Interval type-2 fuzzy predicates for brain magnetic resonance image segmentation

Abstract

The analysis of structural changes in the brain through Magnetic Resonance Imaging (MRI) provides useful information for diagnosis and clinical treatment of patients with pathologies like Alzheimer disease and dementia. While complexity achieved by the MRI equipment is high, quantification of structures and tissues has not been entirely solved. In the present paper, MRI segmentation is discussed using a new classification method called Type-2 Label-based Fuzzy Predicate Classification (T2-LFPC). From labeled data (pixels of different tissues selected by medical experts) a random partition is defined and the obtained subsets are analyzed discovering groups with similar properties called class prototypes. Using theses prototypes, interval type-2 membership functions and fuzzy predicates are defined. Parameters regarding the fuzzy predicates are optimized. Fuzzy predicates are applied on unlabeled pixels performing the segmentation and volumes occupied for the tissues into the intracranial cavity are computed. Results are compared to those of known methods. A method of measuring the progressive atrophy and possible changes compared to a therapeutic effect should be essentially automatic and therefore independent of the radiologist. Results show that the performance of the proposed method is highly acceptable as a contribution for this requirement. Advantages of this approach are presented throughout this paper.

The analysis of structural changes in the brain through Magnetic Resonance Imaging (MRI) provides useful information for diagnosis and clinical treatment of patients with pathologies like Alzheimer disease and dementia. While complexity achieved by the MRI equipment is high, quantification of structures and tissues has not been entirely solved. In the present paper, MRI segmentation is discussed using a new classification method called Type-2 Label-based Fuzzy Predicate Classification (T2-LFPC). From labeled data (pixels of different tissues selected by medical experts) a random partition is defined and the obtained subsets are analyzed discovering groups with similar properties called class prototypes. Using theses prototypes, interval type-2 membership functions and fuzzy predicates are defined. Parameters regarding the fuzzy predicates are optimized. Fuzzy predicates are applied on unlabeled pixels performing the segmentation and volumes occupied for the tissues into the intracranial cavity are computed. Results are compared to those of known methods. A method of measuring the progressive atrophy and possible changes compared to a therapeutic effect should be essentially automatic and therefore independent of the radiologist. Results show that the performance of the proposed method is highly acceptable as a contribution for this requirement. Advantages of this approach are presented throughout this paper.