Extraction of the cardiac waveform from simultaneous multislice fMRI data using slice sorted averaging and a deep...

Area of Science:

• Neuroimaging
• Biomedical Engineering
• Signal Processing

Background:

• Cardiac pulsation significantly contaminates blood-oxygen-level-dependent (BOLD) functional magnetic resonance imaging (fMRI) signals, particularly near blood vessels.
• Traditional fMRI temporal resolution is insufficient to avoid cardiac aliasing, preventing direct cardiac signal removal via spectral filtering.
• Existing modeling methods and visualization techniques for cardiac contamination require physiological data not commonly available in large fMRI databases.

Purpose of the Study:

• To present a novel method for estimating the cardiac waveform directly from multislice fMRI data.
• To enable accurate cardiac signal removal without requiring additional physiological measurements.
• To improve the analysis of BOLD fMRI data contaminated by cardiac activity.

Main Methods:

• Exploited the temporal structure of cardiac contamination in multislice fMRI acquisitions.
• Assumed the cardiac signal is pseudoperiodic, coherent within slices, and has a consistent shape across the brain.
• Developed a deep learning filter to enhance the estimation of the cardiac waveform from fMRI data alone.

Main Results:

• Successfully estimated the cardiac waveform directly from fMRI data without external sensors.
• Demonstrated the potential to significantly improve BOLD signal analysis by removing cardiac contamination.
• The method leverages the rapid slice acquisition in multislice imaging to capture cardiac phase information.

Conclusions:

• A new, non-invasive method effectively estimates cardiac waveforms from fMRI data.
• This technique addresses a major challenge in fMRI analysis, particularly for existing datasets lacking physiological recordings.
• The developed deep learning approach offers a powerful tool for enhancing the quality and interpretability of fMRI studies.