The Performance Evaluation of Pre-trained CNN Architectures for Tumor Classification

Abstract

Tumor is characterized by abnormal and unregulated proliferation of cells in the human body. Slow-growing tumors with clear boundaries and that seldom metastasize, are called benign tumors. Tumors that have the capacity to both intrude nearby normal tissue and propagate throughout the body via the circulatory or lymphatic system are called malignant or cancerous tumors. Both benign as well as malignant tumors are life threatening as they can result in serious dysfunction of the brain. The optimal treatment of brain tumor depends on its early detection and classification. Developments in the field of artificial intelligence have aroused tremendous interest in designing automated diagnostic systems. This study aims to perform an initial investigation of the performance of various pre-trained CNN architectures in tumor detection from medical images in a uniform environment. Results established the superiority of inception class networks.