PHTNet: Characterization and Deep Mining of Involuntary Pathological Hand Tremor using Recurrent Neural Network Models

Abstract

the global aging phenomenon has increased the number of individuals with age-related neurological movement disorders including parkinson's Disease (pD) and essential tremor (et). pathological Hand tremor (pHt), which is considered among the most common motor symptoms of such disorders, can severely affect patients' independence and quality of life. To develop advanced rehabilitation and assistive technologies, accurate estimation/prediction of nonstationary pHt is critical, however, the required level of accuracy has not yet been achieved. The lack of sizable datasets and generalizable modeling techniques that can fully represent the spectrotemporal characteristics of PHT have been a critical bottleneck in attaining this goal. this paper addresses this unmet need through establishing a deep recurrent model to predict and eliminate the PHT component of hand motion. More specifically, we propose a machine learning-based, assumption-free, and real-time pHt elimination framework, the pHtnet, by incorporating deep bidirectional recurrent neural networks. the pHtnet is developed over a hand motion dataset of 81 ET and PD patients collected systematically in a movement disorders clinic over 3 years. The PHTNet is the first intelligent systems model developed on this scale for pHt elimination that maximizes the resolution of estimation and allows for prediction of future and upcoming sub-movements.Age-related neurological movement disorders such as Parkinson's Disease (PD) and Essential tremor (ET) 1-4 are expected to become more prevalent as the population of seniors over the age of sixty is expected to increase from 962 million in 2017 to 2.1 billion by 2050, and to 3.1 billion in 2100 5 . Pathological Hand Tremor (PHT) is a common upper-limb motor symptom of several age-related neurological movement disorders and is described as involuntary and pseudo-rhythmic movements 6 affecting coordination, targeting, and speed of intended motions 7 .Unlike physiological tremor, which is identified with low amplitude vibrations occurring within the spectral range of 6 to 14 Hz 8 and affects the performance of individuals in high precision tasks such as robotic surgery 9 , PHT represents higher amplitude motion occurring in the broader frequency range of 3-14 Hz 10 . The repetitive and oscillating nature of PHT differentiates itself from other involuntary movement disorders such as chorea, athetosis, ballism, tics, and myoclonus 11 . Upper-limb tremor significantly limits individuals in performing Activities of Daily Livings (ADLs) 12 . Thus, during the last decade, several techniques and technologies have been proposed in both rehabilitation and assistive domains 13,14 to compensate for the involuntary movement while promoting the voluntary component of motion. The accuracy of a tremor compensation technology (such as sophisticated wearable exosuits) relies significantly on the efficacy and spectrotemporal resolution of the algorithm, as inaccurate or slow extraction techniques do not allow for proper compensation. PHT consists o...

A deep explainable artificial intelligent framework for neurological disorders discrimination

“…As a final note to our discussion, it is worth comparing the NeurDNet with our previously developed PHTNet framework 25 . Below, we provide a point-by-point comparison between the two works:…”

“…Dugue et al 54 ET/PD classification Accelerometer data, [17 PD,16 recognition 45 , justifying the growing trend of their application in other fields, e.g., tremor assessment 25,40,41,[46][47][48] .…”

“…As the action tremor recordings contain dynamic features from both the person's voluntary movement and the tremorous movements, we believe that the wide range of characteristics and dynamic properties of the voluntary component misleads NeurDNet in the classification tasks and degrades its accuracy. There is a general consensus in the literature of PHT processing works 25,57,58 that the recorded PHT signals can be modeled as the summation of a voluntary and an involuntary component. A major bottleneck in distinguishing between the two components is that no ground truth can be assumed for either of the two components.…”

“…To use time-series recordings for differentiating PD patients from those with similar symptoms but with different diagnosis, several classification schemes are developed over recent years in the literature [25][26][27][28][29][30][31][32][33][34] , including statistical signal processing 35 , support vector machines (SVM) 36 , Naive Bayes classifiers, nearest centroid classifier (NCC), random forest (RF) 37 , decision tree (DT), and linear discriminant analysis (LDA) 38,39 . More recently, deep learning (DL) methods, which are considered as a subcategory of ML techniques and present methodologies to design multi-layer artificial neural networks (ANN) are employed to analyze the tremor signals 40,41 .…”

Real-Time Voluntary Motion Prediction and Parkinson’s Tremor Reduction Using Deep Neural Networks

“…Recently, the WAKE framework [28] was proposed, based on wavelet decomposition and adaptive Kalman filtering; however, it is limited to estimation and one-step prediction. Another framework, PHTNet has been recently proposed by Shahtalebi et al [29]. It is limited to the prediction of only one step ahead, and it introduces a time delay of 100 and 150 ms before it can properly estimate the signal.…”