Quantifying Tremor in Essential Tremor Using Inertial Sensors-Validation of an Algorithm.
Quantifying Tremor in Essential Tremor Using Inertial Sensors-Validation of an Algorithm.
复制标题
DOI：
10.1109/jtehm.2020.3032924
复制
发表时间：
2021
影响因子：
3.4
通讯作者：
Haubenberger D
中科院分区：
文献类型：
--
作者：
Mcgurrin P;Mcnames J;Wu T;Hallett M;Haubenberger D
特发性震颤的评估通常由训练有素的临床医生完成，他们观察不同姿势和动作时的肢体，然后对震颤进行评估。虽然该方法已被证明是可靠的，但评分间和评分内的可靠性以及对训练的需要可能使该方法难以用于症状进展。使用惯性传感器可以潜在地克服临床评定量表的许多局限性，但迄今为止，许多设计用于量化震颤的算法存在关键局限性。方法提出了一种利用惯性传感器表征震颤的新算法。它采用两阶段的方法：1)估计受试者的震颤频率，并仅对该范围附近的震颤进行量化；2)在记录期间，将震颤幅度估计为高于基线活动的信号功率部分，即使存在其他活动，也可以进行震颤估计；3)以平移（cm）和旋转（°）为物理单位估计震颤幅度，与当前的震颤等级量表一致。我们在技术上使用机械臂验证了算法，并通过将算法输出与训练有素的临床医生对原发性震颤患者进行震颤评定量表报告的数据进行了临床比较。结果经技术验证，旋转振幅精度优于±0.2°，位置振幅精度优于±0.1 cm。临床验证显示，旋转和位置成分与震颤评分量表得分显著相关。结论：我们证明，即使在存在其他活动的情况下，我们的算法也可以准确地量化震颤，这可能为家庭监测提供了一步。
Background Assessment of essential tremor is often done by a trained clinician who observes the limbs during different postures and actions and subsequently rates the tremor. While this method has been shown to be reliable, the inter- and intra-rater reliability and need for training can make the use of this method for symptom progression difficult. Many limitations of clinical rating scales can potentially be overcome by using inertial sensors, but to date many algorithms designed to quantify tremor have key limitations. Methods We propose a novel algorithm to characterize tremor using inertial sensors. It uses a two-stage approach that 1) estimates the tremor frequency of a subject and only quantifies tremor near that range; 2) estimates the tremor amplitude as the portion of signal power above baseline activity during recording, allowing tremor estimation even in the presence of other activity; and 3) estimates tremor amplitude in physical units of translation (cm) and rotation (°), consistent with current tremor rating scales. We validated the algorithm technically using a robotic arm and clinically by comparing algorithm output with data reported by a trained clinician administering a tremor rating scale to a cohort of essential tremor patients. Results Technical validation demonstrated rotational amplitude accuracy better than ±0.2 degrees and position amplitude accuracy better than ±0.1 cm. Clinical validation revealed that both rotation and position components were significantly correlated with tremor rating scale scores. Conclusion We demonstrate that our algorithm can quantify tremor accurately even in the presence of other activities, perhaps providing a step forward for at-home monitoring.