Currently, the radiologist needs to distinguish the medical imaging with their multiple classes. In this paper, we work on several steps: segmented ROI, feature extraction of ROI and classification. In this work, we proposed a multiclass kernel based Hellinger decision method HD-Tree and HD-Forest for the classification of brain tumor classes with respect to classification time and accuracy. The calculated features like patient symptoms, centroid, shape, etc. are used in the classification scheme. Total 97 MRI brain tumor images (Astrocytoma (22), Ganglioglioma (6), Glioblastoma (23), Epidermoid (3), Mixed Glioma (5) and Meningnet (41)) were used for the experiment. The Experimental result shows that kernel-based Hellinger HD-Tree was found to be 96.50 % of accuracy and HD-Forest was found to be 99.9%. In this paper, we compare our proposed method LA-SVM method. LA-SVM was found to be 96% of accuracy. We can see that HD-forest gives the best accuracy result.