Idiopathic Parkinson’s syndrome is the second most common neurodegenerative disease with increasing incidence. Prompt diagnosis confirmation at early disease stages is desirable to enable effective therapies to be initiated as soon as possible. One opportunity is the analysis of wrist sensor data.
The diagnosis of idiopathic Parkinson’s syndrome (IPS) is mostly made clinically. Adjuvant diagnostic tests are expensive, time-consuming, insufficiently available, or nonspecific. Therefore, there is a high clinical need for an available, inexpensive, objective, and interpretable method. Machine learning can be used for meaningful analysis of motion sensor data. Such data can be collected, for example, through devices worn on the body (wearables). Combining miniaturized and increasingly connected Internet-of-Things (IoT) sensor diagnostics with machine learning methods offers the opportunity for a low-cost yet highly effective diagnostic methodology.
A proof-of-concept study for a data-driven diagnostic approach using a hierarchical model in early-stage PD was conducted to collect clinimetric metrics of a diagnostic algorithm using machine learning and wrist sensor data. 3D accelerometer data were collected from 25 patients with early stage disease and 25 control subjects. Recordings were made using a Microsoft Band 2, which recorded movements of the arm at a sampling rate of 62.5 Hz with the subjects moving freely (free living setting). The sensor data were fitted into a hierarchical model. The algorithm classified 22 of the 25 PD patients as positive, and 21 of the 25 healthy individuals as negative. The specificity was 0.85 and the sensitivity was 0.84. The positive predictive value was 0.88, and the negative predictive value was 0.84. The accuracy was 0.86. The results are promising and should be validated in further cohorts with optimized models.
Source: 92nd Congress of the German Society of Neurology (DGN)
InFo NEUROLOGY & PSYCHIATRY 2019; 17(6): 35 (published 11/23/19, ahead of print).
