Segmentation of Brain Tumor in MRI Images Using CNN with Edge Detection

摘要

Brain tumor is a collection of mutated cells in brain. Gliomas are the most aggressive brain tumor. Gliomas generally growby diffuse infiltration in to the inner portion of the brain and these are therefore often not directly visible on the brain surface. The efficient treatment planning and detection is the way to improve the condition of oncological patients. The established technique is to detect the gliomas by the method of MRI (Magnetic Resonance Imaging). Manual segmentation takes reasonable time since large number of data generated by MRI have to be processed. For accurate segmentation, a novel completely automatic and reliable segmentation based on CNN is proposed. The 3×3 kernels provide a positive effect against over fitting. In pre-processing step, intensity normalization, which is not normally used in Convolutional Neural Network (CNN), is applied in this work. For applications like keyhole and the Nano robotic surgery it is beneficial to have correct edges of the particular tumor. Canny edge detection and edge detection using wavelet transform methods were performed on the image. Using PCNN method Image enhancement is done. The edge detection with image enhancement will lead to a wide variety of surgical applications in medical field.