Denoise Functional Magnetic Resonance Imaging With Random Matrix Theory Based Principal Component Analysis.

Area of Science:

• Neuroimaging
• Biophysics
• Data Science

Background:

• High-resolution functional MRI (fMRI) is significantly impacted by random thermal noise, limiting its effectiveness.
• Existing denoising techniques for fMRI data lack a consensus on optimal implementation strategies, especially when using Random Matrix Theory (RMT)-based principal component analysis (PCA).
• Low signal-to-noise ratio (SNR) in fMRI data (<5) presents a substantial challenge for accurate analysis.

Purpose of the Study:

• To propose and validate a comprehensive RMT-based denoising method for high-resolution fMRI data.
• To improve the performance of fMRI denoising, particularly for low SNR data, by addressing limitations in current methods.
• To enhance image restoration quality and functional sensitivity in fMRI analysis.

Main Methods:

• Developed a denoising method integrating RMT-based PCA with novel multiple criteria for rank and noise estimation.
• Implemented optimal singular value shrinkage based on RMT principles.
• Incorporated a variance stabilizing approach to effectively handle low SNR fMRI data.

Main Results:

• The proposed method demonstrated superior performance compared to state-of-the-art techniques, especially for fMRI data with SNR < 5.
• Simulation and in-vivo results showed significant improvements in image restoration quality.
• The method enhanced functional sensitivity without increasing functional mapping blurring.

Conclusions:

• The developed RMT-based denoising method effectively reduces thermal noise in high-resolution fMRI.
• This approach offers a robust solution for denoising low SNR fMRI data and can be seamlessly integrated into existing preprocessing pipelines.
• The method holds significant potential for advancing high-quality, high-resolution task fMRI in neuroscience and clinical applications.