Exploring Functional Difference Between Gyri and Sulci via Region-Specific 1D Convolutional Neural Networks

Abstract

The cerebral cortex is highly folded as convex gyri and concave sulci. Accumulating evidence has consistently suggested the morphological, structural, and functional differences between gyri and sulci, which are further supported by recent studies adopting deep learning methodologies. For instance, one of the pioneering studies demonstrated the intrinsic functional difference of neural activities between gyri and sulci by means of a convolutional neural network (CNN) based classifier on fMRI BOLD signals. While those studies revealed the holistic gyro-sulcal neural activity difference in the whole-brain scale, the characteristics of such gyro-sulcal difference within different brain regions, which account for specific brain functions, remains to be explored. In this study, we designed a region-specific one-dimensional (1D) CNN based classifier in order to differentiate gyro-sulcal resting state fMRI signals within each brain region. Time-frequency analysis was further performed on the learned 1D-CNN model to characterize the gyro-sulcal neural activity difference in different frequency scales of each brain region. Experiments results based on 900 subjects across 4 repeated resting-state fMRI scans from Human Connectome Project consistently showed that the gyral and sulcal signals could be differentiated within a majority of regions. Moreover, the gyral and sulcal filters exhibited different frequency characteristics in different scales across brain regions, suggesting that gyri and sulci may play different functional roles for different brain functions. To our best knowledge, this study provided one of the earliest mapping of the functional segregation of gyri/sulci for different brain regions, which helps better understand brain function mechanism.