The value of MRI radiomics features for prediction of lymphovascular invasion in invasive breast cancer
10.3760/cma.j.cn112149-20220106-00016
- VernacularTitle:磁共振成像影像组学特征对乳腺癌淋巴血管浸润的预测价值
- Author:
Haotian WANG
1
;
Min ZHAO
;
Xuejiao FAN
;
Tao YU
;
Shu XU
Author Information
1. 中国医科大学肿瘤医院 辽宁省肿瘤医院放射科，沈阳 110042
- Keywords:
Breast neoplasms;
Magnetic resonance imaging;
Radiomics;
Lymphovascular invasion
- From:
Chinese Journal of Radiology
2022;56(9):982-988
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To investigate the value of MRI radiomics features in predicting breast cancer lymphovascular invasion (LVI).Methods:Totally of 216 patients with breast invasive ductal carcinoma who underwent preoperative MR examination confirmed by postoperative pathology from January to July 2021 in Liaoning Cancer Hospital were analyzed retrospectively. The patients were all females and ranged in age from 27 to 80 (53±11). Among them, 68 patients had LVI and 148 patients had no LVI. Patients were divided into the training set and the validation set in a ratio of 7∶3. The clinical features model was constructed with independent risk factors for LVI. The factors were extracted based on the clinical and MRI performance. Regions of interest in the tumor and peritumoral 1, 2, 3 mm annular region were delineated in the second phase of dynamic contrast-enhanced (DCE) MRI and DWI, respectively, and radiomics features extraction and screening were performed to construct a radiomics feature model. Receiver operating characteristic (ROC) curves were drawn to evaluate the diagnostic efficacy of models.Results:Apparent diffusion coefficient value (ADC) (OR=0.09, 95%CI 0.01-0.97, P=0.047), the axillary lymph node enlargement (OR=2.51, 95%CI 1.18-5.37, P=0.017), the peritumoral edema (OR=2.34, 95%CI 1.15-4.75, P=0.019) were independent risk factors for LVI. The clinical feature model was established with ADC value, the axillary lymph node enlargement and the peritumoral edema. At last, 10 radiomics features were selected to construct the DCE-MRI tumor model, 8 radiomics features were selected to construct the DCE-MRI peritumoral 1 mm model, 9 radiomics features were selected to construct the DCE-MRI peritumoral 2 mm model, 5 radiomics features were selected to construct the DCE-MRI peritumoral 3 mm model, 8 radiomics features were selected to construct the DWI tumor model, 5 radiomics features were selected to construct the DWI peritumoral 1 mm model, 10 radiomics features were selected to construct the DWI peritumoral 2 mm model, 9 radiomics features were selected to construct the DWI peritumoral 3 mm model. The ROC curve analysis showed that DWI peritumoral 1 mm model had the largest area under curve values for predicting breast cancer LVI status both in the training set (0.928) and the validation set (0.907), and there were significant differences compared with other models ( P<0.05). Conclusion:MRI radiomics features can effectively predict LVI of breast invasive ductal carcinoma, and DWI peritumoral 1 mm radiomics features model have the highest prediction efficiency for LVI.
