Smart Detection and Localization of Tumor in MRI Brain by using Deep Learning and Segmentation Techniques

Abstract

Magnetic resonance image processing is a rising area to study dangerous diseases like brain tumors. Tumor detection in MRI brain is assumed to be a difficult task. Several methods are in practice for the scanning of brain like MRI, CT scan, X-ray, SPECT and PET. Images generated by these techniques are analyzed by physicians for the detection of tumor. MRI is generally used to extract meaningful information from images with high accuracy. This article suggests a system that provides technique to detect and get the location of tumor in MRI by the means of deep learning and image segmentation techniques.  The system follows the sequence of processes steps like pre-processing, segmentation, feature extraction, detection and localization of tumor. Pre-processing applies filters to take out the unnecessary noise and artefacts. Threshold, histogram and edge based segmentations are used to partition the MRI into segments.  GLCM is used to extract image features from MRI. Detection and tumor localization is performed through CNN and segmentation algorithms. Practical implementation of this system is accomplished by using Python programming language. MRI images are gathered online from Kaggle dataset. Initially, dataset contains 253 images of T1, T2 weighted and Flair modalities for experimental work. After augmentation the size of dataset is increased to 2318. This system assures 98.02% accuracy of tumor brain detection.