Machine learning based computer aided diagnosis of breast tumor from Dynamic Contrast Enhanced MRI images

Worldwide, breast cancer (BC) is the most common cancer to be diagnosed in women. It is currently the second most common cause of cancer-related death, trailed only by lung cancer. When compared to ultrasound and mammography, breast dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) is helpful for the diagnosis of breast tumors. Breast cancer diagnosis with DCE-MRI has demonstrated good sensitivity. Specifically, a dataset of 500 images including 250 each for benign and malignant is used. Deep learning based Attention U-Net model was used to perform the segmentation of the breast tumor. Various feature extraction techniques such as statistical features, Haralick features, Edge features, Shape features were extracted after segmentation. After performing feature extraction ML classification was implemented to classify benign and malignant tumor. SVM, Random Forest classifier, Gaussian Naïve Bayes and K-Nearest Neighbours were used for classification. Among them K-Nearest Neighbours has performed better than all other ML models with accuracy of 82%