DCE-MRI Based Machine Learning Predictor for HER2-Positive Breast Cancer: A Feasibility and Validation Multicenter Study

DCE-MRI Based Machine Learning Predictor for HER2-Positive Breast Cancer: A Feasibility and Validation Multicenter Study | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article DCE-MRI Based Machine Learning Predictor for HER2-Positive Breast Cancer: A Feasibility and Validation Multicenter Study Chunli Kong, Guihan Lin, Weiyue Chen, Xue Cheng, Shuang Liu, Di Shen, and 7 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4152618/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background Human epidermal growth factor receptor 2 (HER2) status of breast cancer plays a critical role in guiding clinical treatment. We aimed to develop and validate a predictive model for HER2 status using preoperative dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI). Methods A total of 570 patients (282, 121 and 167 patients for training, internal and external test sets, respectively) with pathologically confirmed breast cancer and known HER2 status were recruited. A total of 851 radiomics features for each patient were extracted from preoperative DCE-MRI images. VarianceThreshold, SelectKBest and least absolute shrinkage and selection operator (LASSO) regression were applied to identify the optimal predictive features. Logistic regression was adopted to incorporate the Rad-score and clinical predictors into a nomogram. The performance of the nomogram was evaluated by area under receiver operating characteristic curve (AUC), calibration curve and decision curve. Additionally, gene expression analysis based on the Cancer Image Archive database was conducted to validate the biological interpretability of the model. Results Twenty-three radiomics features were selected to calculate the Rad-score. The Rad-score, along with breast imaging reporting and data system (BI-RADS) parameter, were independent predictors for HER2 status and were incorporated into the predictive model. The combined model achieved AUCs of 0.881, 0.883, and 0.798 in the training, internal and external test sets, respectively. Calibration curves demonstrated well agreement between the model predictions and actual HER2 status. Decision curve analysis further confirmed the clinical utility of the model. Differentially expressed genes between HER2-positive and HER2-negative patients were primarily involved in signaling pathways such as PI3K-AKT, endocrine resistance, and p53. Conclusions The combined model, which incorporated the Rad-score and BI-RADS, representing a potential and efficient alternative tool to evaluate HER2 status in breast cancer. Breast cancer HER2 status BI-RADS Nomogram PI3K-AKT Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Background Breast cancer is the most prevalent malignant cancer in women worldwide, with over 1.3 million cases diagnosed and 400,000 deaths each year [ 1 ]. Approximately 20%-30% of invasive breast cancers overexpress the human epidermal growth factor receptor 2 (HER2) tyrosine kinase receptor [ 2 ]. Breast cancer patients with HER2-positive tumors are known to exhibit more aggressive characteristics, such as lower differentiation, high rates of cell proliferation, early systemic metastasis, and a higher risk of recurrence and mortality [ 3 , 4 ]. Additionally, HER2-positive patients tend to be more resistant to chemotherapy, but they have been found to respond well to targeted antibody therapy [ 5 , 6 ]. Clinical trials have confirmed that patients with HER2-positive treated with anti-HER2 therapy combined with adjuvant chemotherapy have achieved a prolonged time to disease progression, longer overall survival rate, and higher rates of pathological complete response rate compared to patients treated with chemotherapy alone [ 7 – 9 ]. Therefore, it is crucial to accurately identify HER2 status before surgery in order to guide individual treatment plan and predict the prognosis of breast cancer. In clinical practice, it is mainly based on core needle biopsy-based immunohistochemical (IHC) or fluorescence in situ hybridization (FISH) assays to assess HER2 status of breast cancers [ 10 ]. This approach relies on a single biopsy of a heterogeneous tumor, which may only capture a tiny portion of a potentially heterogeneous lesion, and it may not be representative of the whole tumor’s genetic, epigenetic, or phenotypic alterations [ 11 – 13 ]. Meanwhile, the assessment is performed at specific time points, which may not reflect the dynamic changes in tumor biology over time and during treatment [ 14 ]. Hence, it is necessary to develop a non-invasive method that can reflect the overall heterogeneity of tumors and be used for real-time detection of HER2 status for breast cancers. Medical imaging offers a noninvasive approach that can detect a wider range of tumor heterogeneity [ 15 , 16 ]. Radiomics, a rapidly emerging field, can transform medical images into high dimensional computer-based data. Multi-parametric magnetic resonance imaging (MRI), has become a routine examination in clinical practice for diagnosis, preoperative staging assessment, and treatment responsiveness assessment [ 17 ]. In particular, dynamic contrast-enhanced (DCE) imaging is considered the most sensitivity modality for detecting breast cancer as it provide temporal information on the kinetics of the contrast agent in suspicious lesions and offers sufficient spatial resolution [ 18 ]. Prior studies have explored the relationship between radiomics signature and tumor biology characteristics, but the generalizability of these findings is limited due to the use of different MRI protocols and scanners [ 19 , 20 ]. Additionally, these studies had limitations in terms of the number of radiomics features studied and the lack of comparison with the predictive efficiency based on conventional MR features [ 21 ]. Therefore, we aimed to develop an accurate and robust model for predicting HER2 status based DCE-MRI images. Furthermore, we identified differentially expressed genes related to HER2 status in the Cancer Imaging Archive (TCIA) dataset and conducted Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) functional enrichment analyses based on these differential genes to gain insights into the biological behaviors associated with HER2 status. The model developed in this study has the potential to assist in decision-making and patient management in clinical practice, thereby facilitating precision medicine. Methods Patients This study was reviewed and approved by the Institutional Review Boards and Human Ethics Committee of the Fifth Affiliated Hospital of Wenzhou Medical University (center A) and the First Affiliated Hospital of Zhejiang University (center B). The requirement for informed consent was waived in accordance with the Helsinki statement. Patients with pathologically confirmed breast cancer were consecutively enrolled from June 2018 to December 2022 at the two aforementioned centers. The inclusion and exclusion criteria were as follows: (1) The primary tumor was invasive ductal carcinoma; (2) HER2 status was determined using IHC and fluorescence in situ hybridization (FISH) in surgical specimens; (3) DCE-MRI examination was performed prior to surgery; (4) patients with lesion size larger than 1 cm to minimize the impact of partial volume on radiomics analysis; The exclusion criteria were as follows: (1) patients with insufficient clinicopathological data or DCE-MR scans; (2) patients who received neoadjuvant chemotherapy or any other treatment before DCE-MR imaging collection; (3) HER2 status was determined by biopsy before surgery; and (4) patients with a history of malignant tumors. Finally, a total of 570 patients were enrolled in, and the screening process is shown in Fig. 1 . The enrolled patients from center A were divided into a training set of 281 and an internal test set of 122 using a random number table method at a ratio of 7:3. The clinical information, including age, menopausal status, tumor size, breast imaging reporting and data system (BI-RADS), and time-signal intensity curve (TIC) in DCE-MRI scanning, were collected. Patients from center B were used as an independent external test set. Besides, 22 patients from TCIA were recruited as an outcome prediction set to explore the biologic functions. Pathological evaluation The HER2 status was determined according to the American Society of Clinical Oncology (ASCO)/College of American Pathologists (CAP) detection guidelines. HER2 status was divided into HER2-positive and HER2-negative [ 22 ]. The HER2 status was determined by HER2 IHC, and FISH was used if necessary. The IHC score were divided into 0+, 1+, 2 + and 3 + based on the staining status of the membrane and the proportion of the invasive tumor cells. IHC score of 0/1 + were considered HER2-negative, and 3 + was defined as HER2-positive, while the 2 + required further confirmation of amplification status by FISH. MR image acquisition For patients from center A, all breast MR examinations were performed on a 1.5 Tesla MR system (Siemens Healthcare, Erlangen, Germany) utilizing a dedicated eight-channel breast radiofrequency coil with patient placed in the prone position. DCE-MRI was performed using flash low angle shot 3D sequence. The specific scanning parameters are as follows: TR = 4.4 ms, TE = 1.5 ms, flip angle = 12°, slice thickness = 1.5 mm, Matrix = 224×192, FOV = 350 mm×350 mm. After the first dynamic scan was completed, a bolus of 0.2 mmol/kg gadopentetate dimeglumine (Magnevist; Bayer HealthCare, Hanover, NJ, USA) was injected into the cubital vein with a high-pressure syringe at a rate of 2.0 ml/s. Continuous non-interval scans were done in seven stages after intravenous injection, with a scan period of 60 seconds for each phase. For patients from center B, all breast MR examinations were performed on a 3.0 Tesla MR system (Signa HDxt GE healthcare, America) using an eight-channel dedicated surface coil. Patients were positioned in the prone position, with bilateral breasts naturally overhanging in the coil. Gd-DTPA was used as a contrast agent. A dose of 0.1 mmol/kg Gd-DTPA was injected into the elbow vein at a flow rate of 2.0 ml/s, followed by the injection of 15 ml of normal saline at the same flow rate. DCE scans were performed using a cross-sectional three-dimensional fast low angle shot (3D FLASH) technique. The specific scanning parameters are as follows: TR = 4.51 ms, TE = 1.61 ms, flip angle = 10°, slice thickness = 1.0 mm, Matrix = 420 × 420, FOV = 340 mm × 340 mm. The scans were initiated 25 seconds after the start of the contrast agent injection and were repeated for 6 consecutive scans. Each scan had a duration of 58 seconds. In this study, the second phase axial images of DCE-MRI were chosen for further analysis since it showed the lesion borders more clearly. Image segmentation and features extraction All MRI images were imported into the open-source image processing software 3D-Slicer ( https://www.slicer.org/ ). The segmentation of the three-dimensional tumor area is conducted by reader A (JGH, with 10 years of experience) layer by layer along the tumor edge using the 3D-Slicer platform. Then, reader B (XC, with 12 years of experience) verified the segmentation results, and any discrepancies were resolved through discussion. The radiologists are blinded to the pathologic results. Image preprocessing and features extraction is conducted with Pyradiomics package (version 2.12; https://pyradiomics.readthedocs.io/en/2.1.2/ ) in python 3.7.0 [ 23 ]. All images were resampled to a voxel size of 1×1×1 mm 3 with a fixed bin width of 25 to standardize the voxel spacing and decrease the image noise. A total of 851 radiomics features were extracted for each volume of interest (VOI). In order to secure the reproducibility of the extracted features, 60% images were randomly selected and were delineated by the same radiologist (reader A) one month later to calculate the intra-class correlation coefficient (ICC). Then the selected 60% images were also segmented independently by the second radiologist (reader B) to calculate the inter-class ICC. Features with ICCs greater than 0.75 is considered to be good reproducibility and will be remained for further analysis. Intra-class and Inter-class reproducibility of radiomics features For the radiomics features extraction, there was no significant difference between the two radiologists, with p value ranging from 0.769 ~ 0.872. The intra-class ICC calculated with twice segmentation of reader A ranged from 0.858 to 0.957, and the inter-class ICC calculated by the features extracted by reader A and reader B ranged from 0.769 to 0.865. The reproducibility of the radiomics features is relatively well and all the 851 features extracted from the DCE-MRI sequence of each patient were remained for further analysis. Feature Selection and Model Construction Before performing feature dimensionality reduction and feature selection, we standardized the features with ICCs greater than 0.75 using the Z-score method to reduce the influence of the different units imposed by the units of each feature and improve the performance of the model. The Z-score method is defined as follows: z=(x-µ)/σ, where x is the value of current parameter, µ is the mean value of x, and σ is the corresponding deviation. Next, a three-step procedure was conducted for dimensionality reduction and selection of task-specific radiomics features in the training cohort. Firstly, the variance threshold method was applied to preliminarily select features with a variance greater than 0.8. Secondly, SelectKBest method was utilized to calculate the P-value for each feature, which indicates the correlation between the feature and the label (HER2 status). Only features with a P-value less than 0.01 were retained for further analysis. Thirdly, least absolute shrinkage and selection operator logistic regression (LASSO) with five-fold cross-validation were applied to screen the features that were most relevant to HER2 status of breast cancers. For LASSO regression, the regularization parameter λ controls the number of radiomics features and affects the performance of the model. The λ was optimized by five-fold cross-validation in the training cohort to minimize the mean square error. Finally, the radiomics signature for each patient was calculated using logistic regression. To further improve the performance of the prediction model, we also included clinical indicators and imaging examination indicators for analysis. Univariate and multivariate analyses were conducted to screen the predictive variables for HER2 status, which were then integrated with the radiomics signature to establish the predictive nomogram. The performance of the nomogram was evaluated using receiver operating characteristic curve (ROC), calibration curve and decision curve. Outcome Prediction and Exploring Biologic Functions Patients in the TCIA dataset were also enrolled in this study to analyze the biologic functions of breast cancers. We used the HER2 status prediction model built in this study to predict HER2 status for samples in the TCIA dataset. The “limma” package was applied to analyze the differentially expressed genes (DEGs) between the model predicted HER2 positive and HER2 negative groups, and a threshold P < 0.01 was set. Based on the DEGs, GO and KEGG functional enrichment analyses were performed with thresholds of P < 0.05. Statistical Analysis IBM SPSS (Chicago, IL), R software (version 3.6.3) and Python (version 3.7.0) were used for statistical analysis in this study. The continuous variables with a normal distribution were presented as mean and standard deviation, and the continuous variables with non-normal distribution were presented as median or interquartile range. The normality test of the continuous data distribution was performed by Kolmogorov-Smirnov test. Then the t test was applied to analyze the continuous variables with normal distribution, and the Mann-Whitney U test was applied to analyze the continuous variables with abnormal distribution. The categorical variables were compared with chi-square test or Fisher exact test and presented as percentages. A P value less than 0.05 was deemed to be statistically significant in all statistical analysis. Results Baseline Characteristics Baseline characteristics of the patients in the training and internal test set are summarized in Supplementary Table S1 . There is no statistical significance in age, menopausal status, tumor size, BI-RADS and TIC between the training and internal test set. All characteristics are comparable between two cohorts. Clinical and Imaging Examination Indicators for HER2 Status Clinical and imaging examination characteristics of patients with HER2-positive and HER2-negative in training and internal test set are summarized in Table 1. 