Performance Analysis of Low and High-Grade Breast Tumors Using DCE MR Images and LASSO Feature Selection

Breast cancer is a complex genetic disease with diverse morphological and biological characteristics. Generally, the grade of a breast tumor is a prognostic factor and representation of its potential aggressiveness. Presently, Dynamic Contrast-Enhanced MRI (DCE-MRI) has gained a predominant role in assessing tumor grades and vascular physiology. However, due to tumor heterogeneity, tumor-grade classification is still a daunting challenge for radiologists. Therefore, to unburden the tumor grading process, a study has been carried out with 638 patients taken from the Duke-Breast-Cancer-MRI database (431-low-grade & 207-high-grade). The clinicopathological characteristics such as ER receptors, PR receptors, HER2, Pathological Complete Response (PCR or non-PCR), Menopausal status, and Bi-lateral status have shown a high significance of p = <0.00001, <0.00001, 0.0023, <0.00001, 0.0262, and 0.0045 respectively. The LASSO (Least Absolute Shrinkage and Selection Operator) feature selection model has selected 8 optimal features out of 529 feature sets (from Duke-Breast-Cancer-MRI). The extracted features are involved in the classification of high-grade and low-grade tumors by using a collection of classifiers such as Linear Support Vector Machines (LSVM), Logistic regression (LR), Linear discriminant analysis (LDA), Gaussian Naïve Bayes (GNB), k-Nearest Neighbors (KNN), and Random Forest (RF). The outcome of the L-SVM and LR showed better performance metrics values among all classifiers. Hence, the acquired classification results disclose that histological grade prediction using radiomics would aid clinical management and prognosis.