Deep Learning Spinal Cord Segmentation Based on B0 Reference for Diffusion Tensor Imaging Analysis in Cervical...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuroscience

Background:

• Diffusion Tensor Imaging (DTI) is vital for assessing Cervical Spondylotic Myelopathy (CSM).
• Manual segmentation of spinal cord DTI is labor-intensive and subjective.
• Current automatic DTI segmentation methods do not meet clinical needs.

Purpose of the Study:

• To develop an AI-driven segmentation method for spinal cord DTI.
• To improve the accuracy and efficiency of DTI analysis in CSM diagnosis.

Main Methods:

• A deep learning model, SCS-Net (Spinal Cord Segmentation Network), was developed using a U-shaped architecture.
• A lightweight feature extraction module was employed to address data scarcity.
• The model was trained and evaluated on DTI data from 89 CSM patients, utilizing B0 images as input features.

Main Results:

• SCS-Net achieved satisfactory accuracy in general segmentation metrics (precision, recall, Dice coefficient).
• The model demonstrated low error rates for DTI-specific feature indices (e.g., 5.32% on the left lateral side).
• The AI segmentation showed consistent performance, affirming radiological rationality.

Conclusions:

• The proposed AI-driven segmentation model significantly reduces reliance on manual DTI interpretation.
• SCS-Net offers a feasible solution to enhance diagnostic outcomes for CSM patients.
• This advancement supports more objective and efficient DTI-assisted diagnosis.