Fast and robust spatial fuzzy bounded k-plane clustering method for human brain MRI image segmentation

Abstract

Fuzzy k-plane clustering (FkPC) is a soft plane-based clustering that efficiently clusters non-spherically distributed data. However, the FkPC method is sensitive to noise and provides unbounded cluster planes. To overcome these two limitations, we propose two modifications in the conventional FkPC method, referred as fuzzy bounded k-plane clustering method with local spatial information (FBkPC_S1). We introduce FCM objective function to bound the cluster planes and local spatial information in the objective function of FkPC to handle noise. The proposed FBkPC_S1 clustering method is fast and robust as it produces bounded cluster planes and can provide accurate segmentation in presence of noise. To show the effectiveness of the proposed FBkPC_S1 method, extensive experiments are performed on one synthetic image dataset and three publicly available human brain MRI datasets. The performance of the proposed FBkPC_S1 method is compared with 19 related methods in terms of average segmentation accuracy and Dice score. The proposed method achieves 91%, 65% and 75% average segmentation accuracy in the presence of noise artifacts on BrainWeb, IBSR and MRBrainS18 MRI datasets, respectively. Experimental results and statistical test demonstrate superior performance of the proposed FBkPC_S1 method in comparison to related methods.