Investigations of Shallow and Deep Learning Algorithms for Tumor Detection

With the increasing ability of the computer aided detection and diagnosis system, the exploration on the tumors have attained a breakthrough by decreasing the mortality rate. Radiologists perform manual identification for both the diagnosis and prognosis of tumors in the victims. The research on the diagnosis of the tumor have reached a huge milestone using both the shallow and the deep learning algorithms. In this paper, classification of mass lesions in breast cancer dataset has been performed, for studying the comparison of shallow learning algorithms with that of the deep learning algorithms. For shallow learning the texture and statistical features were extracted, used in the contemporary classification algorithms. Of all shallow learning algorithms, support vector machines (SVM) showed 90.6% accuracy, whereas among the deep learning algorithms, the global features are taken for consideration and VGG16 showed 92.6% accuracy. Thus, these deep learning models outperformed the shallow algorithms and have been a state-of-art algorithm for an accurate and automated breast tumor detection.