Automatic Spinal Ultrasound Image Segmentation and Deployment for Real-time Spine Volumetric Reconstruction

Spinal ultrasound image segmentation, as the key for 3D spine reconstruction, has always been one of the most challenging issues in the field of medical image processing because of the intrinsic limitations of ultrasound imaging including low contrast, noise (artifacts, speckle), fuzzy and missing boundaries, etc. In this paper, we propose a real-time 3D spine volumetric reconstruction from spinal ultrasound images, which designs a 3D Slicer middleware to deploy our spine ultrasound segmentation model into the open-source software 3D Slicer. Specifically, we first train and fine-tune a deep residual U-shaped network for ultrasound image segmentation. Then a middleware is designed so that 3D Slicer can directly adopt our segmentation model written in Pytorch. Finally, our system achieves real-time 3D spine segmentation, reconstruction and visualization according to the model inference results. Comparative experiments demonstrate the effectiveness and superiority of our method, and our ultrasound segmentation model can obtain satisfactory Fuzzy-F1 scores and F1 scores.