Studies on Brain Tumor Classification using Learning Techniques in Medical Images

Abstract

Medical image analysis using Computer-Aided Diagnosis (CAD) has recently played a crucial role in diagnosing brain tumors. Early detection of brain tumors is essential for saving lives. Accurate segmentation and classification methods using Magnetic Resonance Imaging (MRI) aid in determining brain tumors. Previous researches have focused on classifying normal and abnormal brain MRI images. The goal of this study is to improve the efficiency and accuracy of brain MRI image segmentation and classification. The proposed methodology classifies the tumors as normal or abnormal and uses different deep-learning models to classify different stages of abnormality. The efficiency of four major approaches for brain MRI segmentation and classification are compared in this study. newlineThe detection of tumors in the brain in medical image analysis is one of the most challenging tasks and can be tackled through MRI, CT, and PET methodologies. In this work, the Support Vector Machine(SVM) classifier is utilized to detect the portion of the image affected bythe tumor. The image is first cleansed of noise through the application ofthe Median filter. Gabor filter performs the detection of edges, extractionof features and removal of noise. Following this, morphological functions such as erosion and dilation are applied to the filtered image, resulting in the separation of the enclosed regions using the SVM classifier. The classifier is effective in identifying the early stages of the tumor and segmenting the affected portion. newlinevi newlineSecondly, a new approach for categorizing brain tumors in Magnetic Resonance (MR) images is introduced. This method aims to improve treatment by differentiating between meningiomas, gliomas, and pituitary tumors. The technique consists of two feature extraction methods newline- Edge Oriented Multi-Texton (EOMT) and Local Coherence Multi- Texton (LCMT). By combining both edge and local coherence information, multi-texton can be separated accurately, leading to improved tumor classification.