Breast cancer (BC) is the most prevalent malignant tumor type threatening women’s health globally (1). According to an immunohistochemistry (IHC) classification system, BC can be divided into four subtypes, basal-like, HER2-enriched, and luminal A and B subtypes, on the basis of the expression of progesterone receptor (PR), estrogen receptor (ER), human epidermal growth factor receptor 2 (HER2), and Ki-67 (2). Ki-67 protein is a recognized marker of tumor proliferation and invasiveness (3), as well as a recognized indicator of BC prognosis (4). Ki-67 can be used as a molecular marker to distinguish the molecular subtypes of luminal A and B (5). A high expression level of Ki-67 is associated with poorer prognosis (3, 6), greater risk of recurrence (7), and worse survival outcomes (8). Hence, accurately identifying the status of the Ki-67 index is crucial for the prognosis of BC.

Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) is useful for assessing tumor anatomical information and angiogenesis (9). Radiomics involves high-throughput extraction of many image-based features from standard medical images and determining the potential links between these features and pathophysiology (10, 11). Radiomics analysis of features extracted from DCE-MRI images can be used to distinguish the HER2 2+ status, predict lymphovascular invasion, determine the status of lymph node metastasis, and identify the degree of tumor malignancy (12–19). The apparent diffusion coefficient (ADC), a quantitative parameter generated in diffusion-weighted imaging (DWI), is the most used clinical parameter reflecting the degree of tissue distribution according to the diffusion of water molecules (20). Since ADC is influenced by cell density and tissue structure, Choi (12) proposed that DCE-MRI combined with DWI is helpful to evaluate the status of lymphovascular invasion in patients with node-negative invasive BC. In addition, ADC values have been shown to correlate with the Ki-67 index (20, 21). Therefore, radiomics analysis based on DCE parameters and ADC might have the potential to predict Ki-67 status and even improve predictive performance.

A previous study has shown that the radiomics features derived from DCE-MRI functional parameter maps achieved the best results in identifying sentinel lymph node metastasis status in patients with BC (22). Another study has predicted the Ki-67 index and HER2 2+ status by using intratumoral and peritumoral radiomics features based on six DCE-MRI functional parameter maps (14). Both of these studies used single-layer lesions and consequently might have overlooked the correlations between layers. Jong et al. have investigated the correlation between quantitative MR parameters and Ki-67 expression status by analyzing DCE-MRI and DWI sequences in ER-positive invasive BC (23). However, their analysis of the interstitial signal enhancement ratio used only univariate and multivariate analysis, without radiomics analysis. To our knowledge, few studies have used a combination of breast MRI functional parametric maps and ADC maps in radiomics analysis. Moreover, in most prior studies, region of interest (ROI) depiction has been performed primarily on the slice images showing the largest tumor size (24–26). In this study, radiomics features were extracted from the entire tumor volume on the basis of three semi-quantitative parametric maps and ADC maps, and the predictive performance of the classification models based on three-dimensional features in terms of Ki-67 expression status was evaluated.

Therefore, the purpose of our study was to evaluate the performance of a radiomics model based on the entire tumor region from three DCE-MRI parametric maps and ADC maps to determine the status of Ki-67 in patients with BC.

In this study, we explored whether the radiomics features of DCE-MRI parameter maps and ADC maps in patients with BC could be used to predict the preoperative Ki-67 proliferation index. The radiomics model constructed in this study performed well in identifying the low and high expression status of Ki-67.

Many previous studies have described Ki-67 expression, on the basis of IHC, as a prognostic and predictive indicator of BC. Higher Ki-67 expression status is associated with poorer response to treatment and poorer prognosis (33). The Ki-67 index can also play a role in distinguishing molecular subtypes of BC together with HER2 expression status (2). However, pathological biopsy only requires sampling in a part of the tumor tissue, so the Ki-67 proliferation index in the test results may not reflect tumor heterogeneity. With the development of imaging technology in recent years, imaging can provide a holistic picture of the anatomy and function of the tumor tissue. Therefore, imaging methods may be more convenient and may even provide more biological information for determining the expression status of Ki-67.

