Research Article
Select and Cluster: A Method for Finding Functional Networks of Clustered Voxels in fMRI
Table 1
Performance on a reference hybrid dataset with increasing CNR (h4 = 1.33; h5 = 1.66; h6 = 2), of the clustering methods used for fMRI data. The value of the weighted Jaccard coefficient is reported for the cluster with maximal number of True Positive voxels found by each method. Numbers in table (the highest the better) measure the number of voxels initially belonging to true cluster (artificially defined) which are put in the same cluster by each compared method. Since Support Vector Clustering resulted in a number of clusters ≤ 10 in each dataset, its performance is compared to other methods performance with the number of clusters initialized to 10. For nonhierarchical methods, clustering solutions depend on initialization conditions, so the mean performance across 50 different initialization conditions is reported. Note that the higher the performance score, the better the quality of the activation cluster. SVC method shows a stable performance across increasing levels of noise and performs better than other methods, partially adapted from Dimitriadou et al. 2004 [28].
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||