Diagnosis of breast cancer for modern mammography using artificial intelligence

Abstract

The diagnosis of breast cancer, one of the most common types of cancer worldwide, is still a challenging task. Localisation of the breast mass and accurate classification is crucial in detecting breast cancer at an early stage. In machine learning-based classification models, performance is dependent on the accuracy of extracted features and is susceptible to saturation problems. Deep learning methods are currently used to learn self-regulating top-level features and achieve remarkable accuracy. It has long been recognised that mammography is competent for the early detection of cancer cells. Thus the technique of image segmentation and artificial intelligence can be applied to the initial stage diagnosis of breast cancer. The proposed method is composed of two major approaches. In the first, the transfer learning method is employed. In the second, convolution neural network architecture is constructed, and its hyper-parameters are adjusted to achieve accurate classification. The result indicates that the proposed methods achieve significant accuracy for MIAS (95.95%), DDSM (99.39%), INbreast (96.53%), and combined datasets (92.27%). Comparison of results of the proposed approach with current schemes demonstrates its efficiency.