281 cases and 122 cases were assigned to the training and internal test set, respectively. Univariate and multivariate logistic regression analyses were performed to identify indicators associated with HER2 status, and corresponding results are summarized in Table 2. As is shown, BI-RADS was screened out as the predictive clinical indicators that was valuable in predicting HER2 status of breast cancer. Table 1 Clinical characteristics of patients with HER2-positive and HER2-negative in the training and internal test sets Characteristic training set P Value internal test set P Value HER2-positive (n=108) HER2-negative (n=173) HER2-positive (n=47) HER2-negative (n=75) Age (years) 0.218 0.713 ＞55 34 (31.5%) 67 (38.7%) 16 (41.1%) 28 (37.3%) ≤55 74 (68.5%) 106 (61.3%) 31 (65.9%) 47 (62.7%) Menopausal status 0.190 0.132 Premenopausal 50 (53.7%) 79 (45.7%) 26 (55.4%) 31 (41.4%) Postmenopausal 58 (46.3%) 94 (54.3%) 21 (44.6%) 44 (58.6%) Tumor size (cm) 0.686 0.236 ＞2 38 (35.1%) 65 (37.5%) 22 (46.8%) 27 (36.0%) ≤2 70 (64.9%) 108 (62.5%) 25 (53.2%) 48 (64.0%) BI-RADS < 0.001 0.002 4A 14 (12.9%) 47 (27.2%) 4 (8.5%) 17 (22.7%) 4B 29 (26.9%) 84 (48.5%) 13 (27.6%) 35 (46.7%) 4C 42 (38.9%) 27 (15.6%) 21 (44.7%) 13 (17.2%) 5 23 (21.3%) 15 (8.7%) 9 (19.2%) 10 (13.4%) TIC 0.255 0.691 Wash-in 35 (32.4%) 44 (25.4%) 11 (23.4%) 22 (29.3%) Platform 56 (51.9%) 107 (61.8%) 27 (57.4%) 42 (56.1%) Wash-out 17 (15.7%) 22 (12.8%) 9 (19.2%) 11 (14.6%) Abbreviations: BI-RADS, breast imaging reporting and data system; TIC, time signal intensity curve. Table 2 Univariate and Multivariate analysis of risk factors for HER2-positive breast cancer in the training set Characteristic Univariate analysis Multivariate analysis OR (95% CI) P Value OR (95% CI) P Value Age (years) 0.727 (0.437-1.209) 0.219 Menopausal Status 0.725 (0.447-1.173) 0.190 Tumor size (cm) 0.902 (0.547-1.488) 0.686 BI-RADS 2.070 (1.568-2.732) < 0.001 2.070(1.568-0.732) < 0.001 TIC 0.906 (0.619-1.326) 0.611 Abbreviations: BI-RADS, breast imaging reporting and data system; TIC, time signal intensity curve. Construction and Validation of the HER2 Status Predictive Model The flowchart of this study is illustrated in Fig. 2 . For each patient, a total of 851 radiomics features were extracted. It is important to note that all 851 features exhibited good reproducibility, with ICCs exceeding 0.75. Thus, all 851 features were utilized to calculate the variance by using the variance threshold method. After that, 851 features were retained as they had a variance greater than 0.8. Next, the SelectKBest method was utilized to calculate the P value for each feature. This step resulted in the retention of 280 features for LASSO regression. LASSO regression process is illustrated in Fig. S1A-B . Finally, 23 key radiomics features were identified ( Appendix E1.1, Fig. S1C ), and the Rad-score for each patient was calculated based these features by using the logistic regression. Performance of the Radiomics Signature Rad-score can be used to distinguish the HER2-positive patients from the HER2-negatives for there is a significance between the two groups. The median value of Rad-score in HER2-positive patients is higher than that in the HER2-negative patients in both the training, internal and external test sets ( Fig. 3A-C ). In order to analyze the performance of the diagnostic signature in detail, we also depicted the ROC curves in different sets ( Fig. 3D-F ). In the training set, it achieved an AUC of 0.825, with an accuracy of 79.9%, a specificity of 85.5%, and a sensitivity of 66.7%. In the internal test set, it achieved an AUC of 0.818, with an accuracy of 81.1%, a specificity of 88.0%, and a sensitivity of 70.2%. In the external test set, it achieved an AUC of 0.748, with an accuracy of 81.4%, a specificity of 91.0%, and a sensitivity of 62.2%. The calibration curves for the diagnostic signature in training, internal and external test set are shown in Fig. 3G-I . Development and Validation of Individualized Prediction Model for HER2 Status To further improve the performance of the predictive model, the imaging examination indicator BI-RADS was incorporated with the Rad-score to construct a more efficiency prediction model. As shown in Fig. 4A , the combined model achieved an AUC of 0.881 in the training set, with an accuracy of 79.4%, a specificity of 92.5%, and a sensitivity of 73.1%. It yields an AUC of 0.883 in internal test set, with an accuracy of 80.3%, a specificity of 82.7%, and a sensitivity of 83.0% ( Fig. 4B ). It yields an AUC of 0.798 in external test set, with an accuracy of 80.2%, a specificity of 84.4%, and a sensitivity of 71.1% ( Fig. 4C ). To evaluate the fitting performance between the model predicted and the actual HER2 status, we developed calibration curves of the training, internal test and external test sets ( Fig. 4D-F ). The calibration curves indicated that the model achieved well agreement between the predicted probabilities and the actual outcomes, which suggested a favorable fitting performance. Based on the reference standard of pathology results, the percentages of true positive, false positive, true negative, and false negative results were calculated and displayed as confusion matrix diagrams ( Fig. 4G-I ). Predictive performance of the clinical model, radiomics model and combined model in the training, internal test and external test sets are summarized in Table 3 . Table 3 Predictive performance of the clinical model, radiomics model and combined model in the training, internal test and external test sets. Model Training Set Internal Test Set External Test Set Model Parameter AUC ACC Specificity Sensitivity AUC ACC Specificity Sensitivity AUC ACC Specificity Sensitivity Clinical Model 0.710 0.712 0.757 0.639 0.664 0.672 0.693 0.638 0.647 0.515 0.361 0.933 Radiomics Model 0.825 0.799 0.855 0.667 0.818 0.811 0.880 0.702 0.748 0.832 0.910 0.622 Combined Model 0.881 0.794 0.925 0.731 0.883 0.803 0.827 0.830 0.798 0.629 0.844 0.711 Abbreviations: AUC, Area under the curve; ACC, Accuracy. Finally, we visualized the combined model as a nomogram to individually predict the HER2 status for each patient ( Fig. 5A ). In addition, the decision curve of the training set was also plotted to verify the clinical utility of the predictive signature and the combined model. As illustrated in Fig. 5B , the integration of the combined model has the potential to greatly enhance the clinical management of breast cancer, maximizing its benefits in practice. Clinical Outcomes and Biologic Functions Associated with HER2 Status 22 patients from TCIA database were enrolled in this study. The exclusion criteria for TCIA database were as follow: (1) Missing of DCE-MR images; (2) poor registration quality; and (3) patients without mRNA data. Finally, a total of 22 patients with both available DCE-MR images and RNA-sequence data were included. The DCE-MR images of the enrolled 22 patients were downloaded from the TCIA dataset (https://www.cancerimagingarchive.net/). Tumor segmentation and features extraction were performed by the same radiologists in this study. Using the HER2 status prediction model developed in this study, we applied it to predict the HER2 status on the 22 samples that were enrolled in the study. Among the 22 samples, 3 samples were predicted to be HER2 positive, and the remaining 19 patients were predicted to be HER2 negative. Analysis of the Biologic Function RNA was extracted from both the tumor specimens. All cells from the tumor specimen were utilized to identify immune cells based on bulk RNA-seq analysis. Gene expression levels were calculated using the FPKM format. A total of 117 HER2 status related genes were identified ( Fig. 6A ). KEGG enrichment analysis revealed that differentially expressed genes (DEGs) were mainly enriched in endocrine resistance, PI3K-AKT signaling pathway, and p53 ( Fig. 6B ). GO enrichment analysis showed that DEGs were mainly involved in signaling pathways such as mammary gland epithelium development, epithelial cell proliferation, and regulation of epithelial cell proliferation ( Fig. 6C ). Discussion HER2 is a pivotal molecular target in the treatment of breast cancer. In present study, we investigated the potential association between HER2 status, quantitative imaging features extracted from DCE-MRI, and the MR reported BI-RADS score. The model could accurately predict HER2 status preoperatively. This predictive model performed exceptionally well in three independent cohorts. Enrichment pathway analysis was conducted on the HER2-positive and HER2-negative groups, which were classified by the model constructed in this study. Pathway enrichment analysis provides valuable biological insights into the prediction model developed in this study. In recent years, numerous researchers have explored the correlation between traditional imaging features and HER2 status. Previous studies have confirmed that conventional imaging features, such as masses with indistinct margins, calcifications within the masses, segmental calcifications, and microcalcifications are significantly associated with HER2-positive status [24-26]. However, these studies only explored the qualitative relationship between traditional imaging features and HER2 status, but didn’t explore the quantitative performance in identifying HER2 status for further step. In contrast, our study also included imaging examination and clinical characteristics for analysis to screen the indicators that were valuable in predicting HER2 status, and quantitatively assessed the performance of the screened indicators for predicting HER2 status. In our study, only BI-RADS showed statistically significant differences between HER2-positive and negative individuals. We integrated BI-RADS with Rad-score to develop a combined model. The model achieved improved AUCs in the training, internal test and external test set, respectively. These results suggest that our model is robust in predicting HER2 status of breast cancer. In recent years, the texture analysis-based radiomics has gained popularity in cancer research. Numerous studies have demonstrated that radiomics has the advantages of reflecting tumor heterogeneity, and there is an intrinsic correlation between tumor genotypes and radiomics features [27, 28]. Zhou et al [29] constructed a predictive model for breast cancer HER2 status based on T2-weighted sequences and DCE-T1 axial sequences. Their model achieved AUCs of 0.86 and 0.81 in the training and validation set, respectively. However, joint modeling based on multiple sequences is time-consuming in clinical practice. Therefore, in our study, we focused on extracting radiomics features and building a predictive model based solely on the DCE-MRI sequence. However, our predictive model outperformed Zhou's model, which may be attributed to our larger sample size, more scientific study design, and more efficient algorithm. Another study attempted to predict HER2 status based on MRI images, but their model only yielded an AUC of 0.65 [30]. This may be due to the fact that they only extracted 38 radiomics features from the DCE-MRI images and had a small sample size of only 91 samples. There have been studies attempting to predict HER2 status using mammography or CT imaging features, but their accuracy, as measured by the AUC, is significantly lower than that in our study [31, 32]. Additionally, mammography and CT scans involve small doses of radiation and have lower soft tissue resolution compared to MRI. This is why the current study relies on dynamic DCE-MRI sequences to predict HER2 status. One major drawback of radiomics models based on machine learning algorithm is their lack of biological interpretability, which hinders its widespread adoption and application in clinical practice. Further research is needed to elucidate the underlying mechanisms that enable clinicians to interpret the biological significance of radiomic features and the predictive power of these models. Therefore, we utilized the model to predict the HER2 status for samples in the TCIA dataset. Differential gene analysis was then conducted based on the HER2-positive and HER2-negative samples classified by the model. Enrichment pathway analysis was performed based on the differentially expressed genes, providing biological interpretability to the radiomics model proposed in this study. The KEGG analysis revealed that differentially expressed genes associated with HER2 status were commonly involved in major signaling pathways such as PI3K-AKT signaling pathway, endocrine resistance, and p53. The PI3K-AKT signaling pathway has been extensively studied and is known to play a crucial role in various cellular processes, including cell proliferation, survival, invasion, migration, apoptosis, glucose metabolism, and DNA repair [33, 34]. A study reported that the PI3K-AKT pathway is activated in approximately half of hormone receptor-positive and HER2-negative breast cancers by means of activating mutations in PIK3CA and AKT1 and inactivating alterations in PTEN [35]. These patients are prone to endocrine therapy resistance. This preliminary research provides biological significance to the radiomics model proposed in this study. However, there are also some limitations to this study. Firstly, it was a retrospective study, which may introduce potential selection bias. Secondly, manual segmentation of tumors was time-consuming and could be replaced by artificial intelligence to improve reproducibility. Conclusions This study presents a radiomics signature and BI-RADS-based