Presentation description

Significant motor impairments can result from neural injuries such as spinal cord injury (SCI) and stroke. Motor neuro-prostheses can significantly enhance rehabilitation outcomes, promoting independence and mobility. While spinal cord stimulation (SCS) is used for chronic pain management, it also has potential for motor function recovery. Accurate electrode placement is crucial because it influences SCS effectiveness. Traditionally, electrode placement is manually confirmed from X-ray images, a trial-and-error process that is time-consuming and prone to inaccuracies. To improve this process, we used automated image analysis and machine learning methods to detect electrode locations.

A U-Net, a type of convolutional neural network, was trained to segment vertebrae T11-12 and L1-5, and individual electrodes from a subset of 6 anterior-posterior X-ray images from patients who received SCS implants in the lumbosacral region. All images were preprocessed, including resizing to the same dimension, and enhancing contrast as needed. Images were annotated by pixel-labeling the corresponding areas for the desired vertebra levels and electrodes. This involved manually outlining each vertebra and electrode using MATLAB's Image Labeling App, generating a ground truth dataset for future training of the U-Net. A larger dataset of 200 additional images will be obtained for more generalized training of the U-Net. Our trained U-Net will be essential for future intraoperative testing, where we will record muscle activity from the leg muscles using electromyography (EMG) following SCS. It will facilitate correlating the anatomical position of the electrodes with the muscle activity from EMG, allowing us to construct a functional map of muscle activation in response to SCS. As SCS becomes more widely used as a rehabilitation intervention, the need for creating a functional map between electrode location and muscle responses will increase. This work will be a valuable tool for researchers and clinicians in automatically generating a functional map from SCS implants.