Classification of Parkinson’s disease and essential tremor based on gait and balance characteristics from wearable motion sensors: A data-driven approach

Abstract

Parkinson's disease (PD) and essential tremor (ET) are movement disorders that can have similar clinical characteristics including tremor and gait difficulty. These disorders can be misdiagnosed leading to delay in appropriate treatment. The aim of the study was to determine whether gait and balance variables obtained with wearable sensors can be utilized to differentiate between PD and ET using machine learning techniques. Additionally, we compared classification performances of several machine learning models. A balance and gait data set collected from 567 people with PD or ET was investigated. Performance of several machine learning techniques including neural networks (NN), support vector machine (SVM), k-nearest neighbor (kNN), decision tree (DT), random forest (RF), and gradient boosting (GB), were compared using F1-scores. Machine learning models classified PD and ET based on balance and gait characteristics better than chance or logistic regression. The highest F1-score was 0.61 of NN, followed by 0.59 of GB, 0.56 of RF, 0.55 of SVM, 0.53 of DT, and 0.49 of kNN. The results demonstrated the utility of machine learning models to classify different movement disorders. Further study will provide a more accurate clinical tool to help clinical decision-making.

A deep explainable artificial intelligent framework for neurological disorders discrimination

“…In other words, NeurDNet offers a more cost-effective solution, which paves the way for utilization of wearable and commercial sensors for the purpose of diagnosis. Finally, it is worth highlighting that the overall classification accuracy of the NeurDNet ( 95.55% ) is higher than the best accuracy reported in reference 53 54 is performed by the built-in accelerometer of a mobile phone, which can potentially offer a cost-effective and accessible solution, there are potential issues regarding generalization of the introduced model over the wide range of characteristics in Parkinsonian and Essential tremor. To be more specific, by considering the tabulated performance results in this article, a huge variation between the average performance and the best performance is observed ( ≥ 20% ), which does not assure a robust and reliable generalization over the training set.…”

“…It also offers a novel machine intelligence pipeline which can be interpreted from the clinical point of view. Considering the predecessors of NeurDNet with the highest classification accuracies (before the invention of NeurDNET in this paper), i.e., References 15,49,[52][53][54]63 , it is understood that NeurDNet not only excels the classification accuracy of the research that is based on accelerometer data but also outperforms the one based on Electromyogram (EMG) signals recorded from a tremorous hand (which was supposed to have richer neurophysiological content in the signal). To be more specific, here we provide an itemized comparison with recent research publications, leading the state-of-the-art classification accuracy for discriminating PD from ET.…”

“…Based on this argument, we believe that the reported accuracy of the framework in reference 52 ( 95.8% ), although numerically comparable with the one for NeurDNet, falls far behind NeurDNet in terms of reliability and generalization. -The work by Moon et al 53 presents a set of processing techniques to address the problem of PD-ET classification based on several hand-crafted features extracted from motion signals recorded from patients with PD and ET. Although the size of the employed dataset is much larger than the one for NeurDNet (567 patients compared to 81 patients), there is a tremendous imbalance between the PD cases ( n = 524 ) and ET cases ( n = 43 ) in the article.…”

“…In fact, we believe that using synthesized samples is somehow equivalent to leakage of information from the training set into the test set, which eventually may result in providing misleading information about performance for the model. Besides, the proposed methodology in reference 53 requires several inertial units to be mounted on the body of patients, which requires medical-grade acquisition devices. However, NeurDNet is developed based on inertial measurements from on accelerometer mounted on the dorsum of hand.…”

“…Moon et al 53 ET/PD classification Gain and balance characteristics, [524 PD, 43 ET] for training and testing Hand-crafted features and classical ML Accuracy = 92%…”