Wearable-based Parkinson's Disease Severity Monitoring using Deep Learning

Abstract

One major challenge in the medication of Parkinson's disease is that the severity of the disease, reflected in the patients' motor state, cannot be measured using accessible biomarkers. Therefore, we develop and examine a variety of statistical models to detect the motor state of such patients based on sensor data from a wearable device. We find that deep learning models consistently outperform a classical machine learning model applied on hand-crafted features in this time series classification task. Furthermore, our results suggest that treating this problem as a regression instead of an ordinal regression or a classification task is most appropriate. For consistent model evaluation and training, we adopt the leave-one-subject-out validation scheme to the training of deep learning models. We also employ a class-weighting scheme to successfully mitigate the problem of high multi-class imbalances in this domain. In addition, we propose a customized performance measure that reflects the requirements of the involved medical staff on the model. To solve the problem of limited availability of high quality training data, we propose a transfer learning technique which helps to improve model performance substantially. Our results suggest that deep learning techniques offer a high potential to autonomously detect motor states of patients with Parkinson's disease.

Granular Motor State Monitoring of Free Living Parkinson's Disease Patients via Deep Learning

“…Thus, it accepts all predictions that lie within this uncertainty bound. In addition, this custom loss was designed for the diagnosis of PD-related symptoms integrating the following clinical requirements [1]: (1) Non-linearity of labels: misclassifying a sample with a label that is two levels away is more than twice as bad as misclassifying by only one level (2) Asymmetry of labels: an exaggerated diagnosis in the true pathological direction should be better than an opposing diagnosis in the wrong pathological direction, even if the label difference between them is the same and (3) Misclassification cost: the cost of misclassifying a patient in the wrong direction is proportional to the motor severity.…”

“…This paper is an extension of a previous work [1] whithin which we compared the performances of classification, ordinal classification and regression on the motor fluctuation estimation problem using state-of-art time series classifcation (TSC) architectures. Therein, we concluded that the default FCN architecture in a regression setting performs best among several baseline methods.…”

“…Finally, previous work often introduced impractical settings for the real-world usage in the PD treatment, which would not match the free-living conditions of patients such as the placement of several sensors over the body to instructing a limited set of specific actions to patients [8,13,22,24]. See [1] for a more detailed literature overview.…”

Parkinson’s Disease Severity Estimation using Deep Learning and Cloud Technology

“…However, these devices have issues with respect to long battery lifetime and lack of cloud based assessment. Lately, researchers have concentrated on the prediction of cardinal motor symptoms, evaluating the progression of the disease by using DL that outperform a classical ML model applied on hand-crafted features in the time series classification task [14], [15]. As in [16]- [18] the authors main objective is to use deep brain simulation and inertial sensors data from single PD subject to quantify PD hand tremors.…”