A mathematical fuzzy fusion framework for whole tumor segmentation in multimodal MRI using Nakagami imaging

Automated lesion segmentation has become one of the most important tasks for the experts and researchers dealing with intelligent image processing of brain magnetic resonance images in clinical and bio-medicine. Fluid attenuated inversion recovery (FLAIR) is the universally accepted sequence for highlighting the lesions while suppressing the surrounding cerebrospinal fluid to create utmost contrast; however, in some cases, FLAIR images are not enough to cover the whole lesions in contrast-based segmentation. Therefore, in this paper, we propose a mathematical fuzzy inference-based fusion framework to increase the dice score coefficient (DSC) of the seg-mentation of FLAIR sequences and to achieve the highest DSC to overcome the inconclusive FLAIR cases. Taken from the BraTS2012 training database, the sample insufficient FLAIR images and the corresponding sequences T1, T1c, and T2 containing high-and low-grade glioma are processed by our fully-mathematical Nakagami imaging method and the lesions are separately segmented and stored. The binary images, generated by the specialized fuzzy c-means segmentation, are fused by a mathematical pixel intensity and distance-based fuzzy inference system to reach the ground truth images with the highest accuracy possible. The average dice score, calculated by segmentation of all images in the BraTS 2012 training database, is computed as 92.78% after fusion of all sequences. As a promising framework, the outputs of this research would be so beneficial for the experts dealing with whole tumor segmentation despite inconclusive FLAIR images.

Automated summary

In this paper, a mathematical fuzzy inference-based fusion framework is proposed to enhance the segmentation of FLAIR sequences and overcome the limitations of FLAIR images. The framework achieved a high dice score coefficient (DSC) and showed promising results for whole tumor segmentation.