Accurate MRI brain tumor segmentation based on rotating triangular section with fuzzy C- means optimization

摘要

This paper proposes an accurate MRI brain tumor segmentation based on a Rotating Triangular Section with Fuzzy C-Means Optimization. Magnetic Resonance Imaging has become so popular due to its capability to differentiate tumors from the non-tumor region. The proposed method initially eliminates most of the background region by two level morphological reconstruction processes followed by thresholding. The two-level morphological reconstruction uses 'erosion' as the first level and 'dilation' as the second level. After eliminating the background, a region for Fuzzy C-Means (FCM) optimization is chosen using the Radius Contraction and Expansion process. The Radius Contraction and Expansion initially, selects the centroid and maximum radius of the region provided by the background elimination. The Radius Contraction and Expansion will give a contour whose shape is approximately the same as the shape of the tumor but larger than the size of the tumor region. The centroid of the new contour which acts as one of the vertices of the triangular region is again estimated. The remaining two vertex pixels are estimated from the contour pixels with a spacing provided by a spacing factor. FCM is then applied to this triangular region to obtain the accurate tumor pixels inside the triangular region. A new triangular region is estimated in the clockwise direction and FCM is again applied to the new triangular region. This process is repeated until the formation of the triangular region based FCM optimization completes one cycle. The performance of the proposed MRI brain tumor segmentation was evaluated using the T-1- weighted contrast-enhanced image dataset with the metrics such as dice score, sensitivity, specificity, Hausdorff Distance, and Probabilistic Rand Index (PRI). Experimental results reveal that the proposed MRI brain tumor segmentation outperforms the other state-of-the-art MRI brain tumor segmentation method.