Breast Cancer Detection in the Equivocal Mammograms by AMAN Method

Abstract

Breast cancer is a primary cause of human deaths among gynecological cancers around the globe. Though it can occur in both genders, it is far more common in women. It is a disease in which the patient’s body cells in the breast start growing abnormally. It has various kinds (e.g., invasive ductal carcinoma, invasive lobular carcinoma, medullary, and mucinous), which depend on which cells in the breast turn into cancer. Traditional manual methods used to detect breast cancer are not only time consuming but may also be expensive due to the shortage of experts, especially in developing countries. To contribute to this concern, this study proposed a cost-effective and efficient scheme called AMAN. It is based on deep learning techniques to diagnose breast cancer in its initial stages using X-ray mammograms. This system classifies breast cancer into two stages. In the first stage, it uses a well-trained deep learning model (Xception) while extracting the most crucial features from the patient’s X-ray mammographs. The Xception is a pertained model that is well retrained by this study on the new breast cancer data using the transfer learning approach. In the second stage, it involves the gradient boost scheme to classify the clinical data using a specified set of characteristics. Notably, the experimental results of the proposed scheme are satisfactory. It attained an accuracy, an area under the curve (AUC), and recall of 87%, 95%, and 86%, respectively, for the mammography classification. For the clinical data classification, it achieved an AUC of 97% and a balanced accuracy of 92%. Following these results, the proposed model can be utilized to detect and classify this disease in the relevant patients with high confidence.

Keywords

• breast cancer classification
• deep learning
• CNN
• equivocal mammogram
• BI-RADS classification
• machine learning