Angeles Paolo, Tai Yen, Pavese Nicola, Wilson Samuel, Vaidyanathan Ravi
IEEE Int Conf Rehabil Robot. 2017 Jul;2017:1512-1517. doi: 10.1109/ICORR.2017.8009462.
Deep brain stimulation (DBS) is currently being used as a treatment for symptoms of Parkinson's disease (PD). Tracking symptom severity progression and deciding the optimal stimulation parameters for people with PD is extremely difficult. This study presents a sensor system that can quantify the three cardinal motor symptoms of PD - rigidity, bradykinesia and tremor. The first phase of this study assesses whether data recorded from the system during physical examinations can be used to correlate to clinician's severity score using supervised machine learning (ML) models. The second phase concludes whether the sensor system can distinguish differences before and after DBS optimisation by a clinician when Unified Parkinson's Disease Rating Scale (UPDRS) scores did not change. An average accuracy of 90.9 % was achieved by the best ML models in the first phase, when correlating sensor data to clinician's scores. Adding on to this, in the second phase of the study, the sensor system was able to pick up discernible differences before and after DBS optimisation sessions in instances where UPDRS scores did not change.
深部脑刺激（DBS）目前被用作治疗帕金森病（PD）症状的方法。跟踪帕金森病患者症状严重程度的进展并确定最佳刺激参数极其困难。本研究提出了一种传感器系统，该系统可以量化帕金森病的三种主要运动症状——僵硬、运动迟缓及震颤。本研究的第一阶段评估在体格检查期间从该系统记录的数据是否可用于通过监督机器学习（ML）模型与临床医生的严重程度评分相关联。第二阶段得出结论，当统一帕金森病评定量表（UPDRS）评分未改变时，该传感器系统能否区分临床医生进行DBS优化前后的差异。在第一阶段，当将传感器数据与临床医生的评分相关联时，最佳ML模型实现了90.9%的平均准确率。此外，在研究的第二阶段，在UPDRS评分未改变的情况下，该传感器系统能够检测到DBS优化前后的明显差异。
