The present work introduces a novel segmentation approach for detection of brain tumor in presence of surrounding obscured tissues. In this view, kernel-based fuzzy clustering algorithm is employed to capture the clear boundary of the tumors. Proposed method also considers two significant features of brain MRI for segmentation; one is regional entropy and the other regional brightness. The most important issue of fuzzy clustering algorithm is the selection of optimal number of clusters prior to the clustering. This work determines the optimal cluster number by introducing the concept of cluster validity indices. Employing five different cluster validity indices, the optimal cluster number is obtained for both of the features. Then, these two features are integrated using principal component analysis method. Following this, shape characteristics of the segmented tumors are extracted for grading the benignancy/malignancy of the tumors. Finally, the superiority of the proposed segmentation approach is compared with similar research works in this field and its efficiency is studied in terms of the classification indices.