nomogram for predicting HER2 status in breast cancer patients before surgery. The proposed model was validated using enrichment pathway analysis on two groups, providing biological insights into the developed model. The model is user-friendly, accurate, and non-invasive, which can aid clinicians in identifying suitable candidates for anti-HER2 treatment in clinical practice. Furthermore, this study introduces novel approaches for more precise and personalized management of breast cancer. Abbreviations MRI Magnetic resonance imaging DCE-MRI Dynamic contrast enhanced MRI HER2 Human epidermal growth factor receptor 2 CT Computed tomography LASSO Least absolute shrinkage and selection operator ROC Receiver operating characteristic curve BI-RADS Breast imaging reporting and data system AUC Area under the curve VOI Volume of interest ICC Intra-class correlation coefficient IHC Immunohistochemistry FISH Fluorescence in situ hybridization TIC Time signal intensity curve GO Gene ontology KEGG Kyoto encyclopedia of genes and genomes Declarations Ethics approval and consent to participate The study was approved by the Institutional Review Boards and Human Ethics Committee of the Fifth Affiliated Hospital of Wenzhou Medical University and the First Affiliated Hospital of Zhejiang University. The requirement for informed consent was waived in accordance with the Helsinki statement. Consent for publication Not applicable. Availability of data and materials The data presented in this study are available on reasonable request from the corresponding author. Competing interests All authors declare that they have no competing interests. Funding This study was supported by Medical and Health General Project of Zhejiang Province (2024KY1860, 2023KY425). Authors’ contributions This study was conceived by JJ, ZP, and MX with inputs from all other authors. XC, SL, DS, JD, JH, and SX were involved in data acquisition and image processing. CK and GL conducted the analyses. CK, GL, WC and MC were involved in literature research, data interpretation and the initial draft of the manuscript. JJ, ZP, and MX were involved in data interpretation and critical revision of the manuscript. All authors approved the final version of the manuscript. Acknowledgements Not Applicable. References Siegel RL, Miller KD, Fuchs HE, Jemal A. Cancer Statistics, 2021. CA: a cancer journal for clinicians. 2021; 71(1):7-33. Bitencourt AGV, Gibbs P, Rossi Saccarelli C, Daimiel I, Lo Gullo R, Fox MJ , et al. MRI-based machine learning radiomics can predict HER2 expression level and pathologic response after neoadjuvant therapy in HER2 overexpressing breast cancer. EBioMedicine. 2020; 61:103042. Ahn S, Woo JW, Lee K, Park SY. HER2 status in breast cancer: changes in guidelines and complicating factors for interpretation. Journal of pathology and translational medicine. 2020; 54(1):34-44. Loibl S, Gianni L. HER2-positive breast cancer. Lancet (London, England). 2017; 389(10087):2415-2429. Lam SW, Jimenez CR, Boven E. Breast cancer classification by proteomic technologies: current state of knowledge. Cancer treatment reviews. 2014; 40(1):129-138. Waks AG, Winer EP. Breast Cancer Treatment: A Review. Jama. 2019; 321(3):288-300. von Minckwitz G, Procter M, de Azambuja E, Zardavas D, Benyunes M, Viale G , et al. Adjuvant Pertuzumab and Trastuzumab in Early HER2-Positive Breast Cancer. The New England journal of medicine. 2017; 377(2):122-131. Swain SM, Miles D, Kim SB, Im YH, Im SA, Semiglazov V , et al. Pertuzumab, trastuzumab, and docetaxel for HER2-positive metastatic breast cancer (CLEOPATRA): end-of-study results from a double-blind, randomised, placebo-controlled, phase 3 study. The Lancet Oncology. 2020; 21(4):519-530. Slamon DJ, Leyland-Jones B, Shak S, Fuchs H, Paton V, Bajamonde A , et al. Use of chemotherapy plus a monoclonal antibody against HER2 for metastatic breast cancer that overexpresses HER2. The New England journal of medicine. 2001; 344(11):783-792. Orlando L, Viale G, Bria E, Lutrino ES, Sperduti I, Carbognin L , et al. Discordance in pathology report after central pathology review: Implications for breast cancer adjuvant treatment. Breast (Edinburgh, Scotland). 2016; 30:151-155. Leithner D, Bernard-Davila B, Martinez DF, Horvat JV, Jochelson MS, Marino MA , et al. Radiomic Signatures Derived from Diffusion-Weighted Imaging for the Assessment of Breast Cancer Receptor Status and Molecular Subtypes. Molecular imaging and biology. 2020; 22(2):453-461. Ma W, Zhao Y, Ji Y, Guo X, Jian X, Liu P , et al. Breast Cancer Molecular Subtype Prediction by Mammographic Radiomic Features. Academic radiology. 2019; 26(2):196-201. Leithner D, Horvat JV, Marino MA, Bernard-Davila B, Jochelson MS, Ochoa-Albiztegui RE , et al. Radiomic signatures with contrast-enhanced magnetic resonance imaging for the assessment of breast cancer receptor status and molecular subtypes: initial results. Breast cancer research : BCR. 2019; 21(1):106. Pisco AO, Huang S. Non-genetic cancer cell plasticity and therapy-induced stemness in tumour relapse: 'What does not kill me strengthens me'. British journal of cancer. 2015; 112(11):1725-1732. Gillies RJ, Kinahan PE, Hricak H. Radiomics: Images Are More than Pictures, They Are Data. Radiology. 2016; 278(2):563-577. Goto M, Le Bihan D, Sakai K, Yamada K. The diffusion MRI signature index is highly correlated with immunohistochemical status and molecular subtype of invasive breast carcinoma. European radiology. 2022; 32(7):4879-4888. Huang Y, Wei L, Hu Y, Shao N, Lin Y, He S , et al. Multi-Parametric MRI-Based Radiomics Models for Predicting Molecular Subtype and Androgen Receptor Expression in Breast Cancer. Frontiers in oncology. 2021; 11:706733. Li W, Yu K, Feng C, Zhao D. Molecular Subtypes Recognition of Breast Cancer in Dynamic Contrast-Enhanced Breast Magnetic Resonance Imaging Phenotypes from Radiomics Data. Computational and mathematical methods in medicine. 2019; 2019:6978650. Waugh SA, Purdie CA, Jordan LB, Vinnicombe S, Lerski RA, Martin P , et al. Magnetic resonance imaging texture analysis classification of primary breast cancer. European radiology. 2016; 26(2):322-330. Sun X, He B, Luo X, Li Y, Cao J, Wang J , et al. Preliminary Study on Molecular Subtypes of Breast Cancer Based on Magnetic Resonance Imaging Texture Analysis. Journal of computer assisted tomography. 2018; 42(4):531-535. Xie T, Zhao Q, Fu C, Bai Q, Zhou X, Li L , et al. Differentiation of triple-negative breast cancer from other subtypes through whole-tumor histogram analysis on multiparametric MR imaging. European radiology. 2019; 29(5):2535-2544. Wolff AC, Hammond MEH, Allison KH, Harvey BE, Mangu PB, Bartlett JMS , 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. Journal of clinical oncology : official journal of the American Society of Clinical Oncology. 2018; 36(20):2105-2122. van Griethuysen JJM, Fedorov A, Parmar C, Hosny A, Aucoin N, Narayan V , et al. Computational Radiomics System to Decode the Radiographic Phenotype. Cancer research. 2017; 77(21):e104-e107. Shin HJ, Kim HH, Huh MO, Kim MJ, Yi A, Kim H , et al. Correlation between mammographic and sonographic findings and prognostic factors in patients with node-negative invasive breast cancer. The British journal of radiology. 2011; 84(997):19-30. Wang X, Chao L, Chen L, Tian B, Ma G, Zang Y , et al. Correlation of mammographic calcifications with Her-2/neu overexpression in primary breast carcinomas. Journal of digital imaging. 2008; 21(2):170-176. Elias SG, Adams A, Wisner DJ, Esserman LJ, van't Veer LJ, Mali WP , et al. Imaging features of HER2 overexpression in breast cancer: a systematic review and meta-analysis. Cancer epidemiology, biomarkers & prevention : a publication of the American Association for Cancer Research, cosponsored by the American Society of Preventive Oncology. 2014; 23(8):1464-1483. Qi Y, Zhao T, Han M. The application of radiomics in predicting gene mutations in cancer. European radiology. 2022; 32(6):4014-4024. Demircioglu A, Grueneisen J, Ingenwerth M, Hoffmann O, Pinker-Domenig K, Morris E , et al. A rapid volume of interest-based approach of radiomics analysis of breast MRI for tumor decoding and phenotyping of breast cancer. PloS one. 2020; 15(6):e0234871. Zhou J, Tan H, Li W, Liu Z, Wu Y, Bai Y , et al. Radiomics Signatures Based on Multiparametric MRI for the Preoperative Prediction of the HER2 Status of Patients with Breast Cancer. Academic radiology. 2021; 28(10):1352-1360. Li H, Zhu Y, Burnside ES, Huang E, Drukker K, Hoadley KA , et al. Quantitative MRI radiomics in the prediction of molecular classifications of breast cancer subtypes in the TCGA/TCIA data set. NPJ breast cancer. 2016; 2:16012-. Zhou J, Tan H, Bai Y, Li J, Lu Q, Chen R , et al. Evaluating the HER-2 status of breast cancer using mammography radiomics features. European journal of radiology. 2019; 121:108718. Yang X, Wu L, Zhao K, Ye W, Liu W, Wang Y , et al. Evaluation of human epidermal growth factor receptor 2 status of breast cancer using preoperative multidetector computed tomography with deep learning and handcrafted radiomics features. Chinese journal of cancer research = Chung-kuo yen cheng yen chiu. 2020; 32(2):175-185. Miricescu D, Totan A, Stanescu S, II, Badoiu SC, Stefani C, Greabu M. PI3K/AKT/mTOR Signaling Pathway in Breast Cancer: From Molecular Landscape to Clinical Aspects. International journal of molecular sciences. 2020; 22(1). Xing Y, Lin NU, Maurer MA, Chen H, Mahvash A, Sahin A , et al. Phase II trial of AKT inhibitor MK-2206 in patients with advanced breast cancer who have tumors with PIK3CA or AKT mutations, and/or PTEN loss/PTEN mutation. Breast cancer research : BCR. 2019; 21(1):78. Turner NC, Oliveira M, Howell SJ, Dalenc F, Cortes J, Gomez Moreno HL , et al. Capivasertib in Hormone Receptor-Positive Advanced Breast Cancer. The New England journal of medicine. 2023; 388(22):2058-2070. Additional Declarations No competing interests reported. Supplementary Files SupplementaryMaterial.docx Fig.S1.pdf Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-4152618","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":284712186,"identity":"1d812006-21ac-4be0-8f69-66bb9fea82ba","order_by":0,"name":"Chunli Kong","email":"","orcid":"","institution":"The Fifth Affiliated Hospital of Wenzhou Medical University","correspondingAuthor":false,"prefix":"","firstName":"Chunli","middleName":"","lastName":"Kong","suffix":""},{"id":284712187,"identity":"06bfd268-ceae-45a1-bcbd-cfac5855f7ff","order_by":1,"name":"Guihan Lin","email":"","orcid":"","institution":"The Fifth Affiliated Hospital of Wenzhou Medical University","correspondingAuthor":false,"prefix":"","firstName":"Guihan","middleName":"","lastName":"Lin","suffix":""},{"id":284712188,"identity":"3fc77bda-2bcb-4850-8cef-a1ad215ea64b","order_by":2,"name":"Weiyue Chen","email":"","orcid":"","institution":"The Fifth Affiliated Hospital of Wenzhou Medical University","correspondingAuthor":false,"prefix":"","firstName":"Weiyue","middleName":"","lastName":"Chen","suffix":""},{"id":284712189,"identity":"4185f8db-e6a7-46b8-8d52-c4bf7c71c1f4","order_by":3,"name":"Xue Cheng","email":"","orcid":"","institution":"The Fifth Affiliated Hospital of Wenzhou Medical University","correspondingAuthor":false,"prefix":"","firstName":"Xue","middleName":"","lastName":"Cheng","suffix":""},{"id":284712190,"identity":"e26d09f6-241b-4fde-b1a5-912c09a91728","order_by":4,"name":"Shuang Liu","email":"","orcid":"","institution":"The Fifth Affiliated Hospital of Wenzhou Medical University","correspondingAuthor":false,"prefix":"","firstName":"Shuang","middleName":"","lastName":"Liu","suffix":""},{"id":284712191,"identity":"a85fa28b-4f01-4987-aa35-950a2a7a1324","order_by":5,"name":"Di Shen","email":"","orcid":"","institution":"The Fifth Affiliated Hospital of Wenzhou Medical University","correspondingAuthor":false,"prefix":"","firstName":"Di","middleName":"","lastName":"Shen","suffix":""},{"id":284712192,"identity":"edc36d5f-5258-464b-b2c4-f471325ce2fe","order_by":6,"name":"Jiayi Ding","email":"","orcid":"","institution":"The Fifth Affiliated Hospital of Wenzhou Medical University","correspondingAuthor":false,"prefix":"","firstName":"Jiayi","middleName":"","lastName":"Ding","suffix":""},{"id":284712193,"identity":"bacb3d22-23a6-4ed4-a167-6ab049c18704","order_by":7,"name":"Junguo Hui","email":"","orcid":"","institution":"The Fifth Affiliated Hospital of Wenzhou Medical University","correspondingAuthor":false,"prefix":"","firstName":"Junguo","middleName":"","lastName":"Hui","suffix":""},{"id":284712194,"identity":"6065530b-628e-4128-897c-94082da2509c","order_by":8,"name":"Minjiang Chen","email":"","orcid":"","institution":"The Fifth Affiliated Hospital of Wenzhou Medical University","correspondingAuthor":false,"prefix":"","firstName":"Minjiang","middleName":"","lastName":"Chen","suffix":""},{"id":284712196,"identity":"2db3631d-cd26-4e32-a95b-71acc6c7f2fd","order_by":9,"name":"Shuiwei Xia","email":"","orcid":"","institution":"The Fifth Affiliated Hospital of Wenzhou Medical University","correspondingAuthor":false,"prefix":"","firstName":"Shuiwei","middleName":"","lastName":"Xia","suffix":""},{"id":284712198,"identity":"5f82bbc1-1f17-48bc-90cf-f9e468e1dde0","order_by":10,"name":"Min Xu","email":"","orcid":"","institution":"The Fifth Affiliated Hospital of Wenzhou Medical University","correspondingAuthor":false,"prefix":"","firstName":"Min","middleName":"","lastName":"Xu","suffix":""},{"id":284712201,"identity":"d7a7e9cc-1820-45c2-92c6-af8cb642e361","order_by":11,"name":"Zhiyi Peng","email":"","orcid":"","institution":"First Affiliated Hospital Zhejiang University","correspondingAuthor":false,"prefix":"","firstName":"Zhiyi","middleName":"","lastName":"Peng","suffix":""},{"id":284712203,"identity":"6a5483ec-66dd-4e57-8e01-addb18115ddc","order_by":12,"name":"Jiansong Ji","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAzElEQVRIiWNgGAWjYBACxoYDYFoOwmUjTgtjA0MCgzHxWsC6gFoSG4jWwtx4/PmDnz8Op2+4dsaA4UPZYQb+2Q2ErDhj2NiTcDh3w+0cA8YZ5w4zSNw5QFALYzMDUMs2oBZm3rbDDAYSCYS0HH8I0pJuBtLylzgtBwxBWhLAWhiJ03LGcGZPWrrh/ttpBQd7zqXzSNwgoMVwxvEHH37YWMtLzk7e+OBHmbUc/wyCWg6AqGYwB8Tkwa8eCOT5G0BUHUGFo2AUjIJRMIIBAEpsS3+f/PXZAAAAAElFTkSuQmCC","orcid":"","institution":"The Fifth Affiliated Hospital of Wenzhou Medical University","correspondingAuthor":true,"prefix":"","firstName":"Jiansong","middleName":"","lastName":"Ji","suffix":""}],"badges":[],"createdAt":"2024-03-23 03:44:17","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-4152618/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-4152618/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":53754536,"identity":"4a3c9c18-77c8-44e7-90d4-e4abaee91a0d","added_by":"auto","created_at":"2024-03-29 18:58:13","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":1328944,"visible":true,"origin":"","legend":"\u003cp\u003eFlowchart of study enrollment.