Level set method for automated 3D brain tumor segmentation using symmetry analysis and kernel induced fuzzy clustering

Automatic brain tumor segmentation in magnetic resonance images (MRIs) is an essential stage for treatment planning. However, MR image segmentation is challenging owing to non-uniformity in the intensity distribution, tumor shape, size, and location variation. The paper proposes a new level set method that is called Fuzzy Kernel Level Set (FKLS) for 3D brain tumor segmentation in MR images. To avoid computational complexity, fast bounding box based on symmetry analysis is used to extract the volume of interest (VOI) in brain MRIs. Then, a level set method is proposed based on fuzzy c-means clustering and kernel mapping. A kernel function is used to transfer the image into another domain, where the new proposed functional is minimized. To assess the proposed FKLS method, a synthetic image and natural brain MR images from BraTS 2017 are segmented. Experimental results show that our method is superior to the state-of-the-art segmentation methods regarding the segmentation accuracy based on Dice, Jaccard, Sensitivity, and Specificity metrics. The mean values of these metrics are 97.62% +/- (0.94%), 95.41% +/- (1.8%), 98.79% +/- (0.63%), and 99.85% +/- (0.09%), respectively.

Automated summary

This paper proposes a new level set method called Fuzzy Kernel Level Set (FKLS) for brain tumor segmentation in MRI images. The method uses fuzzy c-means clustering and kernel mapping to transfer the image and extract the volume of interest. Experimental results show that the method outperforms existing segmentation methods in terms of accuracy.