were further tested using FISH. HER2-negative cases were classified as HER2-low if they had an IHC score of 1+ or 2+ with a negative FISH result, and as HER2-zero if the IHC score was 0.
DECT-based model construction and evaluation
In this study, 40% of patients were randomly selected as the independent validation cohort, resulting in a ratio of 6:4 for training and validation patients. Stratified random sampling was used to ensure that the proportion of the HER2-zero, HER2-low, and HER2-positive patients in the training and validation cohorts remained the same.
In the training cohort, intergroup differences in DECT quantitative parameters were analyzed to compare the variations among the HER2-zero, HER2-low, and HER2-positive groups. Variables showing significant differences (P < 0.05) across the three groups were selected and used to develop the ternary classification model.
Prediction of pCR to NAC
To investigate the predictive performance of HER2 status on the pCR to NAC in patients with breast cancer, we developed two models to predict pCR. At the final surgical histologic examination, pCR was characterized by an experienced pathologist as the lack of invasive residual cancer cells in the breast and axillary nodes (ypT0/is N0). The detection of invasive residual cancer cells in the breast or axillary nodes indicated non-pCR. Preoperative status of estrogen receptor (ER), progesterone receptor (PR), and HER2 differed significantly between the groups with and without pCR (P < 0.05). These variables were thus incorporated into the model in this investigation. Two models were developed to predict the pCR: (1) ER, PR, and HER2 status predicted by DECT; (2) ER, PR, and pathological HER2 status.
Statistical analysis
Continuous variables were assessed using the Wilcoxon rank sum test, Mann–Whitney U test, and Student’s t-test, depending on the normality of distribution. Categorical variables were analyzed using the χ2 test among HER2-zero, HER2-low, and HER2-positive groups. Given the null hypothesis (AUC = 0.50) and alternative hypothesis (AUC = 0.70), along with the proportions of HER2 subtypes in this study (25% for HER2-zero, 55% for HER2-low, and 20% for HER2-positive) and a statistical power of 90%, we calculated that the minimum required sample size was 170 patients. Categorical variables were expressed as numbers and compared using either the chi-square test or Fisher’s exact test, as required. Diagnostic performance was assessed by determining AUC and compared using DeLong’s test. The significance threshold was set at P < 0.05. ROC curve analysis, calibration curve analysis, and DCA were performed using Python (version 3.9.5).
Results
Participant characteristics
Patients’ baseline characteristics of in the two cohorts are summarized in Table 1. A total of 469 patients (281 patients in the training cohort, and 188 patients in the validation cohort) were enrolled for analysis. In the training cohort, 89 of 281 patients (32%) had HER2-positive tumors, 118 of 281 patients (42%) had HER2-low tumors, and 74 of 281 patients (26%) had HER2-zero cancers. Significant differences in ER status, PR status, and molecular subtype (P < 0.05) were observed in the training and validation cohorts. Ki-67 status, clinical T stage and axillary lymph node (ALN) were statistically different only in the training cohort (P < 0.05). No differences were found in the age in the training and validation cohorts (P > 0.05).
Table 1.
Clinicopathologic characteristics of HER2-zero, HER2-low and HER2-positive participants with breast cancer in the training and validation cohorts.
| Training cohort(n = 281) | Validation cohort(n = 188) | |||||||
|---|---|---|---|---|---|---|---|---|
| Characteristics | HER2-zero (n = 74) | HER2-low (n = 118) | HER2-positive (n = 89) | P | HER2-zero (n = 50) | HER2-low (n = 79) | HER2-positive (n = 59) | P |
