Adaptive integration of local region information to detect fine-scale brain activity patterns

With the rapid development of functional magnetic resonance imaging (fMRI) technology, the spatial resolution of fMRI data is continuously growing. This provides us the possibility to detect the fine-scale patterns of brain activities. The established univariate and multivariate methods to analyze fMRI data mostly focus on detecting the activation blobs without considering the distributed fine-scale patterns within the blobs. To improve the sensitivity of the activation detection, in this paper, multivariate statistical method and univariate statistical method are combined to discover the fine-grained activity patterns. For one voxel in the brain, a local homogenous region is constructed. Then, time courses from the local homogenous region are integrated with multivariate statistical method. Univariate statistical method is finally used to construct the interests of statistic for that voxel. The approach has explicitly taken into account the structures of both activity patterns and existing noise of local brain regions. Therefore, it could highlight the fine-scale activity patterns of the local regions. Experiments with simulated and real fMRI data demonstrate that the proposed method dramatically increases the sensitivity of detection of fine-scale brain activity patterns which contain the subtle information about experimental conditions.