BRAIN TUMOUR DETECTION USING CONVOLUTIONAL NEURAL NETWORKS AND DECONVOLUTION

A Tumor is known as the aberrant growth of cells over a particular region of human body. Brain tumor also being one among those and it is capable of causing serious mental disabilities and issues related to the central nervous system, excessive growth of these tissues could ultimately lead to further complications like muscle paralysis, may also lead to fatal death. Considering all these conflicts detection of the tumor in very early stages is essential or else it would end up in causing lethal effects to the nervous system. MRI (Magnetic resonance imaging) scans helps us in diagnosing these brain tumors, but the process involved in detecting these tumors is Human-driven and arduous, also neurologists do generally take reasonable time to detect these tumors, this method of detection can also lead to human errors, so to avoid all these conflicts it is highly required to choose cogentpaths and design an effective model for the detection of brain tumors. This research work has proposed a model that involves an autonomous tumor detection technique for detecting cancerous tumor named Gliomas using convolutional neural networks, robust networks like VGG16 and VGG-19 are used in the process of detection of tumor and further this research has used deconvolution process on the VGG-16 model for a better feature extraction followed by CRF-RNN in the final layer for classification purpose instead of FCN. All these different models used for detecting brain tumor have performed well and has yielded us a very high accuracy rate of 95% and 96% when trained on VGG-16 and VGG19 network respectively. Then, the model has applied deconvolutional process on VGG-16 followed by CRF-RNN is also be able to classify the brain tumor effectively and it have yielded us a good accuracy rate of 92.3%.