KITE scientists use AI to develop at home assessment tool for Parkinson’s patients

This could give clinicians the ability to adjust a patient’s treatment earlier and more effectively.

TORONTO–People living with Parkinson’s may one day be able to get their condition evaluated from the comfort of their own home thanks to a vision-based machine learning model developed by a team at the KITE Research Institute. 

The model was able to detect differences in movement between Parkinson patients who were medicated and unmedicated.  This could lead to improved care for people living with Parkinson’s disease.

Currently, clinicians rely on formal evaluations during visits to detect changes in how a person living with Parkinson's moves. 

Automated assessments, which would rely on vision-based machine learning models, could be used in home or long term care settings to detect changes more frequently. 

This will allow clinicians to adjust a patient’s treatment earlier and more effectively. 

The team tested the model and published the results in Biomedical Engineering OnLine.

The International Conference on Aging, Innovation & Rehabilitation (ICAIR) partnered with BioMedical Engineering OnLine to publish a special edition of the journal featuring full-length papers of the highest scoring abstracts from the conference.

The paper’s first author KITE research associate Andrea Sabo goes in depth on her team’s findings below. 

Which patient groups are most affected by this?

In this work, we trained our model on a cohort of older adults with dementia and drug-induced parkinsonism (DIP) and evaluated it on a population of adults with idiopathic Parkinson’s disease (PD). This work has the potential to improve the treatment of individuals with both DIP and PD. 

What did you find? 

A vision-based machine learning model trained to predict severity of parkinsonism in gait in older adults with drug-induced parkinisonism was poor at predicting clinical severity of parkinsonism in gait in a separate cohort of adults with Parkinson’s disease (PD). However, statistically significant differences in model-predicted clinicial scores of parkinsonism in the ON and OFF medication/ deep brain stimulation (DBS) treatment conditions were found in the PD cohort.

 Why does this matter? 

These results indicate that the model, which was trained on an independent dataset consisting of a different clinical population and collected at a different site, is not precisely calibrated to the clinical rating scale, it captures clinically relavent features of parkinsonism in gait that are present in the test cohort. This in the first study to evaluate a model for estimating parkinsonism severity across datasets.

 What is the potential impact?

Automated vision-based assessment can be used to monitor fluctuations in parkinsonian characteristics in gait in home or long-term care settings. Any changes can be identified sooner as patients can be assessed more frequently than the current practice which requires clinical visits for formal assessments. Timely detection of gait changes allows clinicians to adjust the patient’s medication use and DBS parameters more proactively.

Research Spotlight: 
Evaluating AI's ability to detect changes in gait

Research Associate at the KITE Research Institute

Name of Publication:
Evaluating the ability of a predictive vision-based machine learning model to measure changes in gait in response to medication and DBS within individuals with Parkinson’s disease

Name of Journal:
BioMedical Engineering OnLine