| Age Mean ± SD (years) | 52.38 ± 9.90 | 53.91 ± 11.00 | 51.79 ± 11.08 | 0.951a | 52.70 ± 10.53 | 52.56 ± 10.34 | 53.61 ± 8.73 | 0.264a |
| ER Status | < 0.001b | 0.007b | ||||||
| Positive | 53 (71.6) | 100 (84.7) | 44 (49.4) | 33 (66.0) | 55 (69.6) | 26 (44.1) | ||
| Negative | 21 (28.4) | 18 (15.3) | 45 (50.6) | 17 (34.0) | 24 (30.4) | 33 (55.9) | ||
| PR Status | < 0.001b | < 0.001b | ||||||
| Positive | 29 (39.2) | 67 (56.8) | 23 (25.8) | 14 (28.0) | 36 (45.6) | 7 (11.9) | ||
| Negative | 45 (60.8) | 51 (43.2) | 66 (74.2) | 36 (72.0) | 43 (54.4) | 52 (88.1) | ||
| Ki 67 Status | 0.007b | 0.476b | ||||||
| Positive | 59 (79.7) | 82 (69.5) | 78 (87.6) | 35 (70.0) | 61 (77.2) | 47 (79.7) | ||
| Negative | 15 (20.3) | 36 (30.5) | 11 (12.4) | 15 (30.0) | 18 (22.8) | 12 (20.3) | ||
| Clinical T stage | 0.007b | 0.280b | ||||||
| T1 | 31 (41.9) | 24 (20.3) | 17 (19.1) | 15 (30.0) | 23 (29.1) | 12 (20.3) | ||
| T2 | 29 (39.2) | 60 (50.8) | 55 (61.8) | 23 (46.0) | 33 (41.8) | 27 (45.8) | ||
| T3 | 4 (5.4) | 8 (6.8) | 5 (5.6) | 0 (0.0) | 5 (6.3) | 7 (11.9) | ||
| T4 | 10 (13.5) | 26 (22.0) | 12 (13.5) | 12 (24.0) | 18 (22.8) | 13 (22.0) | ||
| ALN | 0.002b | 0.114b | ||||||
| Positive | 31 (41.9) | 79 (66.9) | 56 (62.9) | 21 (42.0) | 45 (57.0) | 36 (61.0) | ||
| Negative | 43 (58.1) | 39 (33.1) | 33 (37.1) | 29 (58.0) | 34 (43.0) | 23 (39.0) | ||
| Molecular subtype | < 0.001b | <0.001b | ||||||
| Luminal A | 26 (35.1) | 34 (28.8) | 2 (2.2) | 18 (36.0) | 22 (27.8) | 1 (1.7) | ||
| Luminal B | 27 (36.5) | 64 (54.2) | 40 (44.9) | 13 (26.0) | 34 (43.0) | 23 (39.0) | ||
| Triple-negative | 21 (28.4) | 15 (12.7) | 1 (1.1) | 19 (38.0) | 20 (25.3) | 2 (3.4) | ||
| HER2-enriched | 0 (0.0) | 5 (4.2) | 46 (51.7) | 0 (0.0) | 3 (3.8) | 33 (55.9) | ||
SD, standard deviation; ER, estrogen receptor; PR, progesterone receptor; ALN, axillary lymph nodes; HER2, human epidermal growth factor receptor 2.
Mann–Whitney U test.
χ2 test.
Table 2 summarizes the clinical characteristics of patients with breast cancer undergoing NAC. A total of 144 patients received NAC, of whom 40% (57/144) patients achieved pCR. In these two groups, no significant differences were detected between the pCR and non-pCR groups in terms of age, Ki-67 status, clinical T stage, and ALN (P > 0.05). Meanwhile, ER, PR, HER2 status, and molecular subtype were significantly different between groups (P < 0.05).
Table 2.
Clinicopathologic characteristics of participants with breast cancer who received neoadjuvant chemotherapy.
| Characteristics | pCR (n = 57) | non-pCR (n = 87) | P |
|---|---|---|---|
| Age Mean ± SD (years) | 49.16 ± 10.13 | 52.62 ± 8.81 | 0.369a |
| ER Status | < 0.001b | ||
| Positive | 21 (36.8) | 65 (74.7) | |
| Negative | 36 (63.2) | 22 (25.3) | |
| PR Status | < 0.001b | ||
| Positive | 8 (14.0) | 44 (50.6) | |
| Negative | 49 (86.0) | 43 (49.4) | |
| HER 2 Status | 0.001b | ||
| Positive | 34 (59.6) | 25 (28.7) | |
| Low | 14 (24.6) | 43 (49.4) | |
| Zero | 9 (15.8) | 19 (21.8) | |
| Ki 67 Status | 0.344b | ||
| Positive | 45 (78.9) | 74 (85.1) | |
| Negative | 12 (21.1) | 13 (14.9) | |
| Clinical T stage | 0.151b | ||
| T1 | 4 (7.0) | 9 (10.3) | |
| T2 | 34 (59.6) | 37 (42.5) | |
| T3 | 8 (14.0) | 11 (12.6) | |
| T4 | 11 (19.3) | 30 (34.5) | |
| ALN | 0.278b | ||
| Positive | 46 (80.7) | 76 (87.4) | |
| Negative | 11 (19.3) | 11 (12.6) | |
| Molecular subtype | 0.001b | ||
| Luminal A | 2 (3.5) | 18 (20.7) | |
| Luminal B | 18 (31.6) | 46 (52.9) | |
| Triple-negative | 13 (22.8) | 9 (10.3) | |
| HER2-enriched | 24 (42.1) | 14 (16.1) | |
| NAC regimen | < 0.001b | ||
| Anthracycline-based | 3 (5.3) | 10 (11.5) | |
| Taxane-based | 17 (29.8) | 29 (33.3) | |
| Anthracycline- and taxane-based | 37 (64.9) | 48 (55.2) |
SD, standard deviation; ER, estrogen receptor; PR, progesterone receptor; HER2, human epidermal growth factor receptor 2; ALN, axillarylymph nodes; pCR, pathological complete response; non-pCR, non-pathological complete response.
Mann–Whitney U test.
χ2 test.
