RT-ring: a small wearable device for tremulous Parkinson's disease diagnosis in primary care

Abstract

Introduction: Differential diagnosis of rest tremor (RT) disorders is challenging, often requiring 123I-ioflupane single-photon-emission-computed tomography (DaTscan), an expensive technique not available worldwide. In the current study, we investigated the performance of a new wearable mobile device termed "RT-ring" in predicting DaTscan result in patients presenting with RT based on rest tremor inertial features.

Methods: Consecutive RT patients underwent RT-ring tremor analysis, surface electromyography (sEMG), and DaTscan. The RT-ring is a miniaturized mobile device that uses machine learning based on inertial tremor data to estimate the RT pattern. This electrophysiologic tremor feature has proven to accurately predict DaTscan result. The primary outcome was the RT-ring's performance in distinguishing patients with and without striatal dopaminergic deficit.

Results: Sixty-seven RT patients were enrolled, including 42 patients with striatal dopaminergic deficit and 25 with normal DaTscan. The RT-ring showed 85.0% sensitivity, 90.9% specificity, and 87.9% balanced accuracy in predicting DaTscan result, and demonstrated 96.8% agreement with sEMG in RT pattern classification.

Conclusion: The RT-ring is a promising, non-invasive, user-friendly, wearable mobile device for supporting the diagnosis of tremulous Parkinson's disease in primary care settings, especially in low-income countries with limited access to dopamine imaging.

Keywords: DaTscan; Parkinson’s disease; RT-ring; essential tremor plus; machine learning; rest tremor; tremor pattern; wearable device.

Figure 1

(A) The RT-ring device. (B) The RT-ring worn on a subject’s finger. (C) The mobile app’s initial screen shows three boxes: one to start a new session, one to see and access the session list with previous recordings, and one showing the device battery charge level and the connection status between the app and the device. (D) An example of the RT-ring report with the patient’s name or identifier, the estimated predominant rest tremor pattern, the prediction on DaTscan result, and a summary of the characteristics for each of the five tremor recording segments (frequency, estimated pattern and probability of pattern estimation). In the report shown in part label D, in the RT-ring report, frequency is provided in Hz.

Figure 2

The correct hand positioning is checked through tilt sensors (roll and pitch angle evaluation). (A) The correct positioning of the hand for rest tremor recording is shown on the graphic user interface. (B) Incorrectly positioned hand: roll in red. (C) Correctly positioned hand: recording can be started.

Figure 3

RT-ring workflow. (A) When starting a new session, the correct positioning of the hand is checked. The recording begins only when the hand is hanging down from the chair armrest. The app collects five 10-s segments, each undergoing a quality control process. The tremor pattern of each segment is then estimated using a machine-learning (ML) model based on tremor inertial features. If the same pattern (alternating or synchronous) is estimated with a probability >70% by the ML model in at least 4 out of 5 segments, the patient’s tremor is classified as predominantly alternating or synchronous and the prediction on the DaTscan result is shown. (B) The processing of a single 10-s recording is shown. Each segment undergoes a tremor quality check (QC) to ensure there is a rhythmic hand movement with a frequency between 2 and 10 Hz, without significant frequency shifts over time. If a segment does not meet these criteria, it is discarded and a new segment is recorded, until 5 segments passing the QC are collected. Tremor inertial features are then extracted from each recording segment and used as input for a ML model to predict the tremor pattern of the RT segment. RT, rest tremor; ML, machine-learning.