Automated diagnosis of breast cancer from ultrasound images using diverse ML techniques

An accurate diagnosis of breast cancer requires analyzing the tumors and giving appropriate treatment. A cancer diagnosis should be carried out as early as possible to minimize mortality rates. The handcrafted features are utilized for traditional breast cancer diagnosis, and the system’s performance is based on selected features. However, it is a challenging task to analyze shape complexity and different sizes. In recent years, deep learning has become a viable alternative to overcome the drawbacks of conventional cancer diagnosis methods. In this work, machine learning (ML) classifier, deep convolutional neural network (CNN) and transfer learning models (Alexnet, VGG-16, VGG-19, Resnet-50, and Resnet-101) are used to analyse the performance of the classifier. To process data before machine learning classification, an anisotropic diffusion filter is used to extract the tumor. The significant ten features were selected based on statistical testing and achieved an accuracy of 99% in the KNN classifier. The proposed deep CNN achieves a most satisfactory accuracy of 100%. The results outperform current techniques and demonstrate the ability of breast cancer classification. Also, the fast evaluation speed makes the analysis possible in real-time. The accuracy results of Alexnet, VGG-16, VGG-19, Resnet-50, and Resnet-101 are 86%, 82%, 84%, 84% and 74% respectively. Due to the small size of the data, the transfer learning model produces insignificant results, whereas the transfer learning models require a specific data size. The CNN can classify breast ultrasound images of cancer with promising results with relatively few training cases and is suitable for biomedical applications.