Comparison of DECT quantitative parameters among HER2‑zero, HER2‑low, and HER2‑positive groups
The ICCs of DECT quantitative parameters were 0.946 (range, 0.811–0.967), which indicated good-to-excellent agreement.
Comparisons of quantitative parameters among the HER2-zero, HER2-low, and HER2- positive groups are shown in Table 3, Figure 2, and Supplemental Digital Content Figure S1, available at: http://links.lww.com/JS9/H142. Differences in arterial and phase NIC, ΔNIC, venous phase λHu, ΔλHu, and venous phase PEI (HU) were observed among the HER2-zero, HER2-low, and HER2-positive breast cancer groups. In the training cohort, the arterial phase NIC (0.10 ± 0.06 vs 0.08 ± 0.04 vs 0.12 ± 0.05; adjusted P < 0.001), venous phase NIC (0.33 ± 0.14 vs 0.26 ± 0.07 vs 0.39 ± 0.11; adjusted P < 0.001), ΔNIC (0.23 ± 0.13 vs 0.18 ± 0.10 vs 0.26 ± 0.11; adjusted P < 0.001), venous phase λHu (2.29 ± 1.08 vs 2.58 ± 1.36 vs 3.21 ± 1.02; adjusted P < 0.001), ΔλHu (0.93 ± 0.86 vs 1.12 ± 0.85 vs 1.70 ± 0.93; adjusted P < 0.001), and venous phase PEI (HU) (67.16 ± 16.23 vs 72.32 ± 16.58 vs 83.82 ± 15.90; adjusted P < 0.001) were significantly different among the HER2-zero, HER2-low, and HER2-positive groups. No significant difference was found in arterial phase nZeff, venous phase nZeff, ΔnZeff, arterial phase λHu, pre-PEI(HU), and arterial phase PEI (HU) among the groups (P > 0.05). The arterial phase NIC, venous phase NIC, and ΔNIC were significantly lower in the HER2-low breast cancer group compared with both the HER2-zero and HER2-positive breast cancer groups (adjusted P < 0.001 to P = 0.033). The venous phase λHu, ΔλHu, and venous phase PEI (HU) were significantly lower in the HER2-low breast cancer group compared with those in the HER2-positive breast cancer group (adjusted P < 0.001).
Table 3.
DECT quantitative parameters among HER2-zero, HER2-low, and HER2-positive in patients with breast cancer in training cohort and validation cohort.
| Training cohort (n = 281) | Validation cohort (n = 188) | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Quantitative parameters | HER2-zero (n = 74) | HER2-low (n = 118) | HER2-positive (n = 89) | P | P[1] | [P2] | [P3] | HER2-zero (n = 50) | HER2-low (n = 79) | HER2-positive (n = 59) | P | P[1] | [P2] | [P3] |
| Arterial phase NIC | 0.10 ± 0.06 | 0.08 ± 0.04 | 0.12 ± 0.05 | < 0.001 | 0.033 | 0.137 | < 0.001 | 0.09 ± 0.07 | 0.08 ± 0.04 | 0.11 ± 0.06 | 0.026 | 0.543 | 0.318 | 0.019 |
| Venous phase NIC | 0.33 ± 0.14 | 0.26 ± 0.07 | 0.39 ± 0.11 | < 0.001 | < 0.001 | 0.003 | < 0.001 | 0.30 ± 0.15 | 0.27 ± 0.07 | 0.38 ± 0.13 | < 0.001 | 0.271 | 0.003 | < 0.001 |
| ΔNIC | 0.23 ± 0.13 | 0.18 ± 0.10 | 0.26 ± 0.11 | < 0.001 | 0.003 | 0.251 | < 0.001 | 0.20 ± 0.11 | 0.20 ± 0.10 | 0.24 ± 0.14 | 0.114 | NA | NA | NA |
| Arterial phase nZeff | 0.72 ± 0.05 | 0.72 ± 0.09 | 0.72 ± 0.04 | 0.917 | NA | NA | NA | 0.70 ± 0.04 | 0.71 ± 0.04 | 0.71 ± 0.04 | 0.371 | NA | NA | NA |
| Venous phase nZeff | 0.85 ± 0.05 | 0.85 ± 0.08 | 0.86 ± 0.05 | 0.909 | NA | NA | NA | 0.84 ± 0.04 | 0.83 ± 0.13 | 0.85 ± 0.06 | 0.653 | NA | NA | NA |
| ΔnZeff | 0.16 ± 0.18 | 0.22 ± 0.19 | 0.29 ± 0.78 | 0.205 | NA | NA | NA | 0.18 ± 0.12 | 0.20 ± 0.26 | 0.23 ± 0.23 | 0.381 | NA | NA | NA |
| Arterial phase λHu (HU/keV) | 1.37 ± 0.85 | 1.36 ± 0.80 | 1.42 ± 0.65 | 0.863 | NA | NA | NA | 1.25 ± 0.64 | 1.26 ± 0.65 | 1.16 ± 0.58 | 0.628 | NA | NA | NA |
| Venous phase λHu (HU/keV) | 2.29 ± 1.08 | 2.58 ± 1.36 | 3.21 ± 1.02 | < 0.001 | 0.241 | < 0.001 | < 0.001 | 2.22 ± 0.78 | 2.39 ± 1.09 | 2.89 ± 0.80 | < 0.001 | 0.563 | < 0.001 | 0.006 |
