Breast Cancer Detection in Magnetic Resonance Images using Edge-Enhanced Convolutional Neural Network

Abstract

Early detection of breast cancer plays a crucial role in its diagnosis and treatment. This research presents an edge-enhanced Convolutional Neural Network (CNN) model for breast cancer detection and classification using Magnetic Resonance Imaging (MRI) scans. A dataset of 1480 breast MRI scans (comprising of healthy and diseased MRI breast scans) was sourced from Kaggle, an open repository for medical images. The images were pre-processed via resizing, de-noising, normalization, and contrast enhanced using contrast limited adaptive histogram equalization (CLAHE) to obtain uniform and standard images. Pre-processed images were then split into training, testing and validation sets in the ratio 60:20:20. To enhance the structural descriptors, the grayscale MRI images were further processed by means of Canny edge- and Harris corner- detection algorithms, leading to generation of additional two versions of each of the images: Canny edge map and Harris corner mask. These two versions together with the grayscale version constitute the three inputs that were fed into the CNN model. The CNN model uses three convolutional blocks with rectified linear unit activation, max-pooling, dropout regularization, a sigmoid output, binary cross-entropy loss with Adam optimizer for 15 epochs and batch size of 16. The model achieves an accuracy, precision, recall, and F1-score of 97%, 99%, 95%, and 97%, respectively. The results indicate that incorporating explicit edge and structural information into the CNN input can enhance the aid detection of breast cancer in MRI images