Poster evaluation of the level set method by using similarity criterion (KCC) for clustering and analysis of functional MRI data

This paper aims at evaluating the fMRI analysis and data in order to detect the active regions of brain. We present a framework based on clustering by level set technique. The main idea in this approach is that voxels, in the active region, have similar time behavior. To detect the level of similarity between time series of neighboring voxels the Kendall's coefficient concordance (KCC) is used, which is the cause of the level set formulation speed function. Then a two-dimensional curve is defined on the surface, which is in accordance with the speed, forward evolution and propagation. The results of applying the level set method according to real and simulated data were compared according to General Linear Model (GLM) and analysis based on Multiple comparison error by Family Wise Error(FWE) method. one of the benefits of the level set method is that it does not need to detect preliminary clusters and also this method is expressed in a non-parametrical form and is flexible in changes of contours topology and has stable and appropriate results as well as the possibility of being developed from 2D to 3 dimensions. The study of the results and findings illustrates that utilizing both time and spatial data simultaneously provides better segmentation, than the Voxel-wise technique does. And the error rate, with the FWE method in high-noise images, has had a 13% decrease. It also exhibits higher stability against image noises. In this method, accuracy and correctness, compared with the GLM method, have risen by 8% and therefore; it can be regarded as an appropriate way to analyze the noisy data of fMRI.