A Multi-Layer Feed Forward Neural Network for Breast Cancer Diagnosis from Ultrasound Images

Abstract

Diagnosis of breast cancer from ultrasound images (USIs) and images processing are two main stages of medical computing field. In this paper, we propose a Multi-Layer Feed Forward Neural Network (MLFNN) for classification of benign and malignant breast tumors by using a Python based implementation. The neural model is trained using the preprocessed regions of interests (ROIs) from USIs that belong to the Breast Ultrasound Dataset (BUSI dataset). The preprocessing procedure includes extracting the ROIs, resizing, normalizing, and flattening. The ROIs are obtained with our own algorithm that overlaps the original image with its corresponding ground truth image. More images and tumor shapes employed in the training stage of the neural network can lead to better prediction performances. In this study, the binary classification of tumors into benignancy or malignancy gives 99% training accuracy, 86% validation accuracy and 71.43% test accuracy.