| Abstrakt: |
Breast cancer is an extremely prevalent cancer globally and the prominent cause attributing to cancer-related fatalities. The grade of breast cancer is a prognostic marker representing its aggressive potential. Morphologically, tumors that are well differentiated, have a highly noticeable basal membrane, and with moderate proliferation are considered low-grade (Grade I & II). Tumors with a massive nucleus, irregular shape, and size, prominent nucleoli, inadequate cytoplasm, and high intensity are High grade (Grade III & IV). Dynamic Contrast-EnhancedMRI (DCE-MRI) has been extensively used to assess tumors and tumor grades, with an emphasis on heterogeneity and integrated inspections. Neoadjuvant chemotherapy (NAC) for breast cancer is traditionally administered to patients with locally advanced disease and is advantageous for surgical downstaging. Generally, the histological grade and proliferationindex decrease after neoadjuvant chemotherapy and are connected to the therapeutic response. Radiomics is a novel approach for discovering tumor pathophysiological-related image information and possibly a pre-operative predictor of breast cancer pathological grade. Due to the heterogeneous nature of the tumor, histological grading remains challenging for the radiologist. This work extracts radiomics-based features from the QIN BREAST and QIN BREAST-02 datasets (N=47) of the publicly available TCIA database. The extracted features are used in the classification of low- and high-grade tumors by using an array of Support vector machines (SVM) algorithms such as Quadratic SVM, Linear SVM, Cubic SVM, and Medium Gaussian SVM. Results show that the test accuracy for the LinearSVM is 81.2%, AUC of 0.75, a sensitivity of 0.85, and an F-score of 0.89 which is observed to have better performance than other SVM models. Hence, radiomics-based grade differentiation using DCE MRI in patients with breast cancer could help to determine the potential for recovery with the right treatment. [ABSTRACT FROM AUTHOR] |
