Abstract
In the analysis of fMRI, the canonical hemodynamic response model (canonical HRM) based on the hypothesis of the general linear model (GLM) is widely used. However, the canonical HRM, which used the fixed response parameters of onset latency, time-to-peak and duration, limits the flexibility of statistical analysis and results in the reduction of sensitivity for the brain activity detection. In this work, we introduced a method for estimating the subject-specific hemodynamic response without a prior paradigm. First, according to the spatial temporal clustering analysis (STCA), the clusters of potential activated region were selected. Second, each cluster was purified iteratively to increase the concordance of the time courses. Finally, the specific estimated response was computed from the time courses of the cluster being activated. The results show that the response shape and the time parameters of specific response across subjects were different, and the onset latency of motor stimulation of canonical HRM was slower than the specific HRM one about 1 to 2 TR. In addition, the activation maps presented more significant statistical results for the activated voxels and the T value. We conclude that the specific HRM is adaptive to signal variety due to subjects even if the temporal information of stimulation is unknown, and it enhances the sensitivity of the brain activation detection in contrast to the canonical HRM.