| ΔλHu (HU/keV) | 0.93 ± 0.86 | 1.12 ± 0.85 | 1.70 ± 0.93 | < 0.001 | 0.315 | < 0.001 | < 0.001 | 0.91 ± 0.80 | 1.13 ± 0.88 | 1.67 ± 0.80 | < 0.001 | 0.312 | < 0.001 | < 0.001 |
| Pre-PEI(HU) | 41.84 ± 9.70 | 41.69 ± 9.13 | 40.21 ± 9.08 | 0.432 | NA | NA | NA | 40.04 ± 11.33 | 41.57 ± 9.36 | 43.19 ± 10.39 | 0.279 | NA | NA | NA |
| Arterial phase PEI(HU) | 56.65 ± 14.37 | 56.79 ± 13.26 | 58.56 ± 14.66 | 0.599 | NA | NA | NA | 53.12 ± 13.15 | 57.47 ± 12.30 | 56.86 ± 14.36 | 0.169 | NA | NA | NA |
| Venous phase PEI(HU) | 67.16 ± 16.23 | 72.32 ± 16.58 | 83.82 ± 15.90 | < 0.001 | 0.084 | < 0.001 | < 0.001 | 67.40 ± 18.04 | 73.54 ± 14.52 | 80.91 ± 15.36 | < 0.001 | 0.083 | < 0.001 | 0.020 |
DECT, dual-energy CT; HER2, human epidermal growth factor receptor 2; NIC, normalized iodine concentration; ΔNIC, the difference between the arterial phase NIC and the venous phase NIC; nZeff, normalized effective atomic number; ΔnZeff, the difference between the arterial phase nZeff and the venous phase nZeff; λHu, slope of the spectral Hounsfield unit curve; ΔλHu, the difference between the arterial phase λHu and the venous phase λHu; PEI, polyenergetic images; P1 HER2 zero vs HER2 low; P2 HER2 zero vs HER2 positive; P3 HER2 low vs HER2 positive.
Figure 2.
Violin plots presenting the distributions of DECT quantitative parameters among patients with HER2-zero, HER2-low, and HER2-positive breast cancer in the training cohort (A) and validation cohort (B). HER2, human epidermal growth factor receptor 2; NIC, normalized iodine concentration; ΔNIC, the difference between the arterial phase NIC and the venous phase NIC; λHu, slope of the spectral Hounsfield unit curve; ΔλHu, the difference between the arterial phase λHu and the venous phase λHu; PEI, polyenergetic image.
Diagnostic performances of DECT quantitative parameters in differentiating HER2‑zero, HER2‑low, and HER2‑positive groups
To build the predictive model, six key DECT parameters were incorporated: arterial phase NIC, venous phase NIC, ΔNIC, venous phase λHu, ΔλHu, and venous phase PEI. ROC analyses of DECT quantitative parameters to predict HER2 status are shown in Table 4 and Figure 3A and B. Predictions of HER2 categories by the classifier are presented in the confusion matrix (Fig. 3C and D). In the training cohort, the ternary classification model achieved an AUC of 0.821 for distinguishing HER2-zero, with a sensitivity of 39% and specificity of 99%. The model for distinguishing HER2-low achieved an AUC of 0.848, with a sensitivity of 71% and specificity of 82%. The model for distinguishing HER2-positive achieved an AUC of 0.899, with a sensitivity of 92% and specificity of 72%.
Table 4.
Diagnostic performance of DECT quantitative parameters model in discriminating HER2-zero, HER2-low, and HER2-positive status in breast cancers.
| Model | AUC [95% CI] | Sensitivity | Specificity | Accuracy |
|---|---|---|---|---|
| Training Cohort | 69% (195/281) | |||
| HER2-zero | 0.821 [0.781-0.872] | 39% (29/74) | 99% (205/207) | |
| HER2-low | 0.848 [0.799-0.883] | 71% (84/118) | 82% (133/163) | |
| HER2-positive | 0.899 [0.845-0.925] | 92% (82/89) | 72% (138/192) | |
| Validation Cohort | 67% (126/188) | |||
| HER2-zero | 0.825 [0.768-0.872] | 50% (25/50) | 91% (125/138) | |
| HER2-low | 0.850 [0.818-0.897] | 85% (67/79) | 63% (69 /109) | |
| HER2-positive | 0.879 [0.813-0.912] | 58% (34/59) | 93% (120/129) |
DECT, dual-energy CT; HER2, human epidermal growth factor receptor 2.