\u003c/p\u003e","description":"","filename":"1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4152618/v1/2a88ce57ff6d96b568f8e63a.jpg"},{"id":53754535,"identity":"d8eb9fc5-9775-417a-8fa0-343aa1779154","added_by":"auto","created_at":"2024-03-29 18:58:13","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":1259225,"visible":true,"origin":"","legend":"\u003cp\u003eWorkflow of this study. Firstly, segmentation of the tumors based on the DCE-MR images with 3D slicer software. Secondly, 851 features were extracted from manually drawn VOI on preoperative DCE-MRI maps by using the Pyradiomics package. Thirdly, VarianceThreshold, SelectKBest, and LASSO regression were applied for data dimension reduction, feature selection and model construction. Finally, evaluate the performance of the constructed model by ROC curves, calibration curves and decision curves. VOI, volume of interest; MRI, magnetic resonance imaging; LASSO, least absolute shrinkage and selection operator; AUC, area under the curve; ROC, receiver operating characteristic.\u003c/p\u003e","description":"","filename":"2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4152618/v1/ea38a0985b910915bb034688.jpg"},{"id":53756940,"identity":"4a507a3a-fe0a-4291-8bad-28a6ab1d698a","added_by":"auto","created_at":"2024-03-29 19:06:13","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":1435484,"visible":true,"origin":"","legend":"\u003cp\u003eThe diagnostic performance of the radiomics signature. \u003cstrong\u003e(A-C)\u003c/strong\u003e Violin plots illustrated the Rad-scores for HER2-positive and HER2-negative breast cancers in the training, internal test set and external test set. \u003cstrong\u003e(D-F) \u003c/strong\u003eROC curves of the radiomics signature in the training, internal test and external test sets, respectively. \u003cstrong\u003e(G-I) \u003c/strong\u003eCalibration curves of the radiomics signature in the training, internal test and external test sets, respectively. HER2, Human epidermal growth factor receptor 2; ROC, receiver operating characteristic.\u003c/p\u003e","description":"","filename":"3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4152618/v1/8c82a3c9d50a57432fc69fbf.jpg"},{"id":53756941,"identity":"89896daf-8f5e-4962-99dd-7e194762205a","added_by":"auto","created_at":"2024-03-29 19:06:13","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":1174181,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003e(A-C)\u003c/strong\u003e ROC curves of different models in the training, internal test set and external test sets, respectively. \u003cstrong\u003e(D-F)\u003c/strong\u003e Calibration curves between the combined model prediction and actual observation for HER2 status of patients in the training, internal test and external test sets, respectively. The diagonal dotted line represents an ideal evaluation, while the solid lines and dashed lines represent the performance of the corrected and apparent bias, respectively. The closer the fit is to the diagonal dotted line, the better the evaluation. \u003cstrong\u003e(G-I)\u003c/strong\u003e Confusion matrices of the combined models in the training, internal test, and external test sets, respectively. The color shade depends on the percentage within the square: the higher the percentage, the darker the shade. ROC, receiver operating characteristic.\u003c/p\u003e","description":"","filename":"4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4152618/v1/2dd32bf8e2f39fe62474a540.jpg"},{"id":53754541,"identity":"1875ce3d-9b59-46b3-b9a3-71cbbe4460fc","added_by":"auto","created_at":"2024-03-29 18:58:13","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":933736,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003e(A) \u003c/strong\u003eThe radiomics nomogram integrated the radiomics signature with the BI-RADS in the training set. The probability value of each breast cancer patient with HER2-positive is marked on each axis.\u003cstrong\u003e (B) \u003c/strong\u003eThe decision curve of the Radiomics model, clinical model and the combined model. The y-axis represents the standardized net benefit, and the x-axis represents the threshold probability. The red line, green line and the blue line represent the radiomics model, the clinical model and the combined model, respectively. The orange line represents the assumption that all patients are HER2-positive. The purple line represents the assumption that all patients are of HER2-negative. The decision curve reveals that the combined model adds more benefit than either single radiomics model or clinical model in predicting HER2 status when the threshold probability was ranges from 0.05 to 0.9.\u003c/p\u003e","description":"","filename":"5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4152618/v1/8db9dbfe49be9fd6892d96b7.jpg"},{"id":53754537,"identity":"148dbbb5-92ef-4c22-b460-434a3fc80218","added_by":"auto","created_at":"2024-03-29 18:58:13","extension":"jpg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":1085254,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003e(A) \u003c/strong\u003eVolcano plot shows the differentially expressed genes in the model predicted HER2-positive compared with the model predicted HER2-negative group. \u003cstrong\u003e(B) \u003c/strong\u003eKEGG enrichment analysis. GeneRatio means the ratio of genes in this pathway to all genes. Count means the number of genes in that pathway. P.adjust means the P values of KEGG analysis was adjusted by false discovery rate. \u003cstrong\u003e(C) \u003c/strong\u003eGO enrichment analysis. KEGG, Kyoto Encyclopedia of Genes and Genomes; GO, Gene ontology.\u003c/p\u003e","description":"","filename":"6.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4152618/v1/0acc407823fc44bbf1fcb401.jpg"},{"id":71226451,"identity":"5171f18f-b4e6-41cd-846b-162ebc5bbe7b","added_by":"auto","created_at":"2024-12-12 10:02:27","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":8028414,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4152618/v1/e5156882-0d3a-4925-ae77-3c37a653c55e.pdf"},{"id":53754540,"identity":"4f00674c-3ecd-4a0b-a0e4-14f83fda51b1","added_by":"auto","created_at":"2024-03-29 18:58:13","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":480100,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryMaterial.docx","url":"https://assets-eu.researchsquare.com/files/rs-4152618/v1/b14bd2952f7bce6a937b8457.docx"},{"id":53754542,"identity":"a712dd54-dafe-4906-aa54-7599af91ffcd","added_by":"auto","created_at":"2024-03-29 18:58:13","extension":"pdf","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":693545,"visible":true,"origin":"","legend":"","description":"","filename":"Fig.S1.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4152618/v1/5f7cf2397dc555421385b9b6.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"DCE-MRI Based Machine Learning Predictor for HER2-Positive Breast Cancer: A Feasibility and Validation Multicenter Study","fulltext":[{"header":"Background","content":"\u003cp\u003eBreast cancer is the most prevalent malignant cancer in women worldwide, with over 1.3\u0026nbsp;million cases diagnosed and 400,000 deaths each year [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Approximately 20%-30% of invasive breast cancers overexpress the human epidermal growth factor receptor 2 (HER2) tyrosine kinase receptor [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. Breast cancer patients with HER2-positive tumors are known to exhibit more aggressive characteristics, such as lower differentiation, high rates of cell proliferation, early systemic metastasis, and a higher risk of recurrence and mortality [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. Additionally, HER2-positive patients tend to be more resistant to chemotherapy, but they have been found to respond well to targeted antibody therapy [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. Clinical trials have confirmed that patients with HER2-positive treated with anti-HER2 therapy combined with adjuvant chemotherapy have achieved a prolonged time to disease progression, longer overall survival rate, and higher rates of pathological complete response rate compared to patients treated with chemotherapy alone [\u003cspan additionalcitationids=\"CR8\" citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. Therefore, it is crucial to accurately identify HER2 status before surgery in order to guide individual treatment plan and predict the prognosis of breast cancer.\u003c/p\u003e \u003cp\u003eIn clinical practice, it is mainly based on core needle biopsy-based immunohistochemical (IHC) or fluorescence in situ hybridization (FISH) assays to assess HER2 status of breast cancers [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. This approach relies on a single biopsy of a heterogeneous tumor, which may only capture a tiny portion of a potentially heterogeneous lesion, and it may not be representative of the whole tumor\u0026rsquo;s genetic, epigenetic, or phenotypic alterations [\u003cspan additionalcitationids=\"CR12\" citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. Meanwhile, the assessment is performed at specific time points, which may not reflect the dynamic changes in tumor biology over time and during treatment [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. Hence, it is necessary to develop a non-invasive method that can reflect the overall heterogeneity of tumors and be used for real-time detection of HER2 status for breast cancers.\u003c/p\u003e \u003cp\u003eMedical imaging offers a noninvasive approach that can detect a wider range of tumor heterogeneity [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. Radiomics, a rapidly emerging field, can transform medical images into high dimensional computer-based data. Multi-parametric magnetic resonance imaging (MRI), has become a routine examination in clinical practice for diagnosis, preoperative staging assessment, and treatment responsiveness assessment [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. In particular, dynamic contrast-enhanced (DCE) imaging is considered the most sensitivity modality for detecting breast cancer as it provide temporal information on the kinetics of the contrast agent in suspicious lesions and offers sufficient spatial resolution [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. Prior studies have explored the relationship between radiomics signature and tumor biology characteristics, but the generalizability of these findings is limited due to the use of different MRI protocols and scanners [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. Additionally, these studies had limitations in terms of the number of radiomics features studied and the lack of comparison with the predictive efficiency based on conventional MR features [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eTherefore, we aimed to develop an accurate and robust model for predicting HER2 status based DCE-MRI images. Furthermore, we identified differentially expressed genes related to HER2 status in the Cancer Imaging Archive (TCIA) dataset and conducted Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) functional enrichment analyses based on these differential genes to gain insights into the biological behaviors associated with HER2 status. The model developed in this study has the potential to assist in decision-making and patient management in clinical practice, thereby facilitating precision medicine.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003ePatients\u003c/h2\u003e \u003cp\u003e This study was reviewed and approved by the Institutional Review Boards and Human Ethics Committee of the Fifth Affiliated Hospital of Wenzhou Medical University (center A) and the First Affiliated Hospital of Zhejiang University (center B). The requirement for informed consent was waived in accordance with the Helsinki statement. Patients with pathologically confirmed breast cancer were consecutively enrolled from June 2018 to December 2022 at the two aforementioned centers. The inclusion and exclusion criteria were as follows: (1) The primary tumor was invasive ductal carcinoma; (2) HER2 status was determined using IHC and fluorescence in situ hybridization (FISH) in surgical specimens; (3) DCE-MRI examination was performed prior to surgery; (4) patients with lesion size larger than 1 cm to minimize the impact of partial volume on radiomics analysis; The exclusion criteria were as follows: (1) patients with insufficient clinicopathological data or DCE-MR scans; (2) patients who received neoadjuvant chemotherapy or any other treatment before DCE-MR imaging collection; (3) HER2 status was determined by biopsy before surgery; and (4) patients with a history of malignant tumors.\u003c/p\u003e \u003cp\u003eFinally, a total of 570 patients were enrolled in, and the screening process is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. The enrolled patients from center A were divided into a training set of 281 and an internal test set of 122 using a random number table method at a ratio of 7:3. The clinical information, including age, menopausal status, tumor size, breast imaging reporting and data system (BI-RADS), and time-signal intensity curve (TIC) in DCE-MRI scanning, were collected. Patients from center B were used as an independent external test set. Besides, 22 patients from TCIA were recruited as an outcome prediction set to explore the biologic functions.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003ePathological evaluation\u003c/h2\u003e \u003cp\u003e The HER2 status was determined according to the American Society of Clinical Oncology (ASCO)/College of American Pathologists (CAP) detection guidelines. HER2 status was divided into HER2-positive and HER2-negative [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. The HER2 status was determined by HER2 IHC, and FISH was used if necessary. The IHC score were divided into 0+, 1+, 2\u0026thinsp;+\u0026thinsp;and 3\u0026thinsp;+\u0026thinsp;based on the staining status of the membrane and the proportion of the invasive tumor cells. IHC score of 0/1\u0026thinsp;+\u0026thinsp;were considered HER2-negative, and 3\u0026thinsp;+\u0026thinsp;was defined as HER2-positive, while the 2\u0026thinsp;+\u0026thinsp;required further confirmation of amplification status by FISH.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003eMR image acquisition\u003c/h2\u003e \u003cp\u003eFor patients from center A, all breast MR examinations were performed on a 1.5 Tesla MR system (Siemens Healthcare, Erlangen, Germany) utilizing a dedicated eight-channel breast radiofrequency coil with patient placed in the prone position. DCE-MRI was performed using flash low angle shot 3D sequence. The specific scanning parameters are as follows: TR\u0026thinsp;=\u0026thinsp;4.4 ms, TE\u0026thinsp;=\u0026thinsp;1.5 ms, flip angle\u0026thinsp;=\u0026thinsp;12\u0026deg;, slice thickness\u0026thinsp;=\u0026thinsp;1.5 mm, Matrix\u0026thinsp;=\u0026thinsp;224\u0026times;192, FOV\u0026thinsp;=\u0026thinsp;350 mm\u0026times;350 mm. After the first dynamic scan was completed, a bolus of 0.2 mmol/kg gadopentetate dimeglumine (Magnevist; Bayer HealthCare, Hanover, NJ, USA) was injected into the cubital vein with a high-pressure syringe at a rate of 2.0 ml/s. Continuous non-interval scans were done in seven stages after intravenous injection, with a scan period of 60 seconds for each phase.