The value of intra-tumoral and peri-tumoral early dynamic contrast-enhanced MRI-based radiomics models in identifying benign from malignant in breast imaging-reporting and data system 4 breast tumors
10.3760/cma.j.cn112149-20210829-00616
- VernacularTitle:基于瘤内及瘤周早期动态增强MRI的影像组学模型鉴别乳腺影像报告和数据系统4类肿瘤良性与恶性的价值
- Author:
Shuhai ZHANG
1
;
Xiaolei WANG
;
Yun ZHU
;
Zhao YANG
;
Junjian SHEN
;
Qilin NIU
;
Lu CHEN
;
Yichuan MA
;
Zongyu XIE
Author Information
1. 蚌埠医学院研究生院，蚌埠233030
- Keywords:
Breast neoplasms;
Radiomics;
Magnetic resonance imaging;
Peri-tumoral
- From:
Chinese Journal of Radiology
2022;56(7):758-765
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To explore the value of radiomics model based on intratumoral and peritumoral early dynamic contrast-enhanced (DCE) MRI for identifying benign and malignant in breast imaging reporting and data system (BI-RADS) 4 tumors.Methods:A total of 191 patients diagnosed with BI-RADS 4 breast tumors by breast MRI examination with clear pathological diagnosis from January 2016 to December 2020 in the First Affiliated Hospital of Bengbu Medical College were analyzed retrospectively, including 77 benign and 114 malignant cases, aged 23-68 (46±10) years. The one-slice image with the largest area of the lesion of the second stage DCE-MRI images was selected to outline the region of interest, and automatically conformal extrapolated by 5 mm to extract the intra-tumoral and peritumoral radiomics features. The included cases were randomly divided into training and testing cohorts in the ratio of 8∶2. The statistical and machine learning methods were used for feature dimensionality reduction and selection of optimal radiomics features, and logistic regression was used as the classifier to establish the intratumoral, peritumoral, and intratumoral combined with peritumoral radiomics models. The independent risk factors that could predict the benignity and malignancy of breast tumors were retained as clinical-radiological characteristics by univariate and multivariate logistic regression to establish a clinical-radiological model. Finally, the intratumoral and peritumoral radiomics features were combined with clinical-radiological features to develop a combined model of the three. The receiver operating curve was used to analyze the predictive performance of each model and calculate the area under the curve (AUC),the AUC was compared by DeLong test. The stability of the three-component combined diagnostic model was tested by 10-fold cross-validation, and the model was visualized by plotting nomogram and calibration curves.Results:In the training cohort, the AUC of the three-component combined model for identifying benign and malignant BI-RADS 4 breast tumors was significantly higher than that of the intratumoral radiomics model ( Z=3.38, P<0.001), the peritumoral radiomics model ( Z=4.01, P<0.001), the intratumoral combined with peritumoral radiomics model ( Z=3.11, P=0.002), and the clinical-radiological model ( Z=3.24, P=0.001). And the AUC, sensitivity, specificity, accuracy, and F1-score of the three-component combined model were 0.932, 91.2%, 86.9%, 87.0% and 0.89, respectively. In the testing cohort, the three-component combined model also had the highest AUC value (0.875), and diagnostic sensitivity, specificity, accuracy and malignancy F1-score were 95.7%, 62.5%, 76.9%, and 0.89, respectively. The AUC calculated by 10-fold cross-validation was 0.90 (0.85-0.92), and the predicted curve of the three-component combined model in the calibration curve was in good agreement with the ideal curve. Conclusion:The three-component combined diagnostic model based on the intratumoral and peritumoral radiomics features and clinical-radiological features of early DCE-MRI has good performance and stability for identifying the benign and malignant in BI-RADS 4 breast tumors, and it can provide guidance for clinical decision non-invasively.
