Objective: To validate an automatic classification system for the identification of errors in the performance of standardized MDS-UPDRS III exercises by using inertial sensors from smartwatches (SW) and machine learning classifiers.

Background: To improve current methods of diagnosis and objective monitoring of PD symptoms, wearable devices, and machine learning techniques have been used to develop solutions that enable remote monitoring [1]. One of the requirements to optimize monitoring is to ensure the quality of the data collected on ambulatory and remote settings, without the need for continuous supervision.

Method: Six subjects with PD participating in the TECAPARK project were recruited [2]. These participants wore a consumer smartwatch on the wrist of their most symptom-affected side. Data were captured weekly (eight weeks) during the performance of a set of eight standardized exercises selected from the MDS-UPDRS scale part 3. The automatic verification system was developed by using inertial signals of the SW treated with digital signal processing techniques and machine learning algorithms. The data were labeled by a clinical expert by observing video recordings made during the trials.

Results: The highest accuracy in the error identification was obtained using the K-Nearest Neighbours classifier (KNN) trained with both triaxial gyroscope and triaxial accelerometer signals. The combination of feature extraction and principal component analysis (PCA) applied to each sensor independently achieved an accuracy of 89.4% (precision 89.6%; recall 72.9%; F-1 0.804) in the automatic verification of the execution of the exercises. Thus, the system can automatically identify, with high accuracy, the correct or wrong performance of the exercises without the supervision of a clinician or a caregiver.

Conclusion: The implementation of a KNN classifier enables the automatic identification of the correct performance of MDS-UPDSRS exercises. Verification modules like these could be integrated into intelligent data collection systems that operate in a user-guided manner. Also, these kinds of solutions combined with tremor detection systems [3], bradykinesia, FOG… can be used to perform autonomous data collection useful for continuous and remote monitoring of motor symptoms, improving the knowledge of the disease, and facilitating the development of better treatments for PD.

References: [1] Del Din, et al (2021). Front. Neurosci. 11 [2] TECAPARK (2018). Retrieved from www.i2a2.upm.es/tecapark [3] Sigcha L, et al. (2021) Sensors. 1

« Back to 2022 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/automated-error-detection-when-applying-the-protocol-for-the-monitoring-of-motor-function-in-patients-with-parkinsons-disease-using-smartwatches/