Figure 3.
Diagnostic performance of models in the training cohort and validation cohort. ROC curves and confusion matrices for ternary classification of HER2 status in the training (A and C) cohort and validation cohort (B and D). The orange line represents the HER2-zero. The red line represents the HER2-low. The purple line represents the HER2-positive. HER2, human epidermal growth factor receptor 2; ROC, receiver operating characteristic; 95% CI, 95% confidence interval.
In the validation cohort, the model for distinguishing HER2-zero achieved an AUC of 0.825, with a sensitivity of 50% and specificity of 91%. The model for distinguishing HER2-low achieved an AUC of 0.850, with a sensitivity of 85% and specificity of 63%. The model for distinguishing HER2-positive achieved an AUC of 0.879, with a sensitivity of 58% and specificity of 93%. The overall accuracy of differentiating the three groups was 69% in the training cohort and 67% in the validation cohort. Representative patient images for HER2-zero, HER2-low-and HER2-positive breast cancer that were correctly predicted by DECT quantitative parameters are shown in Figure 4.
Figure 4.
Representative images for HER2-zero, HER2-low, and HER2-positive breast cancer that were correctly predicted by DECT quantitative parameters. Patient 1 is a 55-year-old woman with right HER2-zero breast cancer. Patient 2 is a 51-year-old woman with left HER2-low breast cancer. Patient 3 is a 47-year-old woman with right HER2-positive breast cancer. The first column is venous phase PEI. The second column is arterial phase NIC. The third column is venous phase NIC. The fourth column is venous phase λHu. PEI, polyenergetic image; NIC, normalized iodine concentration; λHu, slope of the spectral Hounsfield unit curve.
To further ensure the generalizability of the model, we conducted a subgroup analysis of the HER2 ternary classification prediction model based on T staging. The results of the subgroup analysis showed that the model had stable predictive performance across all T-stage subgroups (Supplemental Digital Content Figure S2 and S3, available at: http://links.lww.com/JS9/H142).
Performance of models in predicting pCR to NAC
For predicting pCR in response to NAC, the model that combined ER, PR and pathological HER2 status achieved an AUC of 0.689 (95% CI: 0.608, 0.766), with a sensitivity of 63%, accuracy of 70%, and specificity of 75%. The model that combined ER, PR, and HER2 status predicted by DECT achieved an AUC of 0.760 (95% CI: 0.675, 0.837), with a sensitivity of 79%, accuracy of 71%, and specificity of 66%, which was superior to the pathological HER2 status with ER and PR model (P = 0.010, DeLong test) (Table 5, Fig. 5).
Table 5.
Diagnostic performance of predictive HER2 status and true HER2 status combined with ER, PR status in predicting pCR in breast cancer.
| Model | Predictive HER2 combined with HR model | True HER2 combined with HR model |
|---|---|---|
| AUC [95% CI] | 0.760 [0.675–0.837] | 0.689 [0.608–0.766] |
| Sensitivity | 79% (45/57) | 63% (36/57) |
| Specificity | 66% (57/87) | 75% (65/87) |
| Accuracy | 71% (102/144) | 70% (101/144) |
| PPV | 60% (45/75) | 62% (36/58) |
| NPV | 83% (57/69) | 76% (65/86) |
ER, estrogen receptor; PR, progesterone receptor; pCR, pathologic complete response; HER2, human epidermal growth factor receptor 2. Predictive HER2 combined with HR model vs True HER2 combined with HR model DeLong test P = 0.010.
Figure 5.
Diagnostic performance of models combining predictive HER2 status or true HER2 status with ER, PR status in predicting pCR in patients with breast cancer who received neoadjuvant chemotherapy. (A) ROC curves; (B) decision curves. ROC: receiver operating characteristic; HER2: human epidermal growth factor receptor 2; ER: estrogen receptor; PR: progesterone receptor; pCR: pathologic complete response; 95% CI: 95% confidence interval.
Discussion
The classification of the three subtypes of HER2 status in breast cancer is attracting increasing attention[23,24]. On the basis of multiparametric DECT, arterial phase NIC, venous phase NIC, ΔNIC, venous phase λHu, ΔλHu, and venous phase PEI (HU) could differentiate among HER2-zero, HER2-low, and HER2-positive status in breast cancer. These six parameters were included in the HER2 prediction model, which showed good performance for predicting HER2-zero, HER2-low, and HER2-positive status, with AUCs of 0.825, 0.850, and 0.879, respectively. Through efficacy assessments, the HER2 predicted by DECT was significantly better than the pathological HER2 status (0.760 vs 0.689) in predicting pCR, further supporting the model’s clinical value.
