Skip to main content
LATEST INFORMATION ABOUT COVID-19

READ MORE

Gina Cody School of Engineering and Computer Science

A PILLAR OF CONCORDIA'S STRATEGIC DIRECTIONS

Gina Cody School of Engineering and Computer Science

A PILLAR OF CONCORDIA'S STRATEGIC DIRECTIONS

Concordia-led researchers study pathological hand tremors in patients to develop a machine learning-based treatment framework

People suffering from Parkinson’s and other neurodegenerative diseases will benefit from smarter, more accurate technology
March 3, 2020
|
The researchers used a dataset of hand motions gathered from 81 neurodegenerative patients over four years.

Diagnosed cases of Parkinson’s disease and other neurodegenerative disorders are likely to increase dramatically as Canada’s population ages. There are more than 100,000 Canadians living with Parkinson’s today, but that number is expected to jump over the coming decade to more than 160,000. The average age of onset is around 60, and with older adults now outnumbering children under 14, dealing with neurodegenerative disorders takes on an increasing urgency.

There is no cure for these diseases. However, there are pharmacological, therapeutic and neurosurgical treatments that can mitigate some effects of the disease, most notably pathological hand tremors (PHT). In a new paper published in the Springer Nature journal Scientific Reports, researchers propose a neural network model that they say can assist rehabilitative and assistive technologies for patients with PHT associated with neurodegenerative diseases like Parkinson’s or essential tremors.

The researchers used a dataset of hand motions gathered from 81 neurodegenerative patients over four years at the University of Western Ontario to build a new framework model they call PHTNet. Arash Mohammadi, assistant professor at the Concordia Institute for Information Systems Engineering, supervised the project. His PhD student Soroosh Shahtalebi is the lead author of the study.

The study's other collaborators include Farokh Atashzar, assistant professor at New York University; Rajni Patel, professor at the University of Western Ontario and director of engineering at Canadian Surgical Technologies and Advanced Robotics; and Mandar Jog, professor at the University of Western Ontario, director of the Parkinson's Foundation Centre of Excellence in Movement Disorders and neurologist at the London Movement Disorders Centre.

Olivia Samotus, a PhD student at the London Movement Disorders Centre and the University of Western Ontario, collected the research data and is also a co-author of the paper.

Overcoming over-predicting

The recruited subjects were undergoing therapy such as robot-assisted arm training to support or direct their movements. These arms calculate the force the patients want to apply and magnify or resist it as needed. This technique requires the arm to dampen or compensate for involuntary hand motions (tremors), and it relies on prediction modelling in order to provide safe and effective treatment.  

The existing models calculate the frequency contents of the hand motion signal to identify and extract involuntary motions. If a therapeutic robot fails to correctly calculate and remove involuntary motion, it will automatically magnify the unwanted motion, and so degrade the technology’s performance and possibly affect user safety. This can be especially problematic because, as the authors point out, there can be significant overlap between voluntary and involuntary movements in frequency domain.

In order to improve the network’s learning abilities, the researchers took nearly 90 hours’ worth of kinematic motion recordings from the 81 patients in a clinical setting. This extra data — Mohammadi says it is one of only two datasets that he knows of — helped them efficiently teach the network how to differentiate between voluntary and involuntary movements. It also allowed the network to operate with a much wider range of possibilities and give it extra accuracy when predicting voluntary hand-motion signals. The model is accurate enough to provide one-sample ahead predictions of voluntary hand movements.

“PHTNet is not considered a cure to Parkinson’s disease, but it can help patients who are using assistive and rehabilitative devices,” says Shahtalebi. “In order for these technologies to work effectively, we need to understand what is happening with the patient’s movement and how the dynamics of those movements are formulated.”

Mohammadi adds that as societies like Canada’s age, we are going to see more age-related neurological issues arise. “Having these technologies is going to prove very timely.”

The ultimate goal is moving toward development of autonomous assistive systems for remote rehabilitation and effective in-home therapy. An autonomous assistive system with artificial intelligence and consciousness can perform specific therapeutic tasks with a high degree of accuracy and autonomy.

Concordia is hosting the 2020 IEEE International Conference on Autonomous Systems (ICAS) from August 12 to 14.

The Natural Sciences and Engineering Research Council of Canada (NSERC) partially funded this study.


Read the cited paper: “PHTNet: Characterization and Deep Mining of Involuntary Pathological Hand Tremor using Recurrent Neural Network Models.”



Back to top Back to top

© Concordia University