Modeling Cognitive Processes via Multi-stage Consistent Functional Response Detection

Abstract

Recent neuroscience research suggested that cognitive processes can be viewed as functional information flows on a complex neural network. However, computational modeling of cognitive processes based on fMRI data has been rarely explored so far due to two key challenges. First, there has been a lack of universal and individualized brain reference system, on which computational modeling of cognitive processes can be performed, integrated, and compared. Second, there has been a lack of ground-truth of cognitive processes. This paper presents a novel framework for computational modeling of working memory processes via a multi-stage consistent functional response detection. We deal with the above two challenges by using a publicly released large-scale cortical landmark system as a universal and individualized brain reference system and as a statistical data integration platform. Specifically, in the first-stage analysis, for each corresponding landmark we measure the consistency of its fMRI BOLD signals from a group of subjects, and the landmarks with high group-wise consistency are found to be highly task-related. In the second stage, the consistency of dynamic functional connection (DFC) time series of each landmark pair from the same group of subjects are measured, and those connections with high consistent patterns are declared as the active interactions during the cognitive task. Here, the group-wise consistent responses inferred from statistical pooling of data from multiple subjects via the universal brain reference system are considered as the benchmark to evaluate the multi-stage framework. Experimental results on working memory task fMRI data revealed that our methods can detect meaningful cognitive processes.