背景:脊髓型颈椎病(CM)会引起一些症状,例如手笨拙,通常需要手术。CM的筛查和早期诊断很重要,因为一些患者不知道他们的早期症状,只有在他们的病情变得严重后才咨询外科医生。10秒手握和释放测试通常用于检查CM的存在。该测试很简单,但如果可以客观地评估CM特有的运动变化,则对筛查更有用。先前的一项研究使用LeapMotion分析了10秒手抓握和释放测试中的手指运动,非接触式传感器,并开发了一个系统,可以诊断CM具有高灵敏度和特异性使用机器学习。然而,之前的研究有局限性,因为该系统记录的参数很少,并且不能区分CM和其他手部疾病.
目的:本研究旨在开发一种能够以更高的灵敏度和特异性诊断CM的系统,并区分CM和腕管综合征(CTS),一种常见的手部疾病.然后,我们使用改进的LeapMotion验证了该系统,该系统可以记录每个手指的关节。
方法:总共,31、27和29名参与者被招募到CM,CTS,和对照组,分别。我们开发了一个使用LeapMotion的系统,该系统记录了229个手指运动参数,而参与者则尽可能快地握住并释放手指。用支持向量机进行机器学习,建立二元分类模型,计算灵敏度,特异性,和曲线下面积(AUC)。我们开发了两种模型,一个在CM和对照组中诊断CM(CM/控制模型),在CM和非CM组中诊断CM(CM/非CM模型)。
结果:CM/对照模型指标如下:灵敏度74.2%,特异性89.7%,和AUC0.82。CM/非CM模型指数如下:灵敏度71%,特异性72.87%,和AUC0.74。
结论:我们开发了一种能够以更高的灵敏度和特异性诊断CM的筛查系统。该系统可以区分患有CM的患者与患有CTS的患者以及健康患者,并且具有在各种患者中筛查CM的潜力。
BACKGROUND: Cervical myelopathy (CM) causes several symptoms such as clumsiness of the hands and often requires surgery. Screening and early diagnosis of CM are important because some patients are unaware of their early symptoms and consult a surgeon only after their condition has become severe. The 10-second hand grip and release test is commonly used to check for the presence of CM. The test is simple but would be more useful for screening if it could objectively evaluate the changes in movement specific to CM. A previous study analyzed finger movements in the 10-second hand grip and release test using the Leap Motion, a noncontact sensor, and a system was developed that can diagnose CM with high sensitivity and specificity using machine learning. However, the previous study had limitations in that the system recorded few parameters and did not differentiate CM from other hand disorders.
OBJECTIVE: This study aims to develop a system that can diagnose CM with higher sensitivity and specificity, and distinguish CM from carpal tunnel syndrome (CTS), a common hand disorder. We then validated the system with a modified Leap Motion that can record the joints of each finger.
METHODS: In total, 31, 27, and 29 participants were recruited into the CM, CTS, and control groups, respectively. We developed a system using Leap Motion that recorded 229 parameters of finger movements while participants gripped and released their fingers as rapidly as possible. A support vector machine was used for machine learning to develop the binary classification model and calculated the sensitivity, specificity, and area under the curve (AUC). We developed two models, one to diagnose CM among the CM and control groups (CM/control model), and the other to diagnose CM among the CM and non-CM groups (CM/non-CM model).
RESULTS: The CM/control model indexes were as follows: sensitivity 74.2%, specificity 89.7%, and AUC 0.82. The CM/non-CM model indexes were as follows: sensitivity 71%, specificity 72.87%, and AUC 0.74.
CONCLUSIONS: We developed a screening system capable of diagnosing CM with higher sensitivity and specificity. This system can differentiate patients with CM from patients with CTS as well as healthy patients and has the potential to screen for CM in a variety of patients.