Advantages of cluster analysis for multifunctional and intercrossing brain area distribution Evaluation by functional magnetic resonance imaging

Hongchang Zhai; Xudong Xu; Shengyong Xiao

摘要：BACKGROUND: Multiple linear regression, general linear test and calculation of correlation values are commonly used in studies of brain function using functional magnetic resonance imaging (fMRI). However, there are some limitations in their applications. In non-signal data statistics, cluster analysis functions as a very mature method, but it is not reliable in signal data statistics. OBJECTIVE: To investigate the spatial distribution of complex function in brain areas during motor tasks by cluster analysis, and to compare this with multiple linear regression. DESIGN, TIME AND SETTING: Block design, performed at the MR laboratory of Guangzhou University of Chinese Medicine. PARTICIPANTS: Fifteen right-handed, healthy university students (10 males and 5 females, aged 19-21 years).METHODS: fMRI was performed while the subjects performed a finger movement task with the right hand. The screen showed a gray hand, with red spots presented in a random order on one of the index, middle, ring and little fingers. The subjects were required to remember the sequence of the red spots on the display. After a delay of 14 seconds, the subjects tapped their fingers according to the order of the red spots, as soon as the red spots turned green. After an interval of 14 seconds, another sequence appeared. Every sequence lasted for 28 seconds, including preparation and execution phases. A total of nine sequences per subject were performed. The data were analyzed using deconvolution and cluster methods, and program "cluster" was used to statistically analyze the coordinate positions of deconvolution and cluster data. MAIN OUTCOME MEASURES: Brain activation maps by deconvolution and brain function maps by clustering of the maximum peak values; blood oxygenation level dependent curves by deconvolution; coordinates of peak values and activation volumes by the two methods. RESULTS: The deconvolution method could not integrate the brain activation maps during different tasks into one activation picture, which made it difficult to identify exactly the spatial distribution of various activities in certain brain areas. Cluster analysis, using maximum peak values, clearly showed brain areas of monofunction and multifunction, even complex function, and presented a clear spatial distribution of multiple functional movements. Using the command "clust" with the same parameters, the volumes of main brain activation areas were consistent, indicating the maximum peak values are completely reliable in contrast to the deconvolution method. CONCLUSION: Cluster analysis is conducive to the analysis of multifunctional complex areas. Clustering using maximum peak values is a reliable method. Brain areas, such as primary motor cortex, supplementary motor area, and posterior parietal cortex, are not monofunctional areas, but multifunctional, complex ones. The activation maps derived by deconvolution statistics are distributed in many maps, which are not convenient to determine the functional distribution of complex brain areas.