An integrated system for breast cancer diagnosis using convolution neural network and attention mechanism

Abstract

In most malignancies, breast cancer is fatal, accounting for approximately 500,000 annual deaths. The subtype of breast cancer known as Invasive Ductal Carcinoma (IDC) is surprisingly common. Pathologists commonly focus on IDC-containing regions when trying to determine if a patient has breast cancer. Although extremely fatal, survival rates and expected lifespans improve dramatically with prompt diagnosis and treatment. The treatment strategy also varies based on the breast cancer patient’s stage. In this research, we use a classification method for a publically available dataset of breast histopathology images obtained from the Kaggle. The IDC regions of the images in this dataset have been restricted for easy retrieval. The breast cancer IDC data set contains 277,524 records, of which 78,786 are positive. The 277,524 images were classified using an IDC breast cancer dataset, with 78,786 positive IDC and 198,738 negative IDC, respectively. The authors introduce a new architecture of deep convolutional neural networks and attention mechanism for classification. The model achieves state-of-the-art levels of accuracy for IDC identification, setting a new benchmark for future studies.