Department of Neurology, Tokushima University Graduate School of Biomedical Sciences, 3-18-15 Kuramoto-cho, Tokushima 770-8503, Japan; Department of Neurology, Nara Medical University School of Medicine, 840 Shijo-cho, Kashihara, Nara 634-8521, Japan.
Department of Neurology, Tokushima University Graduate School of Biomedical Sciences, 3-18-15 Kuramoto-cho, Tokushima 770-8503, Japan.
Clin Neurophysiol. 2022 Aug;140:136-144. doi: 10.1016/j.clinph.2022.06.005. Epub 2022 Jun 17.
Although fasciculation on muscle ultrasonography (MUS) is useful in diagnosing amyotrophic lateral sclerosis (ALS), its applicability to early diagnosis remains unclear. We aimed to develop and validate diagnostic models especially beneficial to early-stage ALS via machine learning.
We investigated 100 patients with ALS, including 50 with early-stage ALS within 9 months from onset, and 100 without ALS. Fifteen muscles were bilaterally observed for 10 s each and the presence of fasciculations was recorded. Hierarchical clustering and nominal logistic regression, neural network, or ensemble learning were applied to the training cohort comprising the early-stage ALS to develop MUS-based diagnostic models, and they were tested in the validation cohort comprising the later-stage ALS.
Fasciculations on MUS in the brainstem or thoracic region had high specificity but limited sensitivities and predictive profiles for diagnosis of ALS. A machine learning-based model comprising eight muscles in the four body regions had a high sensitivity (recall), specificity, and positive predictive value (precision) for both early- and later-stage ALS patients.
We developed and validated MUS-fasciculation-based diagnostic models for early- and later-stage ALS.
Fasciculation detected in relevant muscles on MUS can contribute to the diagnosis of ALS from the early stage.
尽管肌电图（MUS）上的肌束震颤对于诊断肌萎缩侧索硬化症（ALS）很有用，但它在早期诊断中的适用性仍不清楚。我们旨在通过机器学习开发和验证特别有益于早期 ALS 的诊断模型。
我们研究了 100 名 ALS 患者，包括 50 名发病 9 个月内的早期 ALS 患者和 100 名非 ALS 患者。双侧 15 块肌肉各观察 10 秒，记录肌束震颤的存在。分层聚类和名义逻辑回归、神经网络或集成学习应用于包含早期 ALS 的训练队列，以开发基于 MUS 的诊断模型，并在包含晚期 ALS 的验证队列中进行测试。
脑干或胸部区域的 MUS 上的肌束震颤对 ALS 的诊断具有高特异性，但敏感性和预测谱有限。基于四个身体区域中 8 块肌肉的基于机器学习的模型对早期和晚期 ALS 患者均具有高敏感性（召回率）、特异性和阳性预测值（精度）。
我们开发并验证了用于早期和晚期 ALS 的基于 MUS 肌束震颤的诊断模型。
MUS 上相关肌肉中检测到的肌束震颤有助于从早期诊断 ALS。
