Breast Cancer Detection in Deep Learning based Architectures using Mammogram Images

Abstract

Breast cancer is a very devastating type of cancer that typically develops in breast cells and in the past few decades, the incidence rate of breast cancer has risen steadily. The diagnosis mostly depends on the radiologists’ experience, questionable diagnoses are common, and there are worries about lawsuits due to incorrect diagnoses or missed lesions. Therefore, developing deep learning models for breast cancer identification is urgently needed, despite substantial advancements in the treatment of primary breast cancer during the past ten years. In Deep Learning, the convolutional neural network can be a useful technique for classification problems.In this study, some of the Transfer learning models are applied, which are pre-trained models that are trained on huge datasets such as the ImageNet dataset, which contains about 1.2 million images. However, these models are VGG16 VGG19 Mobilenet V2, and it is known that these models are used for classification problems, and therefore it will be used in this study in addition to developing a Convolutional Neural Network, which achieved the highest accuracy and best performance, compared to the applied models. These models were applied to the presented Mammogram Images dataset, which contains eight classes, each of which represents a form of mass that can be cancerous or benign. And since the dataset is unbalanced in terms of the distribution of samples for the classes, the Oversampling technique was applied, which had a key role in improving the performance of the models, especially for the CNN network, which obtained the highest accuracy of 96%.