MNT-DeepSL: Median nerve tracking from carpal tunnel ultrasound images with deep similarity learning and analysis on continuous wrist motions

You-Wei Wang; Ruey-Feng Chang; Yi-Shiung Horng; Chii-Jen Chen

doi:10.1016/j.compmedimag.2019.101687

MNT-DeepSL: Median nerve tracking from carpal tunnel ultrasound images with deep similarity learning and analysis on continuous wrist motions

Comput Med Imaging Graph. 2020 Mar:80:101687. doi: 10.1016/j.compmedimag.2019.101687. Epub 2020 Jan 20.

Authors

You-Wei Wang¹, Ruey-Feng Chang², Yi-Shiung Horng³, Chii-Jen Chen⁴

Affiliations

¹ Department of Computer Science and Information Engineering, National Taiwan University, Taipei, Taiwan.
² Department of Computer Science and Information Engineering, National Taiwan University, Taipei, Taiwan. Electronic address: rfchang@csie.ntu.edu.tw.
³ Department of Physical Medicine and Rehabilitation, Taipei Tzu Chi Hospital, Buddhist Tzu Chi Medical Foundation, Taipei, Taiwan; Department of Medicine, Tzu Chi University, Hualien, Taiwan.
⁴ Department of Medical Imaging and Radiological Technology, Yuanpei University of Medical Technology, Hsinchu, Taiwan. Electronic address: cjchen@mail.ypu.edu.tw.

PMID: 32004994
DOI: 10.1016/j.compmedimag.2019.101687

Abstract

Carpal tunnel syndrome (CTS) is a clinical disease that caused by the compression of median nerve within carpal tunnel. Traditional examining for CTS is electrodiagnostic (EDx), but the evaluation of EDx is more expensive and time-consuming. In the present day, ultrasound (US) image is used to clinical examining to make up the lack of nerve electrical inspection. The diagnostic criteria of US image for CTS are also defined in many researches. In this study, we propose a new tracking model with deep similarity learning for median nerve from CTS US images. Six wrist motions are defined in the clinical rehabilitation, and the proposed method can achieve accuracy more than 90 % for median nerve tracking. In the experiment, we discover some wrist motions, such as hook to full fist, the statistical information of median nerve tracking is more significant (P < 0.001). It means that some wrist motions are more easily to diagnose the problem of median nerve, and can be used as a basis for quick examining for CTS.

Keywords: CTS; Deep learning; Median nerve tracking; US.

MeSH terms

Carpal Tunnel Syndrome / diagnostic imaging*
Carpal Tunnel Syndrome / physiopathology
Deep Learning*
Humans
Image Interpretation, Computer-Assisted / methods*
Median Nerve / diagnostic imaging*
Median Nerve / physiopathology
Motion
Ultrasonography / methods*
Wrist / diagnostic imaging*
Wrist / physiopathology