Novel Approach to Glioma Classification in MRI Images Using Modified Microstructure Descriptors and Hybrid-RBF Kernel SVM

Cancer kills the most people worldwide and will exceed all others in the next decades. Brain cancer mortality is best reduced by early detection and treatment. Modern computer technology and computational imaging have greatly enhanced medical image interpretation, aiding early diagnosis. GBM is characterized by substantial tumor heterogeneity, no effective therapies, and poor survival rates. Current best practice includes maximal resection through surgery, radiation, and adjuvant treatment. Because precise identification is crucial to diagnosis in clinical settings, medical imaging specialists confront challenges when identifying stroke lesions including brain malignancies. Plus, MR photos are scarce. Manually analyzing these photos is time-consuming and problematic. Maintaining a laser-like attention during classification, which increases false hits, is also difficult. For these reasons, CAD offers enormous potential for automated MR image analysis. We use Multi Texton Histogram (MTH) and Support Vector Machine (SVM) to analyze the structure of tumorous and normal tissues to improve brain tumor classification accuracy in T1 post-contrast MRI images. The suggested approach achieves the desired result via the following steps: HSD is used to extract features from tumor regions, combining the benefits of texton co-occurrence matrix and histogram. Hybrid kernel Support Vector Machine is used for final classification. The experimental findings are verified using K-fold cross validation. The suggested HSD+HKSVM model frequently has the greatest accuracy and recall, according to the two databases. The greatest outcomes are 99.24% accuracy and 100% recall using F(V2) feature vector.