Modified total Bregman divergence driven picture fuzzy clustering with local information for brain MRI image segmentation

Abstract

This research work discusses a noise-robust picture fuzzy clustering method with an application to the MRI image segmentation problem. The MRI images suffer from the problem of noise and non-linear structures. Although there are many variants of fuzzy set theory and intuitionistic fuzzy set theory-based clustering approaches to handle the problem of noise and non-linearity present in the image during the segmentation process, still they lack in achieving accurate segmentation. To overcome this problem, we have suggested using the picture fuzzy set theoretic approach which enhances the representational capability of the data and helps in handling the non-linear structures present in the image, and in our proposed work, the picture fuzzy Euclidean distance is replaced with modified picture fuzzy total Bregman divergence using the spatial neighborhood information around the sample. Squared Euclidean distance is itself a special case of Bregman Divergence, hence Bregman divergence helps to explore the details in a better way thus resulting in noise suppression. Furthermore, the proposed algorithm proves to be robust in preserving the image details using picture fuzzy set theory and is free from any parameter selection due to the incorporation of a local information factor. The algorithm was performed on a synthetic image corrupted with “Gaussian”, “salt and pepper”, “mixture of Gaussian and salt and pepper noise” of different intensities, “Brainweb” datasets corrupted with noise, six real “IBSR” datasets along with seven “MRbrainS18” datasets. Performance measures used were partition coefficient, partition entropy, dice score (DS), average segmentation accuracy (ASA), and XB index, and the proposed method was robust when compared to a series of algorithms stated in the literature.