DeepCor: denoising fMRI data with contrastive autoencoders.

Area of Science:

• Neuroimaging
• Machine Learning
• Signal Processing

Background:

• Functional magnetic resonance imaging (fMRI) offers noninvasive neural activity measurement with high spatial resolution.
• fMRI data quality is often compromised by significant noise, limiting analysis accuracy.
• Existing denoising methods struggle to effectively isolate and remove noise from fMRI signals.

Purpose of the Study:

• To introduce and evaluate DeepCor, a novel denoising method for fMRI data.
• To assess DeepCor's ability to disentangle and remove noise using deep generative models.
• To compare DeepCor's performance against state-of-the-art denoising techniques.

Main Methods:

• Development of DeepCor, a deep generative model-based denoising approach.
• Application of DeepCor to simulated and real fMRI datasets.
• Comparative analysis of DeepCor against established methods like CompCor.

Main Results:

• DeepCor demonstrated superior performance in denoising across various simulated fMRI datasets.
• The method successfully enhanced Blood-Oxygen-Level-Dependent (BOLD) signal responses in real fMRI data.
• DeepCor achieved a 215% improvement in BOLD signal enhancement compared to CompCor for face stimuli.

Conclusions:

• DeepCor is a highly effective method for denoising fMRI data, applicable to single-participant analyses.
• The deep generative model approach significantly improves the quality of fMRI data by reducing noise.
• DeepCor offers a substantial advancement in neuroimaging analysis, enabling more accurate measurement of neural activity.