Cervical curvature measurement using multi-view parametric modeling imaging on reconstructed 3D ultrasound volumes

Three-dimensional (3D) ultrasound imaging is a promising tool for assessing cervical lordosis, but the cervical spine structures visualized in ultrasound images are often fragmented, making curvature measurement challenging. In this study, we proposed a 3D multi-view parametric modeling method to reconstruct the complete topological structure of the cervical spine from ultrasound images for curvature estimation. Two 3D ultrasound volumes were acquired from the left and right posterior views, and parametric modeling was used to generate cervical spine models. Signed distance functions (SDFs) of the two models were then reconstructed using a multi-layer perceptron (MLP) and fused based on the Chamfer distance to obtain the final models. Cobb angles were measured on sagittal projection images of the final models to evaluate cervical curvatures under different postures. Experimental results showed that the multi-view modeling method reduced curvature measurement error by 17% compared with the single-view approach. The average difference between ultrasound and MRI measurements was 3.28±0.73°, with a correlation of 0.94. Moreover, the proposed 3D parametric model provided an intuitive visualization of the cervical spine, suggesting its potential as a reliable approach for evaluating cervical lordosis.