Goal: In this paper we investigated the use of smartphone sensors and Artificial Intelligence techniques for the automatic quantification of the MDS-UPDRS-Part III Leg Agility (LA) task, representative of lower limb bradykinesia. Methods: We collected inertial data from 93 PD subjects. Four expert neurologists provided clinical evaluations. We employed a novel Artificial Neural Network approach in order to get a continuous output, going beyond the MDS-UPDRS score discretization. Results: We found a Pearson correlation of 0.92 between algorithm output and average clinical score, compared to an inter-rater agreement index of 0.88. Furthermore, the classification error was less than 0.5 scale point in about 80% cases. Conclusions: We proposed an objective and reliable tool for the automatic quantification of the MDS-UPDRS Leg Agility task. In perspective, this tool is part of a larger monitoring program to be carried out during activities of daily living, and managed by the patients themselves.

中文翻译：

---

基于智能手机的 MDS-UPDRS-III 第 3.8 项评估，用于评估帕金森病患者的腿部灵活性

目标：在本文中，我们研究了智能手机传感器和人工智能技术在 MDS-UPDRS-Part III Leg Agility (LA) 任务（代表下肢运动迟缓）的自动量化中的应用。方法：我们收集了 93 名 PD 受试者的惯性数据。四位专家神经学家提供了临床评估。我们采用了一种新颖的人工神经网络方法来获得连续的输出，超越了 MDS-UPDRS 分数离散化。结果：我们发现算法输出和平均临床评分之间的 Pearson 相关性为 0.92，而评分者间一致性指数为 0.88。此外，在大约 80% 的情况下，分类误差小于 0.5 个刻度点。结论：我们提出了一个客观可靠的工具，用于自动量化 MDS-UPDRS 腿部敏捷性任务。从长远来看，该工具是在日常生活活动期间执行的更大监测计划的一部分，并由患者自己管理。