Early Detection of Brain Tumour in MRI Images using Open by Reconstruction and Convolution Neural Networks

Abstract

Classification and detection of the brain tumour at early stages have always been a concern to reduce the mortality rate. Though the brain tumour detection is possible in Magnetic Resonance Imaging (MRI), the detailed detection of the tumour type has been a concern. This article proposed a comparatively efficient method to detect the dangerous malignant tumour and hence begin the treatment at an early stage. At first, MRI images are filtered by cascading mean, median and Weiner filter. Due to the high density and texture, skull tends to appear as a detected region, which is often mistaken as part of a tumour. The stripping of the skull is done to isolate the Region of Interest (ROI) of the brain from the background. Once an abnormality in the image is confirmed for a tumour, its' classification into Low-Grade Glioma (LGG) and High-Grade Glioma (HGG) are done using Open by Reconstruction followed by thresholding segmentation method & Convolution Neural Networks (CNNs). An accuracy of 92.3% is obtained by first CNN in classifying abnormal brain MRI with normal brain MRI. An accuracy of 98.4% is obtained by second CNN in distinguishing HGG with LGG.