Automatic 3D Ultrasound Modeling Imaging for Spine Deformity using neural networks
- Resource Type
- Conference
- Authors
- Qian, Liyue; Zhao, Jianhao; Gao, Yuchong; Tang, Yiwen; Zhang, Mingbo; Zheng, Rui
- Source
- 2023 IEEE International Ultrasonics Symposium (IUS) Ultrasonics Symposium (IUS), 2023 IEEE International. :1-4 Sep, 2023
- Subject
- Bioengineering
Components, Circuits, Devices and Systems
Fields, Waves and Electromagnetics
Signal Processing and Analysis
Radiography
Deformable models
Solid modeling
Visualization
Three-dimensional displays
Ultrasonic imaging
Ultrasonic variables measurement
3D US imaging
Parametric Spine modeling
HR-Net
Detr
- Language
- ISSN
- 1948-5727
The 3D ultrasound (US) spine image volumes have been used for the assessment of spine deformity, but can be rebuilt only by manually extracting the feature markers of each vertebra. The objectives of this paper are to automatically extract the feature markers of spine using neural networks for the reconstruction of 3D spine model images and to implement the 3D model image for the proxy Cobb angle measurement. Detection Transformer(Detr) and HR-Net were applied to find location of lamina pair on 2D projected coronal US image of spine. Then 3D coordinates were obtained by selecting the high reflectors to reconstruct 3D spine model. The result showed that the overall MAD and correlation of curvature measurement between 3D US model and radiography were 2.03°and 0.95, which was slightly better than direct measurement from 2D ultrasound images (2.53°and 0.92). The 3D spine model images can be built by automatically extracting coordinate information of key points (lamina centers) using neural networks.