Effective fuzzy c-means based kernel function in segmenting medical images

Automatic MR Brain Tumor Detection using Possibilistic C-Means and K-Means Clustering with Color Segmentation

“…Table 2 gives the segmentation accuracy of the two algorithms on brain images. The segmentation accuracy using c-means is 82% [23]. Comparing to recent results the proposed method gives better result.…”

Connectivity‐based parcellation: Critique and implications

“…Cluster validity criterion Rationale Used in previous brain parcellation studies voxel/vertex/node-wise cluster assignments) based on chi-square; put differently, it measures the strength of association between two clustering solutions Dice coefficient assesses the similarity between two samples or adjacency matrices; primarily practically justified rather than backed up theoretically; works well in heterogeneous and outlier-prone data; can be used to compare group and singlesubject clusterings; can be computed in different ways; it is equivalent to the Jaccard index because there is a monotonic transformation between their scores Blumensath et al, 2013;Craddock et al, 2012;Shen et al, 2013;Wang et al, 2015 Inter-versus intra-cluster distance ratio assesses cluster separation by the ratio between the average distance of a voxel/vertex/nodes to its cluster centre and the average distance between the cluster centres; a significantly increased ratio compared to the K-1 solution would indicate a better separation of the obtained clusters Bzdok et al, 2014;Chang et al, 2009 Percentage of misclassification assesses cluster assignment by the amount of noise and potentially local effects in the clustering; the average percentage of voxels/vertexs/nodes that were assigned to a different cluster compared to the most frequent assignment of these voxels/vertices/nodes; used to compared ways to compute a same or different number of clusters Bzdok et al, 2014 Percentage of parent-children congruency assesses cluster topology by how many voxels/vertices/nodes are not related to the dominant parent cluster compared to the solution with K -1 clusters; counts voxels/vertices/nodes that do not reflect a hierarchical organization; related to hierarchy index Clos et al, 2013;Eickhoff et al, in press;Kahnt et al, 2012 Silhouette coefficient assesses cluster separation by measuring how similar that voxel/vertex/node is to voxels/vertices/nodes in its own cluster compared to voxels/vertices/ nodes in the nearest cluster; good solutions are those with a higher silhouette value compared to the K-1 solution; this measure of cluster quality is independent of the number of clusters Bzdok et al, 2014;Craddock et al, 2012;Eickhoff et al, in press;Kannan et al, 2010;Zhang and Li, 2012 Variation of information (VI) assesses how much knowing the cluster assignment for an item in clustering X reduces the uncertainty about the item's cluster in clustering Y; a linear expression involving mutual information and entropy; is not adjusted for chance (contrary to AMI and ARI); used to compare ways to obtain a same or different number of clusters Bzdok et al, 2014;Clos et al, 2013;Eickhoff et al, in press;<...>…”

Segmentation of Brain MRI

“…This iteration will stop when , ε is a termination criterion between 0 and 1, and p is the iteration step (Kannan et al, 2010). Although clustering algorithms do not require training data, they do require an initial segmentation (or equivalently, initial parameters).…”