Department of Electrical and Computer Engineering, University of Michigan - Dearborn, MI, USA; Instituto Politecnico Nacional, Centro de Investigacion en Computacion, Mexico City, Mexico.
Instituto Politecnico Nacional, Centro de Investigacion en Computacion, Mexico City, Mexico.
Int J Med Inform. 2018 Jun;114:6-17. doi: 10.1016/j.ijmedinf.2018.03.002. Epub 2018 Mar 11.
Currently the most consistent, widely accepted and detailed instrument to rate Parkinson's disease (PD) is the Movement Disorder Society sponsored Unified Parkinson Disease Rating Scale (MDS-UPDRS). However, the motor examination is based upon subjective human interpretation trying to capture a snapshot of PD status. Wearable sensors and machine learning have been broadly used to analyze PD motor disorder, but still most ratings and examinations lay outside MDS-UPDRS standards. Moreover, logical connections between features and output ratings are not clear and complex to derive from the model, thus limiting the understanding of the structure in the data.
Fifty-seven PD patients underwent a full motor examination in accordance to the MDS-UPDRS on twelve different sessions, gathering 123 measurements. Overall, 446 different combinations of limb features correlated to rest tremors amplitude are extracted from gyroscopes, accelerometers, and magnetometers and feed into a fuzzy inference system to yield severity estimations.
A method to perform rest tremor quantification fully adhered to the MDS-UPDRS based on wearable sensors and fuzzy inference system is proposed, which enables a reliable and repeatable assessment while still computing features suggested by clinicians in the scale. This quantification is straightforward and scalable allowing clinicians to improve inference by means of new linguistic statements. In addition, the method is immediately accessible to clinical environments and provides rest tremor amplitude data with respect to the timeline. A better resolution is also achieved in tremors rating by adding a continuous range.
目前，评估帕金森病（PD）最一致、最广泛接受和最详细的工具是运动障碍协会赞助的帕金森病统一评定量表（MDS-UPDRS）。然而，运动检查是基于主观的人为解释，试图捕捉 PD 状态的快照。可穿戴传感器和机器学习已被广泛用于分析 PD 运动障碍，但大多数评分和检查仍不符合 MDS-UPDRS 标准。此外，特征与输出评分之间的逻辑联系不清楚，并且难以从模型中得出，从而限制了对数据结构的理解。
五十七名 PD 患者在十二个不同的疗程中按照 MDS-UPDRS 进行了全面的运动检查，共采集了 123 个测量值。总体而言，从陀螺仪、加速度计和磁力计中提取了 446 种与静止震颤幅度相关的肢体特征的 446 种不同组合，并将其输入模糊推理系统以产生严重程度估计。
提出了一种基于可穿戴传感器和模糊推理系统的完全符合 MDS-UPDRS 的静止震颤定量方法，该方法能够进行可靠和可重复的评估，同时仍能计算出量表中临床医生建议的特征。这种量化方法简单直观，具有可扩展性，允许临床医生通过新的语言陈述来改进推理。此外，该方法可立即应用于临床环境，并提供与时间线相关的静止震颤幅度数据。通过添加连续范围，在震颤评分方面也能实现更高的分辨率。
