EPISeg: Automated segmentation of the spinal cord on echo planar images using open-access multi-center data

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Computational Neuroscience

Background:

• Spinal cord functional magnetic resonance imaging (fMRI) is crucial for understanding sensation, movement, and autonomic functions.
• Preprocessing spinal cord fMRI data requires accurate segmentation of the spinal cord from gradient-echo echo planar imaging (EPI) data.
• Existing automated segmentation techniques struggle with low resolution, artifacts, and distortions common in spinal cord EPI, necessitating significant manual correction.

Purpose of the Study:

• To address the challenges in automated spinal cord segmentation for gradient-echo EPI data.
• To develop and validate a deep learning-based model for accurate spinal cord segmentation.
• To create a publicly available dataset and share code for reproducible research.

Main Methods:

• A multi-center dataset of spinal cord gradient-echo EPI images with ground-truth segmentations was compiled and shared on OpenNeuro.
• A novel deep learning model, EPISeg, was developed for automatic segmentation of the spinal cord on gradient-echo EPI data.
• The EPISeg model was integrated into the Spinal Cord Toolbox as a command-line tool.

Main Results:

• EPISeg demonstrated significant improvements in segmentation quality compared to existing spinal cord segmentation models.
• The model proved resilient to variations in acquisition protocols and common artifacts found in fMRI data.
• A multi-center dataset and training code were made publicly available to facilitate further research.

Conclusions:

• The developed EPISeg model offers a robust and accurate solution for automated spinal cord segmentation in gradient-echo EPI data.
• This advancement reduces manual labor and user bias, enabling more efficient and reliable spinal cord fMRI analysis.
• The availability of the dataset and code promotes wider adoption and further development in the field.