Results indicate that the arterial phase NIC, venous phase NIC, and ΔNIC were significantly lower in patients with HER2-low breast cancer compared with those in patients with HER2-zero and HER2-positive breast cancer. Furthermore, these parameters were lower in patients with HER2-zero breast cancer than in those with HER2-positive breast cancer. These findings support previous research[19,21], and the hypothesis that HER2-positive tumors exhibit enhanced angiogenic potential, as NIC parameters reflect intralesional contrast agent distribution patterns associated with neovascularization. Notably, this study included λHu, demonstrating significantly lower venous phase λHu and ΔλHu values in patients with HER2-low breast cancer compared with those with HER2-positive tumors. This observation complements existing evidence[19,21,25–27], suggesting that elevations in λHu correlate with both malignant transformation (versus benign lesions) and specific molecular subtypes (HER2-enriched versus luminal tumors). The collective DECT parameter variations across HER2 subgroups may reflect underlying histopathological differentiation gradients, with HER2-positive tumors demonstrating more aggressive biological features.
Previous studies have demonstrated that radiomics can distinguish between patients with HER2-low-positive and HER2-zero breast cancer with an AUC of 0.855, between patients with HER2-negative and HER2-overexpressing breast cancer with an AUC of 0.868[11], and between patients with HER2-low and HER2-overexpressing breast cancer with an AUC of 0.877[9]. In this study, the performance of the DECT model for predicting HER2 status is comparable to that of previous studies. DECT model directly performed ternary classification rather than pairwise classification, which is more conducive to clinical decision-making. Unlike black-box radiomics features, DECT quantitative parameters provide physiologically interpretable biomarkers, effectively predicting the expression of immunohistochemical biomarkers in breast cancer and reflecting the tumor’s vascular supply.
Identifying HER2-low status can help stratify patients for more personalized therapeutic approaches, optimize clinical decision-making and potentially improve prognosis. In addition to constructing a model to predict HER2 status, the results further predicted the pCR among patients who received NAC. Results showed that DECT-predicted HER2 status exhibited significantly higher predictive performance in predicting pCR compared to pathological HER2 status. This may be because DECT quantitative parameters more accurately reflect the vascular distribution and tissue composition within breast cancer[17,28]. In contrast, pathological HER2 status assessment is reliant on the pathologist’s experience[29], which, along with tumor heterogeneity[30], can introduce potential biases in the pathological results.
This study has several limitations. First, as a prospective single-center study with all DECT data from one equipment brand, and with only 144 of 469 patients undergoing NAC (resulting in a small pCR subgroup sample size), the lack of multi-center, multi-brand CT external validation and limited pCR subgroup size may restrict result generalizability across clinical settings. Second, the model relied solely on DECT single-modal data, whereas multi-modal integration could optimize diagnostic efficacy and improve robustness and accuracy. Third, the model showed low sensitivity for HER2-zero detection – despite >90% specificity, false negatives might miss actual HER2-zero patients and hinder precise treatment selection. Fourth, only short-term pCR was assessed, excluding long-term survival indicators, limiting evaluation of the model’s value for guiding long-term prognosis. Fifth, the biological interpretation of DECT parameters requires further verification.
To address these limitations, future research will expand the total sample size and pCR subgroup size at our hospital to enhance model stability and subgroup analysis efficacy. External validation centers will adopt a “unified scanning protocol framework + device-specific parameter fine-tuning” model with unified image quality control standards, and parameter standardization for cross-device imaging differences will be achieved through batch effect correction. After verifying multi-center generalizability, the model will be integrated into clinical workflows to assist invasive examinations. Additionally, data from different equipment types will be integrated to test parameter adaptability via batch effect correction (grouping centers, scanning parameters, reconstruction methods, and manufacturers for correction and assessing model generalization), preprocessing method evaluation (adopting filtering, resampling, standardization, and normalization techniques to evaluate data harmonization impact), and deep learning applications (e.g., converting images across devices, parameters, or centers to achieve harmonization and examining model improvements).
In conclusion, the results of the single-center study showed that DECT-based quantitative parameters accurately predicted HER2 expression status through ternary classification in breast cancer patients and may serve as a noninvasive approach to identify those suitable for HER2-targeted treatments. Furthermore, the HER2 status predicted by DECT was more useful in forecasting pCR after NAC, underscoring the clinical value of HER2 classification in guiding treatment decisions for breast cancer patients – particularly in the emerging era of HER2-low targeted therapies.
Acknowledgements
Not applicable.
Footnotes
Sponsorships or competing interests that may be relevant to content are disclosed at the end of this article.
Supplemental Digital Content is available for this article. Direct URL citations are provided in the HTML and PDF versions of this article on the journal’s website, www.lww.com/international-journal-of-surgery.
Contributor Information
Huifang Chen, Email: 639377300@qq.com.
Yao Huang, Email: huangyao@stu.cqu.edu.cn.
Xiangfei Zeng, Email: weifang1214@foxmail.com.
Ying Cao, Email: 2280271772@qq.com.
Xueqin Gong, Email: gongxueqin2019@163.com.
Sun Tang, Email: tstangsun@163.com.
Lu Wang, Email: 479783724@qq.com.
