Presentation description

Neural injuries impair mobility. Spinal cord stimulation (SCS) is being explored for rehabilitation. Electrode placement is confirmed via X-ray, but manual identification is slow and inaccurate. We trained a U-Net model on 624 augmented X-ray images for automated segmentation, achieving 83.8% training accuracy but poor generalization (global accuracy: 11.3%, IoU: 4.14%). To improve accuracy, we propose object detection-based segmentation and nnU-Net. It'll enhance SCS mapping for clinical use.