DGPC-Net: Dual Branch Gaussian Process Constrained Network for Spinal Cord Stimulation Effect Prediction

Area of Science:

• Biomedical Engineering
• Computational Neuroscience
• Machine Learning

Background:

• Spinal cord stimulation (SCS) is a therapeutic approach for neurological conditions.
• Optimizing SCS parameters for individual patients is crucial but time-consuming.
• Predicting muscle activation is key to effective SCS.

Purpose of the Study:

• To introduce DGPC-Net, a novel dual-branch Gaussian Process Constrained Network.
• To predict muscle activation intensity in response to SCS.
• To improve the efficiency of SCS parameter optimization.

Main Methods:

• Developed DGPC-Net, a dual-branch Gaussian Process Constrained Network.
• Utilized MRI scans, finite element, and NEURON simulations for personalized virtual environments.
• Generated simulated muscle activation data for epidural electrical stimulation (EES).

Main Results:

• DGPC-Net achieved 81.42% balanced accuracy in binary muscle activation classification.
• Demonstrated a strong correlation of 0.93 between predicted and ground-truth muscle activation intensities.
• Ablation studies validated the necessity of each network component.

Conclusions:

• DGPC-Net accurately predicts muscle activation intensity from SCS.
• Real-time prediction capability can significantly reduce SCS parameter optimization time.
• This approach holds potential for improving treatment efficiency for spinal cord injury and Parkinson's disease gait issues.