Comparison of Four Hierarchical and Sequential Clustering Methods for Analysis of DTI Data of Temporal Lobe Epilepsy

Diffusion tensor imaging (DTI) provides useful information about the anatomy of the brain white matter. This information includes the shape and geometry of the fiber bundles. Temporal Lobe Epilepsy (TLE) is a neurologic disease that damages some fiber bundles in the brain, like fornix. The information in DTI data can be presented by diffusion anisotropy indices. In this paper, Ellipsoidal Area Ratio (EAR) is used as an anisotropy index for extracting the arc length function for each subject. The mean value and the norm value of these arc length functions are used as features for clustering of the data. Four data clustering techniques: Hierarchical Cluster Analysis (HCA); Fuzzy C-Means (FCM) clustering; kmeans clustering; and information-theoretic clustering are used. The subjects are 12 normal control and 19 patients with temporal lobe epilepsy. Decrease of the EAR is found in the TLE group. The performance of the FCM and k-means is similar while information theoretic clustering creates more compact clusters. In comparison, FCM, k-means, and information theoretic clustering have better results than the HCA.

کلیدواژه ها:

Diffusion Tensor Imaging (DTI) ، Temporal LobeEpilepsy (TLE) ، Fuzzy C-Means clustering (FCM) ، ellipsoidalarea ratio (EAR) ، hierarchical cluster analysis (HCA) ، k-means ، information-theoretic clustering