Novel fuzzy clustering‐based bias field correction technique for brain magnetic resonance images. Issue 9 (15th June 2020)

• Record Type: Journal Article
• Authors: Mishro, Pranaba K.; Agrawal, Sanjay; Panda, Rutuparna; Abraham, Ajith
• Is Part Of: IET image processing. Volume 14:Issue 9(2020)
• Page Start: 1929
• Page End: 1936
• Publication Date: 2020-06-15
• Subjects: medical image processing -- fuzzy set theory -- brain -- biological tissues -- pattern clustering -- statistical analysis -- image segmentation -- biomedical MRI -- image classification; segmentation accuracy -- spatial information -- intensity inhomogeneity -- equidistant pixels -- single cluster -- brain MR images -- brain magnetic resonance images -- preprocessing requirement -- brain tissue segmentation task -- authentic brain tissue regions -- classification -- poor resolution magnetic resonance image -- intensity distribution -- fuzzy clustering‐based bias field correction technique -- acquisition procedure -- standard fuzzy C‐means clustering -- statistical analysis; Image processing -- Periodicals; 621.36705
• DOI: 10.1049/iet-ipr.2019.0942
• Languages: English
• ISSNs: 1751-9659

Abstract

Abstract : Bias field correction is an essential pre‐processing requirement for brain tissue segmentation task. Authentic brain tissue regions are highly useful for classification and detection of abnormalities. A poor resolution magnetic resonance (MR) image is produced with irregularities in structure, abnormalities in the intensity distribution and noise during the acquisition procedure. The existing bias field correction methods do not consider the spatial information. Further, the problem of equidistant pixels while clustering is not addressed. These problems lead to poor segmentation accuracy. To solve these problems, the authors suggest a novel biased fuzzy clustering technique for the problem on hand. The basic idea is to incorporate the spatial information by altering the membership matrix of standard fuzzy C‐means clustering to lower the effect of noise and intensity inhomogeneity. It also helps in improving the segmentation accuracies of the tissue regions by assigning the equidistant pixels to a single cluster. The suggested technique is validated with different modalities of brain MR images. Various evaluation indices are computed followed by the statistical analysis to justify the superiority of the suggested technique in comparison to the state‐of‐the‐art methods.