Trademark 3840-480

A boon for Parkinson's patients? New IMU and Machine Learning Based Remote Parkinson's Motion Screening Device

Parkinson's disease (PD) affects approximately 1 million people in the United States and more than 10 million people worldwide. Parkinson's disease is a chronic, incurable disease degenerative disorder that requires close monitoring of patients by clinicians, especially specialists in movement disorders. Physicians often use standard clinical instruments such as the Unified Parkinson's Disease Rating Scale (UPDRS). Typically, each Parkinson's patient is required to visit the clinician's office several times a year to have their condition evaluated. This is a significant burden for Parkinson's patients.

Shehjar Sadhu's team in the US has designed a machine learning based telehealth device that remotely detects and categorizes hand movements using UPDRS tasks. The system consists of Edge Node and Fog Node. where Edge Node records hand activity using a pair of smart gloves with integrated finger flexion sensors and an inertial measurement unit (IMU) and wirelessly transmits the data to the Fog Node for classification. the Fog Node runs a machine learning (ML)-based activity classification model to classify the UPDRS-based hand movement task for classification.

A total of 9 volunteers with smart gloves participated in the testing of this program, 5 of them were healthy people and 4 were Parkinson's patients. Meanwhile the team developed and tested different classification models such as K Nearest Neighbor Method (KNN), Support Vector Machine (SVM) and Decision Tree (DT) to select the one that is most suitable to be used for remote symptom assessment algorithm. The team split the test data into three subsets, training dataset, testing dataset, and validation dataset, with data volume shares of 80%, 10%, and 10%, respectively.The accuracy of the training dataset, the testing dataset, and the validation dataset under the SVM model is 94%, 93%, and 93%, respectively, and the accuracy on the FogNode with an average computation time of 560µs, outperforming other models.

Hand data collection
UPDRS movements of the upper extremities (visualization)
Data Processing Flow

Unfortunately the above tests and data were performed in a laboratory environment. It is expected that in the future, the system can be deployed to real-life situations to provide easier and more accurate diagnosis and treatment for Parkinson's patients.