Morphological and Otsu’s Technique Based Mammography Mass Detection and Deep Neural Network Classifier Based Prediction

Abstract

Breast cancer is a leading cause of death among women. The death rate is reduced when this disease is detected early with the help of mammography. Deep learning is a method that radiologists use and request to help them make more accurate diagnoses and enhance their outcome predictions. This work presents a novel strategy comprising a pre-processing method and a mix of morphological and multi-thresholding using Otsu's technique based segmentation technique, which was tested on the Mini-MIAS dataset of 322 images. For Speed-Up Robust Features (SURF) selection, the inbuilt feature extraction is done utilizing multiple colour and texture features approaches. At the classification level, a new layer is added that performs 70 percent training and 30 percent testing of the deep neural network. There are two primary steps in the training phase: (1) Develop a model for dividing breast tissue into dense and non-dense categories. (2) Develop a model for classifying breast regions into mass and non-mass. The results show that the accuracy rate of the proposed automated DL approach is higher than that of other state-of-the-art models. The average accuracy (ACC) rates of the three types of cancer, i.e., normal, benign, and malignant cancer, utilizing the suggested method are 91 percent, 94 percent, and 93 percent, respectively, according to experimental results.