GLMdenoise: a fast, automated technique for denoising task-based fMRI data.

Area of Science:

• Neuroimaging
• Cognitive Neuroscience
• Biomedical Engineering

Background:

• Task-based functional magnetic resonance imaging (fMRI) is crucial for measuring brain activity during specific tasks.
• Noise in fMRI data often limits the accurate detection of task-related signals.
• Improving the signal-to-noise ratio (SNR) is essential for robust fMRI analysis.

Purpose of the Study:

• To introduce GLMdenoise, a novel technique for enhancing SNR in fMRI data by incorporating noise regressors into General Linear Model (GLM) analysis.
• To demonstrate the effectiveness of GLMdenoise in improving the accuracy of GLM estimates and SNR across various event-related fMRI datasets.
• To provide a practical implementation of GLMdenoise and a benchmark for evaluating denoising methods.

Main Methods:

• GLMdenoise utilizes noise regressors derived from principal components analysis (PCA) of voxels unrelated to the experimental paradigm.
• Cross-validation is employed to select the optimal number of principal components for noise regression.
• The method is particularly suited for datasets with multiple runs, leveraging data resampling.

Main Results:

• GLMdenoise consistently improved cross-validation accuracy of GLM estimates on diverse event-related fMRI datasets.
• Substantial gains in signal-to-noise ratio (SNR) were observed with the application of GLMdenoise.
• In benchmark comparisons, GLMdenoise outperformed other tested denoising methods, including motion regression, ICA-based denoising, and RETROICOR/RVHRCOR.

Conclusions:

• GLMdenoise offers a significant advancement in fMRI data processing, effectively reducing noise and enhancing signal detection.
• The proposed method provides a practical solution for researchers seeking to improve the quality of task-based fMRI analyses.
• The Denoise Benchmark (DNB) facilitates objective comparison and validation of various fMRI denoising techniques.