Automatic ultrasound curve angle measurement via affinity clustering for adolescent idiopathic scoliosis evaluation

• Published in: Expert systems with applications Vol. 269; p. 126410
• Main Authors: Zhou, Yihao, Lee, Timothy Tin-Yan, Lai, Kelly Ka-Lee, Wu, Chonglin, Lau, Hin Ting, Yang, De, Song, Zhen, Chan, Chui-Yi, Chu, Winnie Chiu-Wing, Cheng, Jack Chun-Yiu, Lam, Tsz-Ping, Zheng, Yong-Ping
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 15.04.2025
• Subjects: Intelligent scoliosis diagnosis; Landmark detection; Ultrasound volume projection imaging; Vertebrae; Vertebrae; Ultrasound volume projection imaging; Landmark detection; Intelligent scoliosis diagnosis
• DOI: 10.1016/j.eswa.2025.126410
• ISSN: 0957-4174

The current clinical gold standard for evaluating adolescent idiopathic scoliosis (AIS) is X-ray radiography, specifically through Cobb angle measurement. However, frequent monitoring of AIS progression using X-rays presents a significant challenge due to the risks associated with cumulative radiation exposure. Although 3D ultrasound offers a validated radiation-free alternative, it relies on manual spinal curvature assessment, leading to inter and intra-rater angle variation. In this study, we propose an automated ultrasound curve angle (UCA) measurement system that utilizes a dual-branch network to simultaneously perform landmark detection and vertebra segmentation on ultrasound coronal images. The system incorporates an affinity clustering algorithm within vertebral segments to establish landmark relationships, enabling efficient line delineation for UCA measurement. Our method, specifically optimized for UCA calculation, demonstrates superior performance in landmark and line detection compared to existing approaches. The high correlation between the automatic UCA and Cobb angle (R2=0.858) suggests that our proposed method can potentially replace manual UCA measurement in ultrasound scoliosis assessment. This advancement could significantly enhance the accuracy and reliability of scoliosis monitoring while reducing the need for manual measurement. •We have achieved the fully automatic ultrasound curve angle measurement using a deep learning-based estimation model.•We use a clustering-based strategy to study the relationship between landmarks for line delineation.•The model eliminates inter-observer variability of measurement and supports the vertebral-level analysis, providing a comprehensive understanding of spinal morphology.