DeepRetroMoCo: deep neural network-based retrospective motion correction algorithm for spinal cord functional MRI.

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Machine Learning

Background:

• Spinal cord fMRI preprocessing faces challenges with motion artifacts.
• Existing motion correction algorithms for brain are not optimal for spinal cord.
• Developing effective motion correction is crucial for spinal cord fMRI data quality.

Purpose of the Study:

• To introduce DeepRetroMoCo, a deep learning model for motion artifact detection and correction in spinal cord fMRI.
• To evaluate the performance of DeepRetroMoCo against established methods.

Main Methods:

• A convolutional neural network (CNN) named DeepRetroMoCo was developed using unsupervised learning.
• The model was trained and tested on axial T2*-weighted spinal cord fMRI data from 27 participants.
• Performance was compared to the sct_fmri_moco method using tSNR and DVARS metrics.

Main Results:

• DeepRetroMoCo achieved significantly higher average temporal signal-to-noise ratio (tSNR) in the cervical cord.
• Lower Delta Variation Signal (DVARS) values were observed with DeepRetroMoCo, indicating better motion artifact reduction.
• DeepRetroMoCo demonstrated a significantly faster processing time.

Conclusions:

• DeepRetroMoCo offers a substantial improvement for motion correction in cervical spinal cord fMRI.
• The deep learning approach provides enhanced performance and a promising solution for motion artifacts.
• This method advances the quality of spinal cord fMRI analysis.