Garbey Marc, Lesport Quentin, Girma Helen, Öztosun Gülşen, Kaminski Henry J
Department of Surgery, School of Medicine & Health Sciences, George Washington University, Washington, District of Columbia, USA.
Laboratoire Des Sciences de l'Ingénieur Pour l'Environnement (LaSIE) UMR-CNRS 7356, University of La Rochelle, La Rochelle, France.
Muscle Nerve. 2025 Jul;72(1):34-41. doi: 10.1002/mus.28394. Epub 2025 Apr 2.
INTRODUCTION/AIMS: The adoption of telemedicine is generally considered as advantageous for patients and physicians, but there is limited rigorous assessment of examination strengths and limitations. We set out to perform a quantitative assessment of the limitations of a standardized examination of subjects with myasthenia gravis (MG) during video-taped telemedicine sessions.
We utilized a video bank containing recordings from 51 MG patients who completed two telemedicine-based examinations with neuromuscular experts; each recording included the MG core examination (MG-CE) and the MG activities of daily living (MG-ADL). We then applied artificial intelligence (AI) algorithms from computer vision and speech analysis to natural language processing to generate and assess the reproducibility and inter-rater reliability of the MG-CE and MG-ADL.
We successfully developed a technology to assess video examinations. While overall MG-CE scores were consistent across examiners, individual metrics showed significant variability, with up to a 25% variation in scoring within the MG-CE's range. Additionally, there was wide variability in adherence to MG-ADL instructions. These variations were attributed to differences in examiner instructions, video recording limitations, and patient disease severity.
We were able to develop a system of digital analysis of neuromuscular examinations in order to assess variability in individual scoring measures of the MG-ADL and MG-CE. Our approach enabled post hoc quantitative analysis of neuromuscular examinations. Further refinement of this technology could enhance examiner training and reduce variability in clinical trial outcome measures.
引言/目的：远程医疗的应用通常被认为对患者和医生都有好处，但对其检查优势和局限性的严格评估却很有限。我们着手对重症肌无力（MG）患者在录像远程医疗会诊期间进行的标准化检查的局限性进行定量评估。
我们利用了一个视频库，其中包含51名MG患者的记录，这些患者与神经肌肉专家完成了两次基于远程医疗的检查；每次记录都包括MG核心检查（MG-CE）和MG日常生活活动（MG-ADL）。然后，我们将来自计算机视觉和语音分析的人工智能（AI）算法应用于自然语言处理，以生成和评估MG-CE和MG-ADL的可重复性和评分者间信度。
我们成功开发了一种评估视频检查的技术。虽然不同检查者的总体MG-CE评分一致，但各个指标显示出显著差异，在MG-CE的评分范围内，评分差异高达25%。此外，在遵循MG-ADL指令方面也存在很大差异。这些差异归因于检查者指令的不同、视频记录的局限性以及患者疾病的严重程度。
我们能够开发一种神经肌肉检查的数字分析系统，以评估MG-ADL和MG-CE个体评分指标的差异。我们的方法能够对神经肌肉检查进行事后定量分析。进一步完善这项技术可以加强检查者培训，并减少临床试验结果指标的差异。
