Group-PCA for very large fMRI datasets.

Area of Science:

• Neuroscience
• Data Science
• Computational Biology

Background:

• Large-scale neuroimaging datasets, such as resting-state fMRI from the Human Connectome Project, present significant computational challenges for group-level analysis.
• Existing methods struggle with the memory demands and scalability required for analyzing aggregate data from numerous subjects.

Purpose of the Study:

• To introduce and evaluate two novel group-level principal component analysis (PCA) approaches for analyzing large-scale resting-state fMRI data.
• To provide memory-efficient alternatives that closely approximate PCA results from full data concatenation.

Main Methods:

• Development of two distinct group-PCA methodologies designed for high-dimensional neuroimaging data.
• Simulation studies were conducted to compare the accuracy and memory efficiency of the proposed methods against existing popular approaches for multi-subject resting-state fMRI analysis.

Main Results:

• Both proposed group-PCA methods demonstrated significantly lower memory requirements compared to traditional full data concatenation, irrespective of the number of datasets.
• In realistic simulations, the developed group-PCA approaches yielded more accurate results than commonly used methods for multi-subject resting-state fMRI studies.
• The output from group-PCA facilitates downstream analyses, including group-averaged connectivity estimation, parcellation, and group-independent component analysis (ICA).

Conclusions:

• The presented group-PCA methods offer a scalable and memory-efficient solution for analyzing large resting-state fMRI datasets.
• These approaches enhance the feasibility of advanced group-level neuroimaging analyses, improving the understanding of brain connectivity across populations.