Question about DynamicBC

Submitted by Chen0075 on Thu, 05/05/2016 - 11:45

A new feature "clustering" was added into the new vertion of "DynamicBC" tool, but what's the processes (principle) in your "clustering". In addtion, by your tool, can I obtain the "dwell time" in each state that is regared as the number of windows belonging to the special state related cluster? and the "transition times (numbers) between states"? how can I display these results in the matlab. thanks!

Submitted by Wei Liao on Tue, 05/10/2016 - 23:28

The cluster analysis in DynamicBC toolbox was a prelimeinary attempt, and it would be not well advised. Thank you for your good question. The principle of k-means is very simple. We just reshape the data into 2-dimension matrix (voxel/roi * time-window) and concatenate it subject by subject. Then we use kmean function in Matlab to generate k clusters in group level. After that, we use kmean to generate k clusters in every subject. Pearson correlation is used to re-order every subject's k clusters to match the order of group level. Well, the labels of every time-window map/matrix or the "dwell time" has been generated in the function of kmean. Unfortunately, we didn't save it out in the function. You could add this line of code, 'save([outputd,'mat_',CluMet,'_Kmeans_',num2str(k),'_Order.mat'],'IDX_all','IDX_subj','IDX_subjre');', to the end of the function "dynamicBC_clustermatrix_beta.m"/"dynamicBC_clustermaps_beta.m". The variable IDX_all would give us the distribution of k clusters in the whole maps/matrices. The variable IDX_subj is a 2-dimension matrix (subj num * time-window) The variable IDX_subjre is a 2-dimension matrix (subj num * time-window), which represents the re-ordered lables sorted according the group level by using the pearson correlation. The new version would come soon and it would fix the problem. Hope it could help you. If you have other questions or good ideas, please contract us. It would help us to make the toolbox better.