One-stage Contour Regression Network for Muscle Segmentation

Hongyuan Zhang, Sijin Cai, Xiaofeng Wang, Yongjin Zhou
Proceedings of The First AAAI Bridge Program on AI for Medicine and Healthcare, PMLR 281:42-48, 2025.

Abstract

Muscle segmentation from ultrasound images is crucial for diagnosing and treating musculoskeletal diseases, as it provides geometric properties, such as area and thickness. However, most existing methods often rely on pixel-based segmentation networks to extract the muscle region. This is wasteful, inefficient, failing to meet the real-time requirements of dynamic analysis. Therefore, to address these issues, we introduce a novel One-stage Contour-based Regression network, termed as OCRSeg, which formulates the muscle segmentation problem as contour regression in the vertical direction. In specific, we achieve this using a regression framework that is: 1) simple, consisting of an encoder for feature extraction and an MLP for regressing contour coordinates; 2) flexible, enforcing soft total variation constraints during training to ensure locally smooth muscle edges; and 3) optimized for band-like musculature. Extensive experimental results on two datasets demonstrate that our method outperforms other state-of-the-art segmentation methods in terms of accuracy and efficiency, achieving over 95.0% in mask IoU score and a FPS over 100 in running speed. Furthermore, another key advantage is that OCRSeg can achieve better clinical biometrics estimation compared with other techniques. We hope our proposed framework can serve as a fundamental and strong baseline for muscle segmentation task.

Cite this Paper

BibTeX
@InProceedings{pmlr-v281-zhang25a, title = {One-stage Contour Regression Network for Muscle Segmentation}, author = {Zhang, Hongyuan and Cai, Sijin and Wang, Xiaofeng and Zhou, Yongjin}, booktitle = {Proceedings of The First AAAI Bridge Program on AI for Medicine and Healthcare}, pages = {42--48}, year = {2025}, editor = {Wu, Junde and Zhu, Jiayuan and Xu, Min and Jin, Yueming}, volume = {281}, series = {Proceedings of Machine Learning Research}, month = {25 Feb}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v281/main/assets/zhang25a/zhang25a.pdf}, url = {https://proceedings.mlr.press/v281/zhang25a.html}, abstract = {Muscle segmentation from ultrasound images is crucial for diagnosing and treating musculoskeletal diseases, as it provides geometric properties, such as area and thickness. However, most existing methods often rely on pixel-based segmentation networks to extract the muscle region. This is wasteful, inefficient, failing to meet the real-time requirements of dynamic analysis. Therefore, to address these issues, we introduce a novel One-stage Contour-based Regression network, termed as OCRSeg, which formulates the muscle segmentation problem as contour regression in the vertical direction. In specific, we achieve this using a regression framework that is: 1) simple, consisting of an encoder for feature extraction and an MLP for regressing contour coordinates; 2) flexible, enforcing soft total variation constraints during training to ensure locally smooth muscle edges; and 3) optimized for band-like musculature. Extensive experimental results on two datasets demonstrate that our method outperforms other state-of-the-art segmentation methods in terms of accuracy and efficiency, achieving over 95.0% in mask IoU score and a FPS over 100 in running speed. Furthermore, another key advantage is that OCRSeg can achieve better clinical biometrics estimation compared with other techniques. We hope our proposed framework can serve as a fundamental and strong baseline for muscle segmentation task.} }
Endnote
%0 Conference Paper %T One-stage Contour Regression Network for Muscle Segmentation %A Hongyuan Zhang %A Sijin Cai %A Xiaofeng Wang %A Yongjin Zhou %B Proceedings of The First AAAI Bridge Program on AI for Medicine and Healthcare %C Proceedings of Machine Learning Research %D 2025 %E Junde Wu %E Jiayuan Zhu %E Min Xu %E Yueming Jin %F pmlr-v281-zhang25a %I PMLR %P 42--48 %U https://proceedings.mlr.press/v281/zhang25a.html %V 281 %X Muscle segmentation from ultrasound images is crucial for diagnosing and treating musculoskeletal diseases, as it provides geometric properties, such as area and thickness. However, most existing methods often rely on pixel-based segmentation networks to extract the muscle region. This is wasteful, inefficient, failing to meet the real-time requirements of dynamic analysis. Therefore, to address these issues, we introduce a novel One-stage Contour-based Regression network, termed as OCRSeg, which formulates the muscle segmentation problem as contour regression in the vertical direction. In specific, we achieve this using a regression framework that is: 1) simple, consisting of an encoder for feature extraction and an MLP for regressing contour coordinates; 2) flexible, enforcing soft total variation constraints during training to ensure locally smooth muscle edges; and 3) optimized for band-like musculature. Extensive experimental results on two datasets demonstrate that our method outperforms other state-of-the-art segmentation methods in terms of accuracy and efficiency, achieving over 95.0% in mask IoU score and a FPS over 100 in running speed. Furthermore, another key advantage is that OCRSeg can achieve better clinical biometrics estimation compared with other techniques. We hope our proposed framework can serve as a fundamental and strong baseline for muscle segmentation task.
APA
Zhang, H., Cai, S., Wang, X. & Zhou, Y.. (2025). One-stage Contour Regression Network for Muscle Segmentation. Proceedings of The First AAAI Bridge Program on AI for Medicine and Healthcare, in Proceedings of Machine Learning Research 281:42-48 Available from https://proceedings.mlr.press/v281/zhang25a.html.

Related Material