Reconstructing lost BOLD signal in individual participants using deep machine learning.

Area of Science:

• Neuroimaging
• Machine Learning
• Computational Neuroscience

Background:

• Signal loss in blood oxygen level-dependent (BOLD) functional magnetic resonance imaging (fMRI) is a frequent issue.
• This signal degradation can lead to inaccurate interpretations of brain activity and connectivity.

Purpose of the Study:

• To develop and validate a deep machine learning approach for reconstructing compromised fMRI BOLD signals.
• To assess the model's ability to recover temporal information and functional connectivity from signal-compromised data.
• To determine if the reconstructed signals capture individual-specific brain organization.

Main Methods:

• A deep machine learning model was trained on BOLD activity principles from one dataset.
• The trained model was used to reconstruct artificially compromised regions in an independent dataset, frame by frame.
• The method was validated on healthy datasets and patient data with signal loss due to intracortical electrodes.

Main Results:

• Reconstructed BOLD time series showed significant correlation with original time series, despite frames lacking independent temporal information.
• Reconstructed functional connectivity maps closely corresponded to original maps, indicating recovery of inter-regional brain relationships.
• The reconstructions successfully captured individual-specific functional brain organization features.

Conclusions:

• Deep machine learning offers a powerful method for reconstructing compromised BOLD fMRI signals.
• This approach can restore meaningful temporal dynamics and functional connectivity, crucial for accurate neuroimaging analysis.
• The ability to preserve individual-specific brain organization highlights the potential of AI in advancing neuroimaging research and clinical applications.