Yue Cheng, Email: chengyy96@163.com.
Xiaoxia Wang, Email: wangxx0531@foxmail.com.
Jiuquan Zhang, Email: zhangjq_radiol@foxmail.com.
Ethical approval
This study was approved by Chongqing University Cancer Hospital institutional review board. The reference number was CZLS20200215-A.
Consent
This prospective study was approved by the institutional review board, and written informed consent was collected from every participant.
Sources of funding
This study has received funding by the Natural Science Foundation of Chongqing municipality (CSTB2023NSCQ-MSX0787, CSTB2022NSCQ-MSX1158, CSTB2023NSCQ-MSX0858, CSTB2024NSCQ-MSX0217, CSTB2024NSCQ-MSX0760, and CSTB2024NSCQ-MSX0334), the Fundamental Research Funds for the Central Universities (2023CDJYGRH-YB04), the Chongqing University Cancer Hospital Scientific Research Capacity Improvement Project (2023nlts004), Medical Research Project of Chongqing Municipal Health Commission(2024WSJK072), Chongqing Medical Research Project of Combination of Science and Medicine (No. 2024MSXM171).
Author contributions
H.C., X.W., and J.Z.: conceptualization; Y.C., X.G., S.T., L.W., L.L., T.Y., and Y.C.: data curation; H.C., X.W., and J.Z.: funding acquisition; S.T. and L.W.: formal analysis; H.C. and X.W.: investigation; H.C. and Y.H.: methodology; X.Z.: resources; J.Z.: supervision; H.C.: writing − original draft; H.C., X.W., and J.Z.: writing – review & editing.
Conflicts of interest disclosure
The authors of this manuscript declare no relationships with any companies, whose products or services may be related to the subject matter of the article.
Research registration unique identifying number (UIN)
Not applicable.
Guarantor
Jiuquan Zhang.
Provenance and peer review
Not commissioned; externally peer-reviewed STROCSS GUIDELINE CHECKLIST.
Data availability statement
The datasets generated during and analyzed during the current study are available by the corresponding author J.Z., upon reasonable request.
Summary Statement
DECT parameters noninvasively stratifies HER2-zero, -low, and -positive subtypes and DECT-predicted HER2 status predicts chemotherapy response better than pathological HER2 in breast cancer.
References
- [1].Wolff AC, Hammond MEH, Allison KH, et al. Human epidermal growth factor receptor 2 testing in breast cancer: American Society Of Clinical Oncology/College of American Pathologists Clinical Practice guideline focused update. J Clin Oncol 2018;36:2105–22. [DOI] [PubMed] [Google Scholar]
- [2].Modi S, Park H, Murthy RK, et al. Antitumor activity and safety of trastuzumab deruxtecan in patients with HER2-low-expressing advanced breast cancer: results from a phase ib study. J Clin Oncol 2020;38:1887–96. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [3].Ivanova M, Porta FM, D’Ercole M, et al. Standardized pathology report for HER2 testing in compliance with 2023 ASCO/CAP updates and 2023 ESMO consensus statements on HER2-low breast cancer. Virchows Arch 2024;484:3–14. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [4].Keller M, Rohlf K, Glotzbach A, et al. Inhibiting the glycerophosphodiesterase EDI3 in ER-HER2+ breast cancer cells resistant to HER2-targeted therapy reduces viability and tumour growth. J Exp Clin Cancer Res 2023;42:25. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [5].Montemurro F, Delaloge S, Barrios CH, et al. Trastuzumab emtansine (T-DM1) in patients with HER2-positive metastatic breast cancer and brain metastases: exploratory final analysis of cohort 1 from KAMILLA, a single-arm phase IIIb clinical trial(☆). Ann Oncol 2020;31:1350–58. [DOI] [PubMed] [Google Scholar]
- [6].Tarantino P, Hamilton E, Tolaney SM, et al. HER2-low breast cancer: pathological and clinical landscape. J Clin Oncol 2020;38:1951–62. [DOI] [PubMed] [Google Scholar]
- [7].Denkert C, Seither F, Schneeweiss A, et 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] [PubMed] [Google Scholar]
- [8].Modi S, Jacot W, Yamashita T, et al. Trastuzumab deruxtecan in previously treated HER2-low advanced breast cancer. N Engl J Med 2022;387:9–20. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [9].Mao C, Hu L, Jiang W, et 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 2024;34:2546–59. [DOI] [PubMed] [Google Scholar]
- [10].Ramtohul T, Djerroudi L, Lissavalid E, et al. Multiparametric MRI and radiomics for the prediction of HER2-zero, -low, and -positive breast cancers. Radiology 2023;308:e222646. [DOI] [PubMed] [Google Scholar]
- [11].Guo Y, Xie X, Tang W, et al. Noninvasive identification of HER2-low-positive status by MRI-based deep learning radiomics predicts the disease-free survival of patients with breast cancer. Eur Radiol 2024;34:899–913. [DOI] [PubMed] [Google Scholar]