DCE-MRI is a highly sensitive method but has only moderate specificity for the diagnosis of invasive BC (34). In recent years, DWI with ADC maps has been increasingly used in multiparameter imaging environments for BC examination because it can be performed without a contrast agent (35–37). DWI can quantitatively measure the Brownian motion of free water in the tissue to provide functional information about the tissue structure and does not require intravenous injection (38). In addition, ADC has been found to increase the specificity of breast tumor diagnosis and complement DCE-MRI in tumor qualitative aspects (39, 40). In this study, we selected radiomics features extracted from DCE-MRI parameter maps and ADC maps and finally established six SVM classifier models. The ROC curve showed that the AUC score of the model combined with the parameter maps and ADC maps was higher than that of other single-parameter models in the training set and validation set. Therefore, the radiomics model including features of DCE-MRI parameter maps and ADC maps could improve the performance of Ki-67 expression status discrimination.

Radiomics, which provides potential biomarkers for clinical results through extracting and analyzing image features, is a relatively new technology (11). BC is highly heterogeneous. Compared with traditional genomics and proteomics, radiomics not only can non-invasively assess the tumor and its microenvironment but also can predict the genetic heterogeneity of the tumor (41). Herein, we used radiomics to quantitatively extract features within tumors on the basis of DCE-MRI parameters and ADC maps to reflect the heterogeneity of the internal tumor structure. A previous study has modeled radiomics features extracted from T2-weighted and contrast-enhanced T1-weighted images of BC and indicated that T2-weighted classifiers were important predictors of Ki-67 status (42). Li et al. (14) have combined peritumoral and intratumoral features from DCE-MRI functional parameter maps to determine Ki-67 status. Therefore, in this study, we established a multiparameter model based on the radiomics features of functional parametric maps and ADC maps. Our study achieved better predictive performance than the above multiparameter studies. Another multiparametric MRI study also using DWI sequences has achieved good performance in predicting Ki-67 status (24). However, these prior studies have analyzed only the largest tumor slices in two dimensions and therefore were unlikely to fully assess the heterogeneity of BC. The radiomics features that we extracted in this study were obtained from the three-dimensional volume model of the tumor, taking into account the interlayer correlation, and thus fully revealed the BC heterogeneity.

The current study had several limitations. First, the retrospective design of this study might have introduced inherent variations and biases. One source of variation was the changes in repetition time during DCE-MRI acquisition; however, the difference in MR signal intensity was not the principal factor in our study, because the radiomics features were derived from three parametric maps reflecting the changes in contrast medium concentration. Although the enhancement ratio is a function of repetition time, these functions change relatively slowly (43). Therefore, a well-designed prospective study supporting the results of this study is necessary. Second, this was an independent single-center study, and the number of patients included was limited. Therefore, the model established herein had several limitations, and datasets from other imaging units might have different spatial resolutions. Consequently, further research is required to verify the diagnostic performance of our model in a large multi-center patient sample. Finally, only semiquantitative parametric images were analyzed in this study. The application value of quantitative parametric images in radiomics will be further explored in future studies.

In this study, our experimental results demonstrated that radiomics analysis based on DCE-MRI parameter maps and ADC maps can feasibly be used to predict the Ki-67 status in BC. Given that several different biomarkers must be integrated to make clinical management decisions for patients, our proposed model can be further extended in the future, such as by including more scanning sequences and predicting more molecular subtypes to support clinical decisions.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving human participants were reviewed and approved by Shengjing Hospital of China Medical University. The ethics committee waived the requirement of written informed consent for participation. Written informed consent was not obtained from the individual(s) for the publication of any potentially identifiable images or data included in this article.

SF conducted data analysis and manuscript writing. JY was responsible for the manuscript revision. All authors critically reviewed and revised the manuscript. All authors contributed to the article and approved the submitted version.

This study is supported by Research and development (R&D) foundation for major Science and Technology from Shenyang (No.19-112-4-105), Big data foundation for health care from China Medical University (No. HMB201902105) and Natural Fund Guidance Plan (No. 2019-ZD-0743).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.