MRI Based Glioma Detection and Classification into Low-grade and High-Grade Gliomas

Brain tumors are one of the most rapidly spreading types of tumors known to humans. The worst and most dangerous type of tumor is a brain tumor. However, if diagnosed early, patients with brain tumors have a higher chance of survival acknowledgments to simple and inexpensive treatments. Expert radiologists, equipment, and biopsies are used in the traditional method of diagnosing a brain tumor. Machine learning has proved to deliver cutting-edge methods for early identification of brain tumors with better accuracies, avoiding costly diagnoses and unnecessary biopsies and assisting radiologists. Using a machine learning approach, this study proposes a technique for brain tumor classification and segmentation as HGG and LGG (High-Grade Glioma & Low-Grade Glioma). One of the most inflexible and innovative challenges confronting artificial intelligence approaches is medical diagnostics utilizing image processing and machine learning. The project involves the preprocessing, edge detection, segmentation, feature extraction, and classification of MRI brain images. The preprocessing is implemented by using median filter and canny edge detection is adapted in edge detection stage to inspect the best performing edge detector in terms of accuracy. Then, the MR image is segmented by K-means clustering technique. However, some of the important features are extracted including GLCM features for texture identification. Finally, in the classification phase, the Support Vector Machine (SVM) and k-nearest neighbors (KNN) classifiers are used. After using these classifiers, we distinguished the tumors as HGG or LGG. To determine whether an MRI image of the brain has a tumor and to classify as HGG or LGG, a machine learning methodology is applied. The aim is to develop a system with better tumor detection from MRI images to be used as a tool in real time by employing machine learning approach. The proposed method is validated using the MATLAB environment on the available BRATS 2019 dataset. Then, to illustrate the performance of SVM and KNN classifiers, a confusion matrix is frequently used. The SVM classifier achieves a maximum accuracy of 92%.