- [12].Suvannarerg V, Chitchumnong P, Apiwat W, et al. Diagnostic performance of qualitative and quantitative shear wave elastography in differentiating malignant from benign breast masses, and association with the histological prognostic factors. Quant Imaging Med Surg 2019;9:386–98. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [13].Zhang X, Liu Y, Luo H, et al. PET/CT and MRI for identifying axillary lymph node metastases in breast cancer patients: systematic review and meta-analysis. J Magn Reson Imaging 2020;52:1840–51. [DOI] [PubMed] [Google Scholar]
- [14].Li J, Hao C, Wang K, et al. Chinese Society of Clinical Oncology (CSCO) breast cancer guidelines 2024. Transl Breast Cancer Res 2024;5:18. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [15].Kaatsch HL, Becker BV, Schule S, et al. Gene expression changes and DNA damage after ex vivo exposure of peripheral blood cells to various CT photon spectra. Sci Rep 2021;11:12060. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [16].Lohofer FK, Kaissis GA, Koster FL, et al. Improved detection rates and treatment planning of head and neck cancer using dual-layer spectral CT. Eur Radiol 2018;28:4925–31. [DOI] [PubMed] [Google Scholar]
- [17].Metin NO, Balci S, Metin Y, et al. Correlation between quantitative parameters obtained by dual energy spectral CT and prognostic histopathological factors and biomarkers in breast cancer. Clin Breast Cancer 2024;24:e279–e288. [DOI] [PubMed] [Google Scholar]
- [18].Zhang X, Zheng C, Yang Z, et al. Axillary sentinel lymph nodes in breast cancer: quantitative evaluation at dual-energy CT. Radiology 2018;289:337–46. [DOI] [PubMed] [Google Scholar]
- [19].Wang X, Liu D, Zeng X, et al. Dual-energy CT quantitative parameters for evaluating Immunohistochemical biomarkers of invasive breast cancer. Cancer Imaging 2021;21:4. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [20].Barbara Krug K, Schomig-Markiefka B, Campbell GM, et al. Correlation of CT-data derived from multiparametric dual-layer CT-maps with immunohistochemical biomarkers in invasive breast carcinomas. Eur J Radiol 2022;156:110544. [DOI] [PubMed] [Google Scholar]
- [21].Wang X, Liu D, Zeng X, et al. Dual-energy CT quantitative parameters for the differentiation of benign from malignant lesions and the prediction of histopathological and molecular subtypes in breast cancer. Quant Imaging Med Surg 2021;11:1946–57. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [22].Agha RA, Mathew G, Rashid R, et al. Revised Strengthening the reporting of cohort, cross-sectional and case-control studies in surgery (STROCSS) guideline: an update for the age of Artificial Intelligence. Prem J Sci 2025;10:100081. [Google Scholar]
- [23].Wang L, Wang X, Jiang F, et al. Adding quantitative T1rho-weighted imaging to conventional MRI improves specificity and sensitivity for differentiating malignant from benign breast lesions. Magn Reson Imaging 2024;108:98–103. [DOI] [PubMed] [Google Scholar]
- [24].Dai Y, Lian C, Zhang Z, et al. Development and validation of a deep learning system to differentiate HER2-Zero, HER2-Low, and HER2-positive breast cancer based on dynamic contrast-enhanced MRI. J Magn Reson Imaging 2025;61:2212–20. [DOI] [PubMed] [Google Scholar]
- [25].Xia H, Chen Y, Cao A, et al. Differentiating between benign and malignant breast lesions using dual-energy CT-based model: development and validation. Insights Imaging 2024;15:173. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [26].Wang X, Du L, Cao Y, et al. Comparing extracellular volume fraction with apparent diffusion coefficient for the characterization of breast tumors. Eur J Radiol 2024;171:111268. [DOI] [PubMed] [Google Scholar]
- [27].Lan X, Wang X, Qi J, et al. Application of machine learning with multiparametric dual-energy computed tomography of the breast to differentiate between benign and malignant lesions. Quant Imaging Med Surg 2022;12:810–22. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [28].Klein K, Gabriel Schafigh D, Schomig-Markiefka B, et al. Intermodal correlation of quantitative CT-data and MRI-biomarkers derived from synchronous spectral CT-maps and breast MRI-examinations with molecular biomarkers in invasive ductal breast carcinomas. Eur J Radiol 2023;165:110919. [DOI] [PubMed] [Google Scholar]
- [29].Musulin J, Stifanic D, Zulijani A, et al. An enhanced histopathology analysis: an AI-based system for multiclass grading of oral squamous cell carcinoma and segmenting of epithelial and stromal tissue. Cancers (Basel) 2021;13:1784. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [30].Zhang G, Ren C, Li C, et 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] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
The datasets generated during and analyzed during the current study are available by the corresponding author J.Z., upon reasonable request.