Extraction of group activation features for different sleep stages from whole-brain fMRI data using Tucker decomposition

Hua-Zhe Qi, Yue Han, Bin-Hua Zhao, Qiu-Hua Lin

TL;DR: We used Tucker-2 decomposition to extract group activation features from multi-subject fMRI data across W,N1and N2 sleep stages.

Abstract: Functional magnetic resonance imaging (fMRI) provides a window for studying brain function with high spatial resolution, enabling the investigation of additional sleep-stage features beyond electroencephalography to solve the sleeping problem. However, previous studies mostly focused on analyses of fMRI data from regions of interest at single-subject level, resulting in a lack of group features. In this study, we propose to analyze whole-brain fMRI data at multi-subject level to extract group sleep-stage activation features. More precisely, a Tucker-2 decomposition algorithm is used to analyze multi-subject fMRI data collected during different sleep stages, since this algorithm simultaneously extracts group shared and individual spatial and temporal features from multi-subject fMRI data, minimizing the mixing of group and individual features. For fMRI data of different sleep stages, we extract components of interest such as the default mode network and auditory network, examine activations of group shared spatial maps, analyze frequency fluctuations of group shared time courses, and detect significant voxel-level differences in different sleep stages using individual spatial maps. We perform experiments using fMRI data from ten subjects with available non-rapid eye movement stage 1 and 2 and the wakefulness stage. The results show new findings, e.g., most components exhibit low-frequency oscillations during both wakefulness and sleep stages. However, in some networks, high-frequency signals appear in non-rapid eye movement stages, such as the default mode network and auditory network in non-rapid eye movement stage 1. Therefore, our study provides new evidence for analyzing sleep stage features.

Submission Number: 32