\u003c/p\u003e \u003cp\u003eFor patients from center B, all breast MR examinations were performed on a 3.0 Tesla MR system (Signa HDxt GE healthcare, America) using an eight-channel dedicated surface coil. Patients were positioned in the prone position, with bilateral breasts naturally overhanging in the coil. Gd-DTPA was used as a contrast agent. A dose of 0.1 mmol/kg Gd-DTPA was injected into the elbow vein at a flow rate of 2.0 ml/s, followed by the injection of 15 ml of normal saline at the same flow rate. DCE scans were performed using a cross-sectional three-dimensional fast low angle shot (3D FLASH) technique. The specific scanning parameters are as follows: TR\u0026thinsp;=\u0026thinsp;4.51 ms, TE\u0026thinsp;=\u0026thinsp;1.61 ms, flip angle\u0026thinsp;=\u0026thinsp;10\u0026deg;, slice thickness\u0026thinsp;=\u0026thinsp;1.0 mm, Matrix\u0026thinsp;=\u0026thinsp;420 \u0026times; 420, FOV\u0026thinsp;=\u0026thinsp;340 mm \u0026times; 340 mm. The scans were initiated 25 seconds after the start of the contrast agent injection and were repeated for 6 consecutive scans. Each scan had a duration of 58 seconds.\u003c/p\u003e \u003cp\u003eIn this study, the second phase axial images of DCE-MRI were chosen for further analysis since it showed the lesion borders more clearly.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003eImage segmentation and features extraction\u003c/h2\u003e \u003cp\u003eAll MRI images were imported into the open-source image processing software 3D-Slicer (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.slicer.org/\u003c/span\u003e\u003cspan address=\"https://www.slicer.org/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003e).\u003c/span\u003e The segmentation of the three-dimensional tumor area is conducted by reader A (JGH, with 10 years of experience) layer by layer along the tumor edge using the 3D-Slicer platform. Then, reader B (XC, with 12 years of experience) verified the segmentation results, and any discrepancies were resolved through discussion. The radiologists are blinded to the pathologic results. Image preprocessing and features extraction is conducted with Pyradiomics package (version 2.12; \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://pyradiomics.readthedocs.io/en/2.1.2/\u003c/span\u003e\u003cspan address=\"https://pyradiomics.readthedocs.io/en/2.1.2/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003e)\u003c/span\u003e in python 3.7.0 [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. All images were resampled to a voxel size of 1\u0026times;1\u0026times;1 mm\u003csup\u003e3\u003c/sup\u003e with a fixed bin width of 25 to standardize the voxel spacing and decrease the image noise. A total of 851 radiomics features were extracted for each volume of interest (VOI). In order to secure the reproducibility of the extracted features, 60% images were randomly selected and were delineated by the same radiologist (reader A) one month later to calculate the intra-class correlation coefficient (ICC). Then the selected 60% images were also segmented independently by the second radiologist (reader B) to calculate the inter-class ICC. Features with ICCs greater than 0.75 is considered to be good reproducibility and will be remained for further analysis.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003eIntra-class and Inter-class reproducibility of radiomics features\u003c/h2\u003e \u003cp\u003eFor the radiomics features extraction, there was no significant difference between the two radiologists, with p value ranging from 0.769\u0026thinsp;~\u0026thinsp;0.872. The intra-class ICC calculated with twice segmentation of reader A ranged from 0.858 to 0.957, and the inter-class ICC calculated by the features extracted by reader A and reader B ranged from 0.769 to 0.865. The reproducibility of the radiomics features is relatively well and all the 851 features extracted from the DCE-MRI sequence of each patient were remained for further analysis.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eFeature Selection and Model Construction\u003c/h2\u003e \u003cp\u003eBefore performing feature dimensionality reduction and feature selection, we standardized the features with ICCs greater than 0.75 using the Z-score method to reduce the influence of the different units imposed by the units of each feature and improve the performance of the model. The Z-score method is defined as follows: z=(x-\u0026micro;)/σ, where x is the value of current parameter, \u0026micro; is the mean value of x, and σ is the corresponding deviation. Next, a three-step procedure was conducted for dimensionality reduction and selection of task-specific radiomics features in the training cohort. Firstly, the variance threshold method was applied to preliminarily select features with a variance greater than 0.8. Secondly, SelectKBest method was utilized to calculate the P-value for each feature, which indicates the correlation between the feature and the label (HER2 status). Only features with a P-value less than 0.01 were retained for further analysis. Thirdly, least absolute shrinkage and selection operator logistic regression (LASSO) with five-fold cross-validation were applied to screen the features that were most relevant to HER2 status of breast cancers. For LASSO regression, the regularization parameter λ controls the number of radiomics features and affects the performance of the model. The λ was optimized by five-fold cross-validation in the training cohort to minimize the mean square error. Finally, the radiomics signature for each patient was calculated using logistic regression.\u003c/p\u003e \u003cp\u003eTo further improve the performance of the prediction model, we also included clinical indicators and imaging examination indicators for analysis. Univariate and multivariate analyses were conducted to screen the predictive variables for HER2 status, which were then integrated with the radiomics signature to establish the predictive nomogram. The performance of the nomogram was evaluated using receiver operating characteristic curve (ROC), calibration curve and decision curve.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003eOutcome Prediction and Exploring Biologic Functions\u003c/h2\u003e \u003cp\u003ePatients in the TCIA dataset were also enrolled in this study to analyze the biologic functions of breast cancers. We used the HER2 status prediction model built in this study to predict HER2 status for samples in the TCIA dataset. The \u0026ldquo;limma\u0026rdquo; package was applied to analyze the differentially expressed genes (DEGs) between the model predicted HER2 positive and HER2 negative groups, and a threshold P\u0026thinsp;\u0026lt;\u0026thinsp;0.01 was set. Based on the DEGs, GO and KEGG functional enrichment analyses were performed with thresholds of P\u0026thinsp;\u0026lt;\u0026thinsp;0.05.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003eStatistical Analysis\u003c/h2\u003e \u003cp\u003eIBM SPSS (Chicago, IL), R software (version 3.6.3) and Python (version 3.7.0) were used for statistical analysis in this study. The continuous variables with a normal distribution were presented as mean and standard deviation, and the continuous variables with non-normal distribution were presented as median or interquartile range. The normality test of the continuous data distribution was performed by Kolmogorov-Smirnov test. Then the t test was applied to analyze the continuous variables with normal distribution, and the Mann-Whitney U test was applied to analyze the continuous variables with abnormal distribution. The categorical variables were compared with chi-square test or Fisher exact test and presented as percentages. A P value less than 0.05 was deemed to be statistically significant in all statistical analysis.\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cdiv\u003e\n\u003ch2\u003eBaseline Characteristics\u003c/h2\u003e\n\u003cp\u003eBaseline characteristics of the patients in the training and internal test set are summarized in \u003cstrong\u003eSupplementary Table S1\u003c/strong\u003e. There is no statistical significance in age, menopausal status, tumor size, BI-RADS and TIC between the training and internal test set. All characteristics are comparable between two cohorts.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv\u003e\n\u003ch2\u003eClinical and Imaging Examination Indicators for HER2 Status\u003c/h2\u003e\n\u003cp\u003eClinical and imaging examination characteristics of patients with HER2-positive and HER2-negative in training and internal test set are summarized in Table\u0026nbsp;1. 281 cases and 122 cases were assigned to the training and internal test set, respectively. Univariate and multivariate logistic regression analyses were performed to identify indicators associated with HER2 status, and corresponding results are summarized in Table\u0026nbsp;2. As is shown, BI-RADS was screened out as the predictive clinical indicators that was valuable in predicting HER2 status of breast cancer.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 1\u003c/strong\u003e Clinical characteristics of patients with HER2-positive and HER2-negative in the training and internal test sets\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd rowspan=\"2\" width=\"16.867469879518072%\"\u003e\n\u003cp\u003e\u003cstrong\u003eCharacteristic\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" width=\"29.27710843373494%\"\u003e\n\u003cp\u003e\u003cstrong\u003e\u0026nbsp;training set\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd rowspan=\"2\" width=\"8.433734939759036%\"\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eP\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003cstrong\u003eValue\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" width=\"28.91566265060241%\"\u003e\n\u003cp\u003e\u003cstrong\u003einternal test set\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd rowspan=\"2\" width=\"8.072289156626505%\"\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eP\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003cstrong\u003eValue\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"22.06148282097649%\"\u003e\n\u003cp\u003e\u003cstrong\u003eHER2-positive (n=108)\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"21.880650994575046%\"\u003e\n\u003cp\u003e\u003cstrong\u003eHER2-negative (n=173)\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"21.15732368896926%\"\u003e\n\u003cp\u003e\u003cstrong\u003eHER2-positive (n=47)\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"22.24231464737794%\"\u003e\n\u003cp\u003e\u003cstrong\u003eHER2-negative (n=75)\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"16.867469879518072%\"\u003e\n\u003cp\u003eAge (years)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.698795180722891%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.578313253012048%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"8.433734939759036%\"\u003e\n\u003cp\u003e0.218\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.096385542168674%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.819277108433734%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"8.072289156626505%\"\u003e\n\u003cp\u003e0.713\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"16.867469879518072%\"\u003e\n\u003cp\u003e＞55\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.698795180722891%\"\u003e\n\u003cp\u003e34 (31.5%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.578313253012048%\"\u003e\n\u003cp\u003e67 (38.7%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"8.433734939759036%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.096385542168674%\"\u003e\n\u003cp\u003e16 (41.1%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.819277108433734%\"\u003e\n\u003cp\u003e28 (37.3%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"8.072289156626505%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"16.867469879518072%\"\u003e\n\u003cp\u003e\u0026le;55\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.698795180722891%\"\u003e\n\u003cp\u003e74 (68.5%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.578313253012048%\"\u003e\n\u003cp\u003e106 (61.3%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"8.433734939759036%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.096385542168674%\"\u003e\n\u003cp\u003e31 (65.9%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.819277108433734%\"\u003e\n\u003cp\u003e47 (62.7%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"8.072289156626505%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"16.867469879518072%\"\u003e\n\u003cp\u003eMenopausal status\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.698795180722891%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.578313253012048%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"8.433734939759036%\"\u003e\n\u003cp\u003e0.190\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.096385542168674%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.819277108433734%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"8.072289156626505%\"\u003e\n\u003cp\u003e0.132\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"16.867469879518072%\"\u003e\n\u003cp\u003ePremenopausal\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.698795180722891%\"\u003e\n\u003cp\u003e50 (53.7%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.578313253012048%\"\u003e\n\u003cp\u003e79 (45.7%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"8.433734939759036%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.096385542168674%\"\u003e\n\u003cp\u003e26 (55.4%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.819277108433734%\"\u003e\n\u003cp\u003e31 (41.4%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"8.072289156626505%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"16.867469879518072%\"\u003e\n\u003cp\u003ePostmenopausal\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.698795180722891%\"\u003e\n\u003cp\u003e58 (46.3%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.578313253012048%\"\u003e\n\u003cp\u003e94 (54.3%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"8.433734939759036%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.096385542168674%\"\u003e\n\u003cp\u003e21 (44.6%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.819277108433734%\"\u003e\n\u003cp\u003e44 (58.6%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"8.072289156626505%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"16.867469879518072%\"\u003e\n\u003cp\u003eTumor size (cm)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.698795180722891%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.578313253012048%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"8.433734939759036%\"\u003e\n\u003cp\u003e0.686\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.096385542168674%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.819277108433734%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"8.072289156626505%\"\u003e\n\u003cp\u003e0.236\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"16.867469879518072%\"\u003e\n\u003cp\u003e＞2\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.698795180722891%\"\u003e\n\u003cp\u003e38 (35.1%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.578313253012048%\"\u003e\n\u003cp\u003e65 (37.5%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"8.433734939759036%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.096385542168674%\"\u003e\n\u003cp\u003e22 (46.8%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.819277108433734%\"\u003e\n\u003cp\u003e27 (36.0%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"8.072289156626505%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"16.867469879518072%\"\u003e\n\u003cp\u003e\u0026le;2\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.698795180722891%\"\u003e\n\u003cp\u003e70 (64.9%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.578313253012048%\"\u003e\n\u003cp\u003e108 (62.5%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"8.433734939759036%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.096385542168674%\"\u003e\n\u003cp\u003e25 (53.2%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.819277108433734%\"\u003e\n\u003cp\u003e48 (64.0%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"8.072289156626505%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"16.867469879518072%\"\u003e\n\u003cp\u003eBI-RADS\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.698795180722891%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.578313253012048%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"8.433734939759036%\"\u003e\n\u003cp\u003e\u0026lt; 0.001\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.096385542168674%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.819277108433734%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"8.072289156626505%\"\u003e\n\u003cp\u003e0.002\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"16.867469879518072%\"\u003e\n\u003cp\u003e4A\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.698795180722891%\"\u003e\n\u003cp\u003e14 (12.9%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.578313253012048%\"\u003e\n\u003cp\u003e47 (27.2%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"8.433734939759036%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.096385542168674%\"\u003e\n\u003cp\u003e4 (8.5%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.819277108433734%\"\u003e\n\u003cp\u003e17 (22.7%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"8.072289156626505%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"16.867469879518072%\"\u003e\n\u003cp\u003e4B\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.698795180722891%\"\u003e\n\u003cp\u003e29 (26.9%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.578313253012048%\"\u003e\n\u003cp\u003e84 (48.5%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"8.433734939759036%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.096385542168674%\"\u003e\n\u003cp\u003e13 (27.6%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.819277108433734%\"\u003e\n\u003cp\u003e35 (46.7%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"8.072289156626505%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"16.867469879518072%\"\u003e\n\u003cp\u003e4C\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.698795180722891%\"\u003e\n\u003cp\u003e42 (38.9%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.578313253012048%\"\u003e\n\u003cp\u003e27 (15.6%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"8.433734939759036%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.096385542168674%\"\u003e\n\u003cp\u003e21 (44.7%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.819277108433734%\"\u003e\n\u003cp\u003e13 (17.2%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"8.072289156626505%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"16.867469879518072%\"\u003e\n\u003cp\u003e5\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.698795180722891%\"\u003e\n\u003cp\u003e23 (21.3%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.578313253012048%\"\u003e\n\u003cp\u003e15 (8.7%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"8.433734939759036%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.096385542168674%\"\u003e\n\u003cp\u003e9 (19.2%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.819277108433734%\"\u003e\n\u003cp\u003e10 (13.4%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"8.072289156626505%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"16.867469879518072%\"\u003e\n\u003cp\u003eTIC\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.698795180722891%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.578313253012048%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"8.433734939759036%\"\u003e\n\u003cp\u003e0.255\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.096385542168674%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.819277108433734%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"8.072289156626505%\"\u003e\n\u003cp\u003e0.691\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"16.867469879518072%\"\u003e\n\u003cp\u003eWash-in\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.698795180722891%\"\u003e\n\u003cp\u003e35 (32.4%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.578313253012048%\"\u003e\n\u003cp\u003e44 (25.4%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"8.433734939759036%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.096385542168674%\"\u003e\n\u003cp\u003e11 (23.4%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.819277108433734%\"\u003e\n\u003cp\u003e22 (29.3%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"8.072289156626505%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"16.867469879518072%\"\u003e\n\u003cp\u003ePlatform\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.698795180722891%\"\u003e\n\u003cp\u003e56 (51.9%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.578313253012048%\"\u003e\n\u003cp\u003e107 (61.8%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"8.433734939759036%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.096385542168674%\"\u003e\n\u003cp\u003e27 (57.4%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.819277108433734%\"\u003e\n\u003cp\u003e42 (56.1%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"8.072289156626505%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"16.867469879518072%\"\u003e\n\u003cp\u003eWash-out\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.698795180722891%\"\u003e\n\u003cp\u003e17 (15.7%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.578313253012048%\"\u003e\n\u003cp\u003e22 (12.8%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"8.433734939759036%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.096385542168674%\"\u003e\n\u003cp\u003e9 (19.2%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"14.819277108433734%\"\u003e\n\u003cp\u003e11 (14.6%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"8.072289156626505%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003c/table\u003e\n\u003c/div\u003e\n\u003cdiv\u003e\n\u003cp\u003eAbbreviations: BI-RADS, breast imaging reporting and data system; TIC, time signal intensity curve.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 2\u003c/strong\u003e Univariate and Multivariate analysis of risk factors for HER2-positive breast cancer in the training set\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd rowspan=\"2\" width=\"23.46938775510204%\"\u003e\n\u003cp\u003e\u003cstrong\u003eCharacteristic\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" width=\"38.775510204081634%\"\u003e\n\u003cp\u003e\u003cstrong\u003e\u0026nbsp;Univariate analysis\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" width=\"37.755102040816325%\"\u003e\n\u003cp\u003e\u003cstrong\u003e\u0026nbsp;Multivariate analysis\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"33.78378378378378%\"\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eOR (95% CI)\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"17.56756756756757%\"\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eP\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003cstrong\u003eValue\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"32.432432432432435%\"\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eOR (95% CI)\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"16.216216216216218%\"\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eP\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003cstrong\u003eValue\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"23.711340206185568%\"\u003e\n\u003cp\u003eAge (years)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"25.77319587628866%\"\u003e\n\u003cp\u003e0.727 (0.437-1.209)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"13.402061855670103%\"\u003e\n\u003cp\u003e0.219\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"24.742268041237114%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"12.371134020618557%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"23.711340206185568%\"\u003e\n\u003cp\u003eMenopausal Status\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"25.77319587628866%\"\u003e\n\u003cp\u003e0.725 (0.447-1.173)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"13.402061855670103%\"\u003e\n\u003cp\u003e0.190\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"24.742268041237114%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"12.371134020618557%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"23.711340206185568%\"\u003e\n\u003cp\u003eTumor size (cm)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"25.77319587628866%\"\u003e\n\u003cp\u003e0.902 (0.547-1.488)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"13.402061855670103%\"\u003e\n\u003cp\u003e0.686\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"24.742268041237114%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"12.371134020618557%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"23.711340206185568%\"\u003e\n\u003cp\u003eBI-RADS\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"25.77319587628866%\"\u003e\n\u003cp\u003e2.070 (1.568-2.732)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"13.402061855670103%\"\u003e\n\u003cp\u003e\u0026lt; 0.001\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"24.742268041237114%\"\u003e\n\u003cp\u003e2.070(1.568-0.732)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"12.371134020618557%\"\u003e\n\u003cp\u003e\u0026lt; 0.001\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"23.711340206185568%\"\u003e\n\u003cp\u003eTIC\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"25.77319587628866%\"\u003e\n\u003cp\u003e0.906 (0.619-1.326)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"13.402061855670103%\"\u003e\n\u003cp\u003e0.611\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"24.742268041237114%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"12.371134020618557%\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eAbbreviations: BI-RADS, breast imaging reporting and data system; TIC, time signal intensity curve.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConstruction and Validation of the HER2 Status Predictive Model\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe flowchart of this study is illustrated in \u003cstrong\u003eFig. 2\u003c/strong\u003e. For each patient, a total of 851 radiomics features were extracted. It is important to note that all 851 features exhibited good reproducibility, with ICCs exceeding 0.75. Thus, all 851 features were utilized to calculate the variance by using the variance threshold method. After that, 851 features were retained as they had a variance greater than 0.8. Next, the SelectKBest method was utilized to calculate the P value for each feature. This step resulted in the retention of 280 features for LASSO regression. LASSO regression process is illustrated in \u003cstrong\u003eFig. S1A-B\u003c/strong\u003e. Finally, 23 key radiomics features were identified (\u003cstrong\u003eAppendix E1.1,\u003c/strong\u003e\u003cstrong\u003eFig. S1C\u003c/strong\u003e), and the Rad-score for each patient was calculated based these features by using the logistic regression.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePerformance of the Radiomics Signature\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eRad-score can be used to distinguish the HER2-positive patients from the HER2-negatives for there is a significance between the two groups. The median value of Rad-score in HER2-positive patients is higher than that in the HER2-negative patients in both the training, internal and external test sets (\u003cstrong\u003eFig. 3A-C\u003c/strong\u003e). In order to analyze the performance of the diagnostic signature in detail, we also depicted the ROC curves in different sets (\u003cstrong\u003eFig. 3D-F\u003c/strong\u003e). In the training set, it achieved an AUC of 0.825, with an accuracy of 79.9%, a specificity of 85.5%, and a sensitivity of 66.7%. In the internal test set, it achieved an AUC of 0.818, with an accuracy of 81.1%, a specificity of 88.0%, and a sensitivity of 70.2%. In the external test set, it achieved an AUC of 0.748, with an accuracy of 81.4%, a specificity of 91.0%, and a sensitivity of 62.2%. The calibration curves for the diagnostic signature in training, internal and external test set are shown in \u003cstrong\u003eFig. 3G-I\u003c/strong\u003e.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDevelopment and Validation of Individualized Prediction Model for HER2 Status\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo further improve the performance of the predictive model, the imaging examination indicator BI-RADS was incorporated with the Rad-score to construct a more efficiency prediction model. As shown in \u003cstrong\u003eFig. 4A\u003c/strong\u003e, the combined model achieved an AUC of 0.881 in the training set, with an accuracy of 79.4%, a specificity of 92.5%, and a sensitivity of 73.1%. It yields an AUC of 0.883 in internal test set, with an accuracy of 80.3%, a specificity of 82.7%, and a sensitivity of 83.0% (\u003cstrong\u003eFig. 4B\u003c/strong\u003e). It yields an AUC of 0.798 in external test set, with an accuracy of 80.2%, a specificity of 84.4%, and a sensitivity of 71.1% (\u003cstrong\u003eFig. 4C\u003c/strong\u003e). To evaluate the fitting performance between the model predicted and the actual HER2 status, we developed calibration curves of the training, internal test and external test sets (\u003cstrong\u003eFig. 4D-F\u003c/strong\u003e). The calibration curves indicated that the model achieved well agreement between the predicted probabilities and the actual outcomes, which suggested a favorable fitting performance. Based on the reference standard of pathology results, the percentages of true positive, false positive, true negative, and false negative results were calculated and displayed as confusion matrix diagrams (\u003cstrong\u003eFig. 4G-I\u003c/strong\u003e). Predictive performance of the clinical model, radiomics model and combined model in the training, internal test and external test sets are summarized in \u003cstrong\u003eTable 3\u003c/strong\u003e.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 3\u003c/strong\u003e Predictive performance of the clinical model, radiomics model and combined model in the training, internal test and external test sets.