Motor Evoked Potentials (MEPs) are used to monitor disability progression in multiple sclerosis (MS). Their morphology plays an important role in this process. Currently, however, there is no clear definition of what constitutes a normal or abnormal morphology. To address this, five experts independently labeled the morphology (normal or abnormal) of the same set of 1,000 MEPs. The intra- and inter-rater agreement between the experts indicates they agree on the concept of morphology, but differ in their choice of threshold between normal and abnormal morphology. We subsequently performed an automated extraction of 5,943 time series features from the MEPs to identify a valid proxy for morphology, based on the provided labels. To do this, we compared the cross-validation performances of one-dimensional logistic regression models fitted to each of the features individually. We find that the approximate entropy (ApEn) feature can accurately reproduce the majority-vote labels. The performance of this feature is evaluated on an independent test set by comparing to the majority vote of the neurologists, obtaining an AUC score of 0.92. The model slightly outperforms the average neurologist at reproducing the neurologists consensus-vote labels. We can conclude that MEP morphology can be consistently defined by pooling the interpretations from multiple neurologists and that ApEn is a valid continuous score for this. Having an objective and reproducible MEP morphological abnormality score will allow researchers to include this feature in their models, without manual annotation becoming a bottleneck. This is crucial for large-scale, multi-center datasets. An exploratory analysis on a large single-center dataset shows that ApEn is potentially clinically useful. Introducing an automated, objective, and reproducible definition of morphology could help overcome some of the barriers that are currently obstructing broad adoption of evoked potentials in daily care and patient follow-up, such as standardization of measurements between different centers, and formulating guidelines for clinical use.

中文翻译：

---

破译运动诱发电位的形态学

运动诱发电位 (MEP) 用于监测多发性硬化症 (MS) 的残疾进展。它们的形态在这个过程中起着重要作用。然而，目前对于什么构成正常或异常形态没有明确的定义。为了解决这个问题，五位专家独立标记了同一组 1,000 名 MEP 的形态（正常或异常）。专家之间的评分者内和评分者间的一致性表明他们同意形态学的概念，但他们在正常和异常形态学之间的阈值选择上有所不同。随后，我们从 MEP 中自动提取了 5,943 个时间序列特征，以根据提供的标签识别有效的形态学代理。去做这个，我们分别比较了适合每个特征的一维逻辑回归模型的交叉验证性能。我们发现近似熵 (ApEn) 特征可以准确地再现多数投票标签。通过与神经科医生的多数投票进行比较，在独立测试集上评估此功能的性能，获得 0.92 的 AUC 分数。该模型在重现神经学家共识投票标签方面略胜于一般神经学家。我们可以得出结论，MEP 形态可以通过汇集多位神经学家的解释来一致地定义，并且 ApEn 是一个有效的连续分数。拥有客观且可重复的 MEP 形态异常评分将使研究人员能够将此特征包含在他们的模型中，没有手动注释成为瓶颈。这对于大规模、多中心的数据集至关重要。对大型单中心数据集的探索性分析表明 ApEn 具有潜在的临床用途。引入自动、客观和可重复的形态学定义可以帮助克服目前阻碍在日常护理和患者随访中广泛采用诱发电位的一些障碍，例如不同中心之间测量的标准化，以及制定临床指南。用。