Efficient Brain Tumor Segmentation using Kernel Representation

The segmentation of a brain tumour is essential for diagnosis, treatment planning, and surgical simulation. A brain tumor’s location, size, and shape can all be better understood by clinicians with the use of precise segmentation. This study proposes kernel sparse coding for fully automatic segmentation of brain tumours. It is validated using 3D multimodal Magnetic Resonance Imaging (MRI). This method uses kernel dictionary learning to generate five adaptive dictionaries for healthy tissues, necrosis, edoema, non-enhancing tumour, and enhancing tumour tissues by extracting the nonlinear properties from pre-processed MRI images. Interest in automatically segmenting tumour regions from Magnetic Resonance (MR) images of brain tumours has grown because of the demand for quantitative assessment and three-dimensional) visualization of brain tumours. A representation model based on the joint constraints of kernel and sparsity is presented to mine the features and structural prior knowledge of brain tumour pictures in the spectral kernel space considering the uneven grey distribution of MR images and the fuzzy boundaries of brain tumours. Additionally, by creating an ideal kernel and sparsity based on uniform regions on MRI images will give optimum results. Using a kernel-clustering method based on dictionary learning, the voxels are coded. The experimental outcomes show that the segmentation accuracy of the suggested strategy is better than several other methods under a variety of indicators and that the kernel low-rank model’s sparsity constraint, which is incorporated into the kernel space, has certain influence on maintaining the regional structure and particulars of brain tumours.