Transfer Learning based Deep Neural Network for Autonomous identification of Brain Tumor Disease

Due to the significant role of brain in controlling as well as coordinating multi-functions of body, it is considered as a complex and intricate organ due to which any damage to the tissues lying within brain can cause serious problem for an individual if not identified and treated timely. Though most of the medical professionals ask their patients to undergo magnetic resonance imaging (MRI), computed tomography etc. which further rely on experts for providing interpretations to identify presence of abnormal growth of tissues. Thus, asymptotic behaviour at an early stage may cause them to give false interpretations especially when tumor growth is too small and doesn’t get identified at initial stage. Developing computer-assisted system can aid radiologists to overcome this issue while making any decisions in this regard for achieving more accurate and efficient results while diagnosing brain tumor disease in healthcare as compared to the conventional approaches. In this work, extended version of residual network namely ResNet152 has been integrated with support vector machines for extracting significant features and distinguishing infected images from healthy brain images respectively. Thus, proposed model is trained well following transfer learning process and validated on unseen data after which it gained 97.50% accuracy, 98.33% sensitivity, 96.67% specificity, 96.72% precision and 97.52% F1-Score due to which this hybrid model performed effectively while diagnosing brain tumor rather than existing state-of-art approaches.