Automatic Ultrasound Curve Angle Measurement via Affinity Clustering for Adolescent Idiopathic Scoliosis Evaluation

TL;DR

The paper addresses the radiation exposure concerns of repeated Cobb angle monitoring in AIS by proposing a fully automatic ultrasound curve angle (UCA) measurement system. It introduces a dual-branch network that detects vertebral landmarks and segments vertebrae, then uses a vertebral-affinity clustering mechanism to group landmarks and form lines for UCA calculation. The authors show a strong relationship between automatic UCA and radiographic Cobb angle (R^2=0.858) on a sizeable VPI ultrasound dataset, with robust line detection metrics and ablation studies confirming design choices. This method reduces operator dependence, supports vertebral-level analysis, and has potential to replace manual UCA measurement in ultrasound-based scoliosis assessment, facilitating safer and more accessible monitoring.

Abstract

The current clinical gold standard for evaluating adolescent idiopathic scoliosis (AIS) is X-ray radiography, using Cobb angle measurement. However, the frequent monitoring of the AIS progression using X-rays poses a challenge due to the cumulative radiation exposure. Although 3D ultrasound has been validated as a reliable and radiation-free alternative for scoliosis assessment, the process of measuring spinal curvature is still carried out manually. Consequently, there is a considerable demand for a fully automatic system that can locate bony landmarks and perform angle measurements. To this end, we introduce an estimation model for automatic ultrasound curve angle (UCA) measurement. The model employs a dual-branch network to detect candidate landmarks and perform vertebra segmentation on ultrasound coronal images. An affinity clustering strategy is utilized within the vertebral segmentation area to illustrate the affinity relationship between candidate landmarks. Subsequently, we can efficiently perform line delineation from a clustered affinity map for UCA measurement. As our method is specifically designed for UCA calculation, this method outperforms other state-of-the-art methods for landmark and line detection tasks. The high correlation between the automatic UCA and Cobb angle (R$^2$=0.858) suggests that our proposed method can potentially replace manual UCA measurement in ultrasound scoliosis assessment.

Figures (8)

• Figure 1: . (a) An Illustration of the generation of a volume projection imaging (VPI). The probe is being moved at a constant speed from bottom to top along the spine curve on the patients' skin. B-mode images combined with recorded spatial information are grouped for the generation of 3D ultrasound volume. The coronal ultrasound image is then generated using the VPI method, which incorporates a customized depth profile based on the distance from the skin to the laminae. (b) Ultrasound spinous process angle (SPA). The scoliotic curve on the medial shadow of the spinous processes is used to measure the angle for AIS diagnosis. Cheung2015_VPI (c) Ultrasound curve angle (UCA). For thoracic region, line is placed on the center of the shadow of a transverse process (purple dotted line). For lumbar region, lines are drawn towards the center of the bilateral sides of lump (green dotted line) Lee2021.
• Figure 2: . The model identifies all potential anatomical landmarks along both sides of the bone curvature. The landmarks corresponding to the same vertebrae are connected based on the clustered affinity map. The most tilted lines in different regions are selected to form the UCA for the assessment of scoliosis.
• Figure 3: . Overview pipeline of automatic UCA measurement. The model extracts latent features through a feature backbone for both landmark detection and spine segmentation, respectively. The features in the segmentation decoder are shared with the landmark detection decoder. In the reference stage, the vertebrae segmentation map is parsed to represent the affinity relationship among the detected candidate points. Points belonging to the same vertebra are grouped together to form lines.
• Figure 4: . The pipeline of automatic UCA measurement on stage of training and reference.
• Figure 5: . An example of vertebral-level line detection. The local extreme values are formed the UCA for angle measurement.