Objective: To investigate if motion data collected from inertial measurement units (IMUs) mounted on implements used in activities of daily living (ADLs) can characterize the functional impairment of a person with Parkinson’s disease (pwPD).

Background: Structured assessments of PD in clinics lack quantitative measures of functional impairment while performing ADLs. Clinicians involved in the long-term management of PD can benefit from a quantitative method to assess patient impairment at home.

Method: PwPD (n = 32) and healthy controls (HC) (n = 2) capable of performing ADLs were enrolled in the study. IMUs were calibrated and mounted to a plastic toothbrush and hair comb to track motion during ADL performance. Study participants were first asked to brush their teeth and comb their hair for an unspecified length of time. Patients were then instructed to repeat 2-3 additional one-minute trials, depending on the length of their first trial. Motion capture data and a video recording were collected for all trials. Patient UPDRS subscores were recorded and summed to one total score [1]. Signals under 30s long and those with sensor artifacts were removed. Accelerometer and gyroscope data were isolated from the IMU, preprocessed, and analyzed to extract 30 signal features [table1]. Features were used for a classification task to differentiate among HC (UPDRS 0), mild severity PD (UPDRS 0-32), and moderate severity PD (UPDRS 33-59). Signals were split into 7-second, non-overlapping intervals, and used to train a decision tree model.

Results: The model trained on toothbrushing data achieved F1-scores of 0.67, 0.93, and 0.86 for HCs, mild severity PD, and moderate severity PD, respectively, and an accuracy of 89.52% on a held-out test set (20% of total data). Decision tree models trained with 10-fold cross-validation achieved an average accuracy of 87.38% ± 3.85%. The model trained on combing data achieved F1-scores of 0.73, 0.92, and 0.75 respectively, and an accuracy of 86.96% on the test set. Average diagnostic accuracy was 85.82% ± 5.34% when trained with 10-fold cross-validation. Confusion matrices for both tasks are shown [figure1].

Conclusion: Specific features extracted from the sensor data was able to differentiate between HCs and two different severities of diagnosed PD, suggesting potential for this approach to be used to characterize patient functionality for ADLs in the home setting.

Table 1

Figure 1

References: [1] Fahn S, Elton RL, members of the UPDRS Development Committee. The Unified Parkinson’s Disease Rating Scale. In: S Fahn, CD Marsden, DB Calne, M Goldstein, eds. Recent Developments in Parkinson’s Disease. Vol 2. Florham Park, New Jersey: Macmillan Healthcare Information; 1987: 153-163, 293-304.

[2] Zhao, A., Cui, E., Leroux, A., Lindquist, M. A., & Crainiceanu, C. M. (2023). Evaluating the prediction performance of objective physical activity measures for incident Parkinson’s disease in the UK Biobank. Journal of Neurology, 270(12), 5913–5923. https://doi.org/10.1007/s00415-023-11939-0

« Back to 2025 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/a-novel-quantitative-assessment-to-evaluate-functional-impairment-in-parkinsons-disease/