ROI Segmentation for Breast Cancer Classification : Deep Learning Perspective

Breast cancer mammogram images are types of unstructured data that need to be analyzed in order to classify and detect abnormalities at an early stage of cancer. In this research, ROI segmentation for the breast cancer classification CAD model is proposed. The proposed model is trained and tested on the MIAS dataset, evaluating the performance of CNN (ResNet-18 and ResNet-50) classifiers along with transfer learning. The medical imaging dataset consists of noise and low-quality images, so to overcome this problem, medium filtration for denoising and segmentation with the region of interest process are applied. Region-based segmentation is used to segment the images based on local parameters. For training the adopted model, ResNet-18 (two layers) and ResNet-50 (three layers) were used, achieving the best result at an 80:20 data split ratio in both cases (without and with transfer learning) of 81.6% and 96.6% accuracy by using ResNet-18, and ResNet-50 achieving 89 % and 98.45 % accuracy (without and with transfer learning). Such promising outcomes will provide a great opportunity to use the ResNet classifier with transfer learning as an early detection model in hospitals or in the Healthcare sector.