BRAIN TUMOUR SEGMENTATION AND GRADE CLASSIFICATION USING MODIFIED VGG19 WITH SUPPORT VECTOR MACHINE

A brain tumour is an abnormal cell growth that occurs either inside or outside the brain. Brain tumours are among the worst and most dangerous diseases in medicine. Convolutional neural networks produce the greatest results for tumour segmentation these days. BraTS2020 and real-time hospital images were employed in this method. Pre-processing, segmentation, and post-processing are the three phases of this methodology. During the pre-processing phase, Digital imaging & communications in medicine (DICOM) images converted to Neuroimaging informatics technology initiative (NIFTI), the images are resized, noise is removed, skulls removed process are done. During the segmentation phase, the Local binary pattern (LBP) features are identified and the modified VGG19 (M-VGG19) is employed to segment the tumour. The suggested approach obtained Dice similarity coefficient (DSC) values of 0.9415, 0.8898, 0.8862, and Hausdorff distance - 95th percentile (HD95) values of 5.65, 11.11, and 11.88 for whole tumour (WT), tumour core (TC), and enhanced tumour (ET) respectively, during the segmentation process. The usage of Graphics processing unit (GPU) reduced the training and testing computation time for segmentation and reached the speedup folds up to 3× compared with CPU. Tumour volumes are computed at the post-processing phase. Subsequently, this technique classified tumours into High-grade glioma (HGG) and Low-grade glioma (LGG) using Support vector machine (SVM). The BraTS dataset was trained using real-time hospital pictures, and this method got 85.71 classification accuracy.