A Multi-Layer Gaussian Process for Motor Symptom Estimation in People With Parkinson's Disease

Abstract

The assessment of Parkinson's disease (PD) poses a significant challenge as it is influenced by various factors which lead to a complex and fluctuating symptom manifestation. Thus, a frequent and objective PD assessment is highly valuable for effective health management of people with Parkinson's disease (PwP). Here, we propose a method for monitoring PwP by stochastically modeling the relationships between their wrist movements during unscripted daily activities and corresponding annotations about clinical displays of movement abnormalities. We approach the estimation of PD motor signs by independently modeling and hierarchically stacking Gaussian process models for three classes of commonly observed movement abnormalities in PwP including tremor, (non-tremulous) bradykinesia, and (non-tremulous) dyskinesia. We use clinically adopted severity measures as annotations for training the models, thus allowing our multi-layer Gaussian process prediction models to estimate not only their presence but also their severities. The experimental validation of our approach demonstrates strong agreement of the model predictions with these PD annotations. Our results show the proposed method produces promising results in objective monitoring of movement abnormalities of PD in the presence of arbitrary and unknown voluntary motions, and makes an important step towards continuous monitoring of PD in the home environment.

Thanks to repetition, dustbathing detection can be automated combining accelerometry and wavelet analysis

“…Second, these authors used unsupervised cluster analysis to categorize each time point of the spectrum into one of 20 behaviour groups (Sakamoto et al, 2009). In human patients with Parkinson's disease, Lang et al (2019) showed behaviours, such as sitting and standing, appear in accelerometer recording of wrist movement as periodic at a different frequency than tremors. Thus, authors used wavelet analysis as a processing step for modelling the relationships between wrist movements during daily activities and clinical displays of movement abnormalities (Lang et al, 2019).…”

“…For example, Chimienti et al (2022) showed, in two penguin species, that by using a large sample size and including behavioural variability within training datasets resulted in a high level of accuracy (>80%) between the supervised and unsupervised approaches. The detection of tremors in Parkinson's patients with multi-layer Gaussian process prediction models showed an accuracy of 83% (Lang et al, 2019). Support vector machine (SVM) algorithms have shown overall mean accuracies between 80% and 87% and, when reported, a sensitivity of around 80% for classifying behavioural events correctly from acceleration data in different species such as penguins (Carroll et al, 2014), cows (Martiskainen et al, 2009) and broilers (Yang et al, 2021).…”

“…(2019) showed behaviours, such as sitting and standing, appear in accelerometer recording of wrist movement as periodic at a different frequency than tremors. Thus, authors used wavelet analysis as a processing step for modelling the relationships between wrist movements during daily activities and clinical displays of movement abnormalities (Lang et al., 2019). Noticeably, for dustbathing, repetitive sequences were observed for lower frequencies in the swaying ( a y ) profile, showing cycles every 25–60 s.…”

Parkinson’s Disease Tremor Detection in the Wild Using Wearable Accelerometers

“…Many researchers have explored the use of machine learning to automatically detect PD motor symptoms with wearable sensor [ 9 , 10 , 11 ], considering several motor symptoms at the same time in several cases [ 12 ]. Despite this strong interest, there are no commercial systems that provide a good user experience or sufficient accuracy to complement doctor assessments.…”

Detection and Quantification of Resting Tremor in Parkinson’s Disease Using Long-Term Acceleration Data

“…In addition, multiple methods such as joint time-frequency analysis, statistical analysis, and machine learning have been employed to study the tremor so far. Spectral analysis and time-frequency analysis were well utilized in several previous studies [18,19]. Salarian et al [20] used frequency and amplitude of the signals to detect and quantify resting tremor through fixed thresholding.…”