Type-2 Fuzzy Soft Learning Algorithms for MR Image Segmentation

The integration of the merits of the fuzzy set and neural network called neuro-fuzzy approach has attracted considerable attention in segmentation of the magnetic resonance imaging (MRI). In this paper, we extend the membership values of each pattern in the well known fuzzy soft learning segmentation techniques viz., fuzzy soft Kohonen’s competitive learning (FSKCL) and fuzzy soft learning vector quantization (FSLVQ) to the type-2 membership functions and two algorithms viz., type-2 FSKCL (T2FSKCL) and type-2 FSLVQ (T2FSLVQ) respectively have been proposed. The T2FSKCL and T2FSLVQ are the alternative optimal versions of the fuzzy soft learning methods (FSKCL and FSLVQ respectively) based on the type-2 membership function. The proposed algorithms are applied on several magnetic resonance brain images and are compared with the existing fuzzy soft learning algorithms viz., FSKCL and FSLVQ. Comparisons are based on the quality of the MRI segmentation results and the optimal value of the objective function. Superiority of the proposed algorithms over the existing algorithms viz., fuzzy c- means (FCM), FSKCL and FSLVQ clustering algorithms are demonstrated quantitatively and it is found that the segmented results of the T2FSKCL and the T2FSLVQ obtained better detection of abnormal tissues with optimal objective function value.