\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd width=\"7.936507936507937%\"\u003e\n\u003cp\u003eModel\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"4\" width=\"30.793650793650794%\"\u003e\n\u003cp\u003eTraining Set\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"4\" width=\"30.687830687830687%\"\u003e\n\u003cp\u003eInternal Test Set\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"4\" width=\"30.582010582010582%\"\u003e\n\u003cp\u003eExternal Test Set\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"7.944915254237288%\"\u003e\n\u003cp\u003eModel Parameter\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.521186440677966%\"\u003e\n\u003cp\u003eAUC\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.521186440677966%\"\u003e\n\u003cp\u003eACC\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.944915254237288%\"\u003e\n\u003cp\u003eSpecificity\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.838983050847458%\"\u003e\n\u003cp\u003eSensitivity\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.415254237288136%\"\u003e\n\u003cp\u003eAUC\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.415254237288136%\"\u003e\n\u003cp\u003eACC\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.944915254237288%\"\u003e\n\u003cp\u003eSpecificity\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.838983050847458%\"\u003e\n\u003cp\u003eSensitivity\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.415254237288136%\"\u003e\n\u003cp\u003eAUC\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.415254237288136%\"\u003e\n\u003cp\u003eACC\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.944915254237288%\"\u003e\n\u003cp\u003eSpecificity\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.838983050847458%\"\u003e\n\u003cp\u003eSensitivity\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"7.944915254237288%\"\u003e\n\u003cp\u003eClinical Model\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.521186440677966%\"\u003e\n\u003cp\u003e0.710\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.521186440677966%\"\u003e\n\u003cp\u003e0.712\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.944915254237288%\"\u003e\n\u003cp\u003e0.757\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.838983050847458%\"\u003e\n\u003cp\u003e0.639\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.415254237288136%\"\u003e\n\u003cp\u003e0.664\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.415254237288136%\"\u003e\n\u003cp\u003e0.672\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.944915254237288%\"\u003e\n\u003cp\u003e0.693\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.838983050847458%\"\u003e\n\u003cp\u003e0.638\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.415254237288136%\"\u003e\n\u003cp\u003e0.647\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.415254237288136%\"\u003e\n\u003cp\u003e0.515\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.944915254237288%\"\u003e\n\u003cp\u003e0.361\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.838983050847458%\"\u003e\n\u003cp\u003e0.933\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"7.944915254237288%\"\u003e\n\u003cp\u003eRadiomics Model\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.521186440677966%\"\u003e\n\u003cp\u003e0.825\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.521186440677966%\"\u003e\n\u003cp\u003e0.799\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.944915254237288%\"\u003e\n\u003cp\u003e0.855\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.838983050847458%\"\u003e\n\u003cp\u003e0.667\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.415254237288136%\"\u003e\n\u003cp\u003e0.818\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.415254237288136%\"\u003e\n\u003cp\u003e0.811\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.944915254237288%\"\u003e\n\u003cp\u003e0.880\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.838983050847458%\"\u003e\n\u003cp\u003e0.702\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.415254237288136%\"\u003e\n\u003cp\u003e0.748\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.415254237288136%\"\u003e\n\u003cp\u003e0.832\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.944915254237288%\"\u003e\n\u003cp\u003e0.910\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.838983050847458%\"\u003e\n\u003cp\u003e0.622\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"7.944915254237288%\"\u003e\n\u003cp\u003eCombined Model\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.521186440677966%\"\u003e\n\u003cp\u003e0.881\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.521186440677966%\"\u003e\n\u003cp\u003e0.794\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.944915254237288%\"\u003e\n\u003cp\u003e0.925\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.838983050847458%\"\u003e\n\u003cp\u003e0.731\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.415254237288136%\"\u003e\n\u003cp\u003e0.883\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.415254237288136%\"\u003e\n\u003cp\u003e0.803\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.944915254237288%\"\u003e\n\u003cp\u003e0.827\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.838983050847458%\"\u003e\n\u003cp\u003e0.830\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.415254237288136%\"\u003e\n\u003cp\u003e0.798\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.415254237288136%\"\u003e\n\u003cp\u003e0.629\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.944915254237288%\"\u003e\n\u003cp\u003e0.844\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"7.838983050847458%\"\u003e\n\u003cp\u003e0.711\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eAbbreviations: AUC, Area under the curve; ACC, Accuracy.\u003c/p\u003e\n\u003cp\u003eFinally, we visualized the combined model as a nomogram to individually predict the HER2 status for each patient (\u003cstrong\u003eFig. 5A\u003c/strong\u003e). In addition, the decision curve of the training set was also plotted to verify the clinical utility of the predictive signature and the combined model. As illustrated in \u003cstrong\u003eFig. 5B\u003c/strong\u003e, the integration of the combined model has the potential to greatly enhance the clinical management of breast cancer, maximizing its benefits in practice.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eClinical Outcomes and Biologic Functions Associated with HER2 Status\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e22 patients from TCIA database were enrolled in this study. The exclusion criteria for TCIA database were as follow: (1) Missing of DCE-MR images; (2) poor registration quality; and (3) patients without mRNA data. Finally, a total of 22 patients with both available DCE-MR images and RNA-sequence data were included.\u003c/p\u003e\n\u003cp\u003eThe DCE-MR images of the enrolled 22 patients were downloaded from the TCIA dataset (https://www.cancerimagingarchive.net/). Tumor segmentation and features extraction were performed by the same radiologists in this study. Using the HER2 status prediction model developed in this study, we applied it to predict the HER2 status on the 22 samples that were enrolled in the study. Among the 22 samples, 3 samples were predicted to be HER2 positive, and the remaining 19 patients were predicted to be HER2 negative.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAnalysis of the Biologic Function\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eRNA was extracted from both the tumor specimens. All cells from the tumor specimen were utilized to identify immune cells based on bulk RNA-seq analysis. Gene expression levels were calculated using the FPKM format.\u003c/p\u003e\n\u003cp\u003eA total of 117 HER2 status related genes were identified (\u003cstrong\u003eFig. 6A\u003c/strong\u003e). KEGG enrichment analysis revealed that differentially expressed genes (DEGs) were mainly enriched in endocrine resistance, PI3K-AKT signaling pathway, and p53 (\u003cstrong\u003eFig. 6B\u003c/strong\u003e). GO enrichment analysis showed that DEGs were mainly involved in signaling pathways such as mammary gland epithelium development, epithelial cell proliferation, and regulation of epithelial cell proliferation (\u003cstrong\u003eFig. 6C\u003c/strong\u003e).\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eHER2 is a pivotal molecular target in the treatment of breast cancer. In present study, we investigated the potential association between HER2 status, quantitative imaging features extracted from DCE-MRI, and the MR reported BI-RADS score. The model could accurately predict HER2 status preoperatively. This predictive model performed exceptionally well in three independent cohorts. Enrichment pathway analysis was conducted on the HER2-positive and HER2-negative groups, which were classified by the model constructed in this study. Pathway enrichment analysis provides valuable biological insights into the prediction model developed in this study.\u003c/p\u003e\n\u003cp\u003eIn recent years, numerous researchers have explored the correlation between traditional imaging features and HER2 status. Previous studies have confirmed that conventional imaging features, such as masses with indistinct margins, calcifications within the masses, segmental calcifications, and microcalcifications are significantly associated with HER2-positive status [24-26]. However, these studies only explored the qualitative relationship between traditional imaging features and HER2 status, but didn\u0026rsquo;t explore the quantitative performance in identifying HER2 status for further step. In contrast, our study also included imaging examination and clinical characteristics for analysis to screen the indicators that were valuable in predicting HER2 status, and quantitatively assessed the performance of the screened indicators for predicting HER2 status. In our study, only BI-RADS showed statistically significant differences between HER2-positive and negative individuals. We integrated BI-RADS with Rad-score to develop a combined model. The model achieved improved AUCs in the training, internal test and external test set, respectively. These results suggest that our model is robust in predicting HER2 status of breast cancer.\u003c/p\u003e\n\u003cp\u003eIn recent years, the texture analysis-based radiomics has gained popularity in cancer research. Numerous studies have demonstrated that radiomics has the advantages of reflecting tumor heterogeneity, and there is an intrinsic correlation between tumor genotypes and radiomics features [27, 28]. Zhou et al [29] constructed a predictive model for breast cancer HER2 status based on T2-weighted sequences and DCE-T1 axial sequences. Their model achieved AUCs of 0.86 and 0.81 in the training and validation set, respectively. However, joint modeling based on multiple sequences is time-consuming in clinical practice. Therefore, in our study, we focused on extracting radiomics features and building a predictive model based solely on the DCE-MRI sequence. However, our predictive model outperformed Zhou\u0026apos;s model, which may be attributed to our larger sample size, more scientific study design, and more efficient algorithm. Another study attempted to predict HER2 status based on MRI images, but their model only yielded an AUC of 0.65 [30]. This may be due to the fact that they only extracted 38 radiomics features from the DCE-MRI images and had a small sample size of only 91 samples. There have been studies attempting to predict HER2 status using mammography or CT imaging features, but their accuracy, as measured by the AUC, is significantly lower than that in our study [31, 32]. Additionally, mammography and CT scans involve small doses of radiation and have lower soft tissue resolution compared to MRI. This is why the current study relies on dynamic DCE-MRI sequences to predict HER2 status.\u003c/p\u003e\n\u003cp\u003eOne major drawback of radiomics models based on machine learning algorithm is their lack of biological interpretability, which hinders its widespread adoption and application in clinical practice. Further research is needed to elucidate the underlying mechanisms that enable clinicians to interpret the biological significance of radiomic features and the predictive power of these models. Therefore, we utilized the model to predict the HER2 status for samples in the TCIA dataset. Differential gene analysis was then conducted based on the HER2-positive and HER2-negative samples classified by the model. Enrichment pathway analysis was performed based on the differentially expressed genes, providing biological interpretability to the radiomics model proposed in this study. The KEGG analysis revealed that differentially expressed genes associated with HER2 status were commonly involved in major signaling pathways such as PI3K-AKT signaling pathway, endocrine resistance, and p53. The PI3K-AKT signaling pathway has been extensively studied and is known to play a crucial role in various cellular processes, including cell proliferation, survival, invasion, migration, apoptosis, glucose metabolism, and DNA repair [33, 34]. A study reported that the PI3K-AKT pathway is activated in approximately half of hormone receptor-positive and HER2-negative breast cancers by means of activating mutations in PIK3CA and AKT1 and inactivating alterations in PTEN [35]. These patients are prone to endocrine therapy resistance. This preliminary research provides biological significance to the radiomics model proposed in this study.\u003c/p\u003e\n\u003cp\u003eHowever, there are also some limitations to this study. Firstly, it was a retrospective study, which may introduce potential selection bias. Secondly, manual segmentation of tumors was time-consuming and could be replaced by artificial intelligence to improve reproducibility. \u003c/p\u003e"},{"header":"Conclusions","content":"\u003cp\u003eThis study presents a radiomics signature and BI-RADS-based nomogram for predicting HER2 status in breast cancer patients before surgery. The proposed model was validated using enrichment pathway analysis on two groups, providing biological insights into the developed model. The model is user-friendly, accurate, and non-invasive, which can aid clinicians in identifying suitable candidates for anti-HER2 treatment in clinical practice. Furthermore, this study introduces novel approaches for more precise and personalized management of breast cancer.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cp\u003eMRI Magnetic resonance imaging\u003c/p\u003e\n\u003cp\u003eDCE-MRI Dynamic contrast enhanced MRI\u003c/p\u003e\n\u003cp\u003eHER2 Human epidermal growth factor receptor 2\u003c/p\u003e\n\u003cp\u003eCT Computed tomography\u003c/p\u003e\n\u003cp\u003eLASSO Least absolute shrinkage and selection operator\u003c/p\u003e\n\u003cp\u003eROC Receiver operating characteristic curve\u003c/p\u003e\n\u003cp\u003eBI-RADS Breast imaging reporting and data system\u003c/p\u003e\n\u003cp\u003eAUC Area under the curve\u003c/p\u003e\n\u003cp\u003eVOI Volume of interest\u003c/p\u003e\n\u003cp\u003eICC Intra-class correlation coefficient\u003c/p\u003e\n\u003cp\u003eIHC Immunohistochemistry\u003c/p\u003e\n\u003cp\u003eFISH Fluorescence in situ hybridization\u003c/p\u003e\n\u003cp\u003eTIC Time signal intensity curve\u003c/p\u003e\n\u003cp\u003eGO Gene ontology\u003c/p\u003e\n\u003cp\u003eKEGG Kyoto encyclopedia of genes and genomes\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe study was approved by the Institutional Review Boards and Human Ethics Committee of the Fifth Affiliated Hospital of Wenzhou Medical University and the First Affiliated Hospital of Zhejiang University. The requirement for informed consent was waived in accordance with the Helsinki statement.