Walsh Hadamard kernel‐based texture feature for multimodal MRI brain tumour segmentation

Abstract

AbstractThe automated brain tumor segmentation methods are challenging due to the diverse nature of tumors. Recently, the graph based Spectral Clustering (SC) method is utilized for brain tumor segmentation to create high‐quality clusters. In this article, a new superpixel based SC using the Walsh Hadamard texture feature for multimodal brain tumor segmentation from Magnetic Resonance Image is proposed. The selected kernels of Walsh Hadamard Transform (WHT) are projected on equal size blocks of the image for texture feature extraction. The texture feature strength of each block is considered as superpixels, and these superpixels become nodes in the graph of SC. Finally, the original members of superpixels are recovered to represent Complete Tumor (CT), Tumor Core (TC), and Enhancing Tumor (ET) tissues. The observational results are brought out on BRATS 2015 data sets and evaluated using the Dice Score (DS), Hausdorff Distance, and Volumetric Difference metrics. The proposed method has produced competitive results with a DS of 0.83 for CT, 0.75 for TC, and 0.73 for ET, respectively, for high‐grade images. In case of low‐grade images, the proposed method achieves DS of 0.78 for CT, 0.68 for TC, and 0.60 for ET, respectively. The proposed method produces results better than other existing clustering methods.