Segmentation of Brain Tissues in MRI Images Using Improved Gustafson-Kessel Clustering Algorithm

Magnetic Resonance Imaging (MRI) often suffers from noise and intensity non-uniformity (INU), making image segmentation a challenging task. The fuzzy c-means (FCM) algorithm is a popular clustering-based method for image segmentation, but it is sensitive to noise, and its convergence rate is influenced by data distribution. Traditional FCM approaches use the Euclidean distance, which does not account for the variation in data point distances within similar and compact clusters. To address these limitations, a new cost function is proposed for the Gustafson-Kessel (GK) clustering algorithm that utilizes the Mahalanobis distance for similarity measurement. This approach enhances robustness to noise and INU conditions compared to other FCM-based methods. Additionally, information theory is incorporated to further improve noise robustness and segmentation accuracy. Conventional GK clustering is typically affected by the choice of fuzziness value; hence, a membership function entropy is introduced to mitigate this issue. In the proposed algorithm, morphological reconstruction is used as a pre-processing step to reduce noise while preserving object contours. Experimental results on various MRI datasets demonstrate that the proposed algorithm performs well in segmenting multiple tissues under different noise and INU conditions.