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe data presented in this study are available on reasonable request from the corresponding author.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll authors declare that they have no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was supported by Medical and Health General Project of Zhejiang Province (2024KY1860, 2023KY425).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors\u0026rsquo; contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was conceived by JJ, ZP, and MX with inputs from all other authors. XC, SL, DS, JD, JH, and SX were involved in data acquisition and image processing. CK and GL conducted the analyses. CK, GL, WC and MC were involved in literature research, data interpretation and the initial draft of the manuscript. JJ, ZP, and MX were involved in data interpretation and critical revision of the manuscript. All authors approved the final version of the manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot Applicable.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eSiegel RL, Miller KD, Fuchs HE, Jemal A. Cancer Statistics, 2021. CA: a cancer journal for clinicians.\u003cem\u003e \u003c/em\u003e2021; 71(1):7-33.\u003c/li\u003e\n\u003cli\u003eBitencourt AGV, Gibbs P, Rossi Saccarelli C, Daimiel I, Lo Gullo R, Fox MJ\u003cem\u003e, et al.\u003c/em\u003e MRI-based machine learning radiomics can predict HER2 expression level and pathologic response after neoadjuvant therapy in HER2 overexpressing breast cancer. EBioMedicine.\u003cem\u003e \u003c/em\u003e2020; 61:103042.\u003c/li\u003e\n\u003cli\u003eAhn S, Woo JW, Lee K, Park SY. HER2 status in breast cancer: changes in guidelines and complicating factors for interpretation. Journal of pathology and translational medicine.\u003cem\u003e \u003c/em\u003e2020; 54(1):34-44.\u003c/li\u003e\n\u003cli\u003eLoibl S, Gianni L. HER2-positive breast cancer. Lancet (London, England).\u003cem\u003e \u003c/em\u003e2017; 389(10087):2415-2429.\u003c/li\u003e\n\u003cli\u003eLam SW, Jimenez CR, Boven E. Breast cancer classification by proteomic technologies: current state of knowledge. Cancer treatment reviews.\u003cem\u003e \u003c/em\u003e2014; 40(1):129-138.\u003c/li\u003e\n\u003cli\u003eWaks AG, Winer EP. Breast Cancer Treatment: A Review. Jama.\u003cem\u003e \u003c/em\u003e2019; 321(3):288-300.\u003c/li\u003e\n\u003cli\u003evon Minckwitz G, Procter M, de Azambuja E, Zardavas D, Benyunes M, Viale G\u003cem\u003e, et al.\u003c/em\u003e Adjuvant Pertuzumab and Trastuzumab in Early HER2-Positive Breast Cancer. The New England journal of medicine.\u003cem\u003e \u003c/em\u003e2017; 377(2):122-131.\u003c/li\u003e\n\u003cli\u003eSwain SM, Miles D, Kim SB, Im YH, Im SA, Semiglazov V\u003cem\u003e, et al.\u003c/em\u003e Pertuzumab, trastuzumab, and docetaxel for HER2-positive metastatic breast cancer (CLEOPATRA): end-of-study results from a double-blind, randomised, placebo-controlled, phase 3 study. The Lancet Oncology.\u003cem\u003e \u003c/em\u003e2020; 21(4):519-530.\u003c/li\u003e\n\u003cli\u003eSlamon DJ, Leyland-Jones B, Shak S, Fuchs H, Paton V, Bajamonde A\u003cem\u003e, et al.\u003c/em\u003e Use of chemotherapy plus a monoclonal antibody against HER2 for metastatic breast cancer that overexpresses HER2. The New England journal of medicine.\u003cem\u003e \u003c/em\u003e2001; 344(11):783-792.\u003c/li\u003e\n\u003cli\u003eOrlando L, Viale G, Bria E, Lutrino ES, Sperduti I, Carbognin L\u003cem\u003e, et al.\u003c/em\u003e Discordance in pathology report after central pathology review: Implications for breast cancer adjuvant treatment. Breast (Edinburgh, Scotland).\u003cem\u003e \u003c/em\u003e2016; 30:151-155.\u003c/li\u003e\n\u003cli\u003eLeithner D, Bernard-Davila B, Martinez DF, Horvat JV, Jochelson MS, Marino MA\u003cem\u003e, et al.\u003c/em\u003e Radiomic Signatures Derived from Diffusion-Weighted Imaging for the Assessment of Breast Cancer Receptor Status and Molecular Subtypes. Molecular imaging and biology.\u003cem\u003e \u003c/em\u003e2020; 22(2):453-461.\u003c/li\u003e\n\u003cli\u003eMa W, Zhao Y, Ji Y, Guo X, Jian X, Liu P\u003cem\u003e, et al.\u003c/em\u003e Breast Cancer Molecular Subtype Prediction by Mammographic Radiomic Features. Academic radiology.\u003cem\u003e \u003c/em\u003e2019; 26(2):196-201.\u003c/li\u003e\n\u003cli\u003eLeithner D, Horvat JV, Marino MA, Bernard-Davila B, Jochelson MS, Ochoa-Albiztegui RE\u003cem\u003e, et al.\u003c/em\u003e Radiomic signatures with contrast-enhanced magnetic resonance imaging for the assessment of breast cancer receptor status and molecular subtypes: initial results. Breast cancer research : BCR.\u003cem\u003e \u003c/em\u003e2019; 21(1):106.\u003c/li\u003e\n\u003cli\u003ePisco AO, Huang S. Non-genetic cancer cell plasticity and therapy-induced stemness in tumour relapse: \u0026apos;What does not kill me strengthens me\u0026apos;. British journal of cancer.\u003cem\u003e \u003c/em\u003e2015; 112(11):1725-1732.\u003c/li\u003e\n\u003cli\u003eGillies RJ, Kinahan PE, Hricak H. Radiomics: Images Are More than Pictures, They Are Data. Radiology.\u003cem\u003e \u003c/em\u003e2016; 278(2):563-577.\u003c/li\u003e\n\u003cli\u003eGoto M, Le Bihan D, Sakai K, Yamada K. The diffusion MRI signature index is highly correlated with immunohistochemical status and molecular subtype of invasive breast carcinoma. European radiology.\u003cem\u003e \u003c/em\u003e2022; 32(7):4879-4888.\u003c/li\u003e\n\u003cli\u003eHuang Y, Wei L, Hu Y, Shao N, Lin Y, He S\u003cem\u003e, et al.\u003c/em\u003e Multi-Parametric MRI-Based Radiomics Models for Predicting Molecular Subtype and Androgen Receptor Expression in Breast Cancer. Frontiers in oncology.\u003cem\u003e \u003c/em\u003e2021; 11:706733.\u003c/li\u003e\n\u003cli\u003eLi W, Yu K, Feng C, Zhao D. Molecular Subtypes Recognition of Breast Cancer in Dynamic Contrast-Enhanced Breast Magnetic Resonance Imaging Phenotypes from Radiomics Data. Computational and mathematical methods in medicine.\u003cem\u003e \u003c/em\u003e2019; 2019:6978650.\u003c/li\u003e\n\u003cli\u003eWaugh SA, Purdie CA, Jordan LB, Vinnicombe S, Lerski RA, Martin P\u003cem\u003e, et al.\u003c/em\u003e Magnetic resonance imaging texture analysis classification of primary breast cancer. European radiology.\u003cem\u003e \u003c/em\u003e2016; 26(2):322-330.\u003c/li\u003e\n\u003cli\u003eSun X, He B, Luo X, Li Y, Cao J, Wang J\u003cem\u003e, et al.\u003c/em\u003e Preliminary Study on Molecular Subtypes of Breast Cancer Based on Magnetic Resonance Imaging Texture Analysis. Journal of computer assisted tomography.\u003cem\u003e \u003c/em\u003e2018; 42(4):531-535.\u003c/li\u003e\n\u003cli\u003eXie T, Zhao Q, Fu C, Bai Q, Zhou X, Li L\u003cem\u003e, et al.\u003c/em\u003e Differentiation of triple-negative breast cancer from other subtypes through whole-tumor histogram analysis on multiparametric MR imaging. European radiology.\u003cem\u003e \u003c/em\u003e2019; 29(5):2535-2544.\u003c/li\u003e\n\u003cli\u003eWolff AC, Hammond MEH, Allison KH, Harvey BE, Mangu PB, Bartlett JMS\u003cem\u003e, et al.\u003c/em\u003e Human Epidermal Growth Factor Receptor 2 Testing in Breast Cancer: American Society of Clinical Oncology/College of American Pathologists Clinical Practice Guideline Focused Update. Journal of clinical oncology : official journal of the American Society of Clinical Oncology.\u003cem\u003e \u003c/em\u003e2018; 36(20):2105-2122.\u003c/li\u003e\n\u003cli\u003evan Griethuysen JJM, Fedorov A, Parmar C, Hosny A, Aucoin N, Narayan V\u003cem\u003e, et al.\u003c/em\u003e Computational Radiomics System to Decode the Radiographic Phenotype. Cancer research.\u003cem\u003e \u003c/em\u003e2017; 77(21):e104-e107.\u003c/li\u003e\n\u003cli\u003eShin HJ, Kim HH, Huh MO, Kim MJ, Yi A, Kim H\u003cem\u003e, et al.\u003c/em\u003e Correlation between mammographic and sonographic findings and prognostic factors in patients with node-negative invasive breast cancer. The British journal of radiology.\u003cem\u003e \u003c/em\u003e2011; 84(997):19-30.\u003c/li\u003e\n\u003cli\u003eWang X, Chao L, Chen L, Tian B, Ma G, Zang Y\u003cem\u003e, et al.\u003c/em\u003e Correlation of mammographic calcifications with Her-2/neu overexpression in primary breast carcinomas. Journal of digital imaging.\u003cem\u003e \u003c/em\u003e2008; 21(2):170-176.\u003c/li\u003e\n\u003cli\u003eElias SG, Adams A, Wisner DJ, Esserman LJ, van\u0026apos;t Veer LJ, Mali WP\u003cem\u003e, et al.\u003c/em\u003e Imaging features of HER2 overexpression in breast cancer: a systematic review and meta-analysis. Cancer epidemiology, biomarkers \u0026amp; prevention : a publication of the American Association for Cancer Research, cosponsored by the American Society of Preventive Oncology.\u003cem\u003e \u003c/em\u003e2014; 23(8):1464-1483.\u003c/li\u003e\n\u003cli\u003eQi Y, Zhao T, Han M. The application of radiomics in predicting gene mutations in cancer. European radiology.\u003cem\u003e \u003c/em\u003e2022; 32(6):4014-4024.\u003c/li\u003e\n\u003cli\u003eDemircioglu A, Grueneisen J, Ingenwerth M, Hoffmann O, Pinker-Domenig K, Morris E\u003cem\u003e, et al.\u003c/em\u003e A rapid volume of interest-based approach of radiomics analysis of breast MRI for tumor decoding and phenotyping of breast cancer. PloS one.\u003cem\u003e \u003c/em\u003e2020; 15(6):e0234871.\u003c/li\u003e\n\u003cli\u003eZhou J, Tan H, Li W, Liu Z, Wu Y, Bai Y\u003cem\u003e, et al.\u003c/em\u003e Radiomics Signatures Based on Multiparametric MRI for the Preoperative Prediction of the HER2 Status of Patients with Breast Cancer. Academic radiology.\u003cem\u003e \u003c/em\u003e2021; 28(10):1352-1360.\u003c/li\u003e\n\u003cli\u003eLi H, Zhu Y, Burnside ES, Huang E, Drukker K, Hoadley KA\u003cem\u003e, et al.\u003c/em\u003e Quantitative MRI radiomics in the prediction of molecular classifications of breast cancer subtypes in the TCGA/TCIA data set. NPJ breast cancer.\u003cem\u003e \u003c/em\u003e2016; 2:16012-.\u003c/li\u003e\n\u003cli\u003eZhou J, Tan H, Bai Y, Li J, Lu Q, Chen R\u003cem\u003e, et al.\u003c/em\u003e Evaluating the HER-2 status of breast cancer using mammography radiomics features. European journal of radiology.\u003cem\u003e \u003c/em\u003e2019; 121:108718.\u003c/li\u003e\n\u003cli\u003eYang X, Wu L, Zhao K, Ye W, Liu W, Wang Y\u003cem\u003e, et al.\u003c/em\u003e Evaluation of human epidermal growth factor receptor 2 status of breast cancer using preoperative multidetector computed tomography with deep learning and handcrafted radiomics features. Chinese journal of cancer research = Chung-kuo yen cheng yen chiu.\u003cem\u003e \u003c/em\u003e2020; 32(2):175-185.\u003c/li\u003e\n\u003cli\u003eMiricescu D, Totan A, Stanescu S, II, Badoiu SC, Stefani C, Greabu M. PI3K/AKT/mTOR Signaling Pathway in Breast Cancer: From Molecular Landscape to Clinical Aspects. International journal of molecular sciences.\u003cem\u003e \u003c/em\u003e2020; 22(1).\u003c/li\u003e\n\u003cli\u003eXing Y, Lin NU, Maurer MA, Chen H, Mahvash A, Sahin A\u003cem\u003e, et al.\u003c/em\u003e Phase II trial of AKT inhibitor MK-2206 in patients with advanced breast cancer who have tumors with PIK3CA or AKT mutations, and/or PTEN loss/PTEN mutation. Breast cancer research : BCR.\u003cem\u003e \u003c/em\u003e2019; 21(1):78.\u003c/li\u003e\n\u003cli\u003eTurner NC, Oliveira M, Howell SJ, Dalenc F, Cortes J, Gomez Moreno HL\u003cem\u003e, et al.\u003c/em\u003e Capivasertib in Hormone Receptor-Positive Advanced Breast Cancer. The New England journal of medicine.\u003cem\u003e \u003c/em\u003e2023; 388(22):2058-2070.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Breast cancer, HER2 status, BI-RADS, Nomogram, PI3K-AKT","lastPublishedDoi":"10.21203/rs.3.rs-4152618/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4152618/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eHuman epidermal growth factor receptor 2 (HER2) status of breast cancer plays a critical role in guiding clinical treatment. We aimed to develop and validate a predictive model for HER2 status using preoperative dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI).\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eA total of 570 patients (282, 121 and 167 patients for training, internal and external test sets, respectively) with pathologically confirmed breast cancer and known HER2 status were recruited. A total of 851 radiomics features for each patient were extracted from preoperative DCE-MRI images. VarianceThreshold, SelectKBest and least absolute shrinkage and selection operator (LASSO) regression were applied to identify the optimal predictive features. Logistic regression was adopted to incorporate the Rad-score and clinical predictors into a nomogram. The performance of the nomogram was evaluated by area under receiver operating characteristic curve (AUC), calibration curve and decision curve. Additionally, gene expression analysis based on the Cancer Image Archive database was conducted to validate the biological interpretability of the model.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eTwenty-three radiomics features were selected to calculate the Rad-score. The Rad-score, along with breast imaging reporting and data system (BI-RADS) parameter, were independent predictors for HER2 status and were incorporated into the predictive model. The combined model achieved AUCs of 0.881, 0.883, and 0.798 in the training, internal and external test sets, respectively. Calibration curves demonstrated well agreement between the model predictions and actual HER2 status. Decision curve analysis further confirmed the clinical utility of the model. Differentially expressed genes between HER2-positive and HER2-negative patients were primarily involved in signaling pathways such as PI3K-AKT, endocrine resistance, and p53.\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e \u003cp\u003eThe combined model, which incorporated the Rad-score and BI-RADS, representing a potential and efficient alternative tool to evaluate HER2 status in breast cancer.\u003c/p\u003e","manuscriptTitle":"DCE-MRI Based Machine Learning Predictor for HER2-Positive Breast Cancer: A Feasibility and Validation Multicenter Study","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-03-29 18:58:08","doi":"10.21203/rs.3.rs-4152618/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"ae689dca-cf80-4ad5-a9ca-6db15e9ce6bb","owner":[],"postedDate":"March 29th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2024-12-12T09:54:14+00:00","versionOfRecord":[],"versionCreatedAt":"2024-03-29 18:58:08","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-4152618","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4152618","identity":"rs-4152618","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}