Researchers associated with the Physical Activity Laboratory (Movi-Lab) at São Paulo State University (UNESP) in Bauru, Brazil, are using artificial intelligence to help diagnose Parkinson’s disease and assess its progression.
Article published in the magazine Gait and posture reports the results of a study in which a machine learning algorithm diagnosed cases of the disease by analyzing spatial and temporal gait parameters.
The researchers identified four gait characteristics most relevant to Parkinson’s diagnosis: stride length, speed, width, and consistency (or variation in width). To assess the severity of the disease, the most important factors were step width variation and double support time (while both feet are in contact with the ground).
“Our study was innovative compared to the scientific literature by using a larger database than usual for diagnostic purposes. We chose gait parameters as a key criterion because gait impairment occurs early in Parkinson’s disease and worsens over time, and also because they are not associated with physiological parameters such as and age, height and weight,” said Fabio Augusto Barbieri, co-author of the paper, to Agência FAPESP. Barbieri is a professor in the Sports Department of the UNESP Faculty of Sciences (FC).
The study sample consisted of 63 participants in Ativa Parkinson, a multidisciplinary structured exercise program for Parkinson’s patients implemented at FC-UNESP, and 63 healthy controls. All volunteers were over 50 years old. Data was collected and fed into the repository used in the machine learning processes for seven years.
Baseline estimates were produced by analyzing gait parameters for the healthy control group and comparing them to expected values for this age group. This involved using a special kinematics camera to measure each person’s stride for length, width, length, speed, cadence and such as double support time, as well as stride variability and asymmetry.
The researchers used the data to create two different machine learning models – one to diagnose the disease and the other to assess its severity based on the patient’s assessment. Researchers at the Faculty of Engineering at the University of Porto in Portugal participated in this part of the study.
They ran the data through six algorithms: Naïve Bayes (NB), Support Vector Machine (SVM), Decision Tree (DT), Random Forest (RF), Logistic Regression (LR) and Multilayer Perceptron (MLP). NB achieved a diagnostic accuracy of 84.6%, but NB and RF performed best in severity assessment.
“Typical accuracy for clinical assessment is around 80%. We could significantly reduce the chance of diagnostic errors by combining clinical assessment with artificial intelligence,” said Barbieri.
Parkinson’s disease is caused, at least in part, by the degeneration of neurons in the areas of the brain that control movement, due to a lack of dopamine production. Dopamine is the neurotransmitter that sends signals to the extremities. Low dopamine levels impair movement, causing symptoms such as tremors, slow gait, stiffness and poor balance, as well as changes in speech and writing.
Diagnosis is now based on the patient’s clinical history and neurological examination, without specific tests. Exact information is not available, but it is estimated that 3%-4% of the population over the age of 65 has Parkinson’s disease.
According to another co-author, Ph.D. candidate Tiago Penedo, whose research is supervised by Barbieri, the results of the study will be useful to improve diagnostic evaluation in the future, but cost could be a limiting factor. “We made progress with the tool and contributed to the expansion of the database, but we used expensive equipment that is difficult to find in clinics and doctors’ offices,” he said.
The equipment used in the study costs about US$100,000. “Gait analysis can be done with cheaper methods, using a timer, a force plate and so on, but the results are not accurate,” Penedo said.
The methods used in the study can contribute to a better understanding of the mechanisms underlying the disease, especially gait patterns, the researchers believe.
A previous study, reported in a paper published in 2021, with Barbieri as the last author, showed a 53% lower step length synergy in overcoming obstacles in Parkinson’s patients than in healthy subjects of the same age and weight. Synergy in this case refers to the ability of a movement (or musculoskeletal system) to adapt movements and integrate factors such as speed and foot position, while stepping off a curb, for example.
Another study, also published in Gait and posture, showed that Parkinson’s patients were less able to maintain postural and tremor stability than their neurologically healthy peers. The authors said the findings provide new insights to explain the larger, faster and more variable oscillations seen in Parkinson’s patients.
Marta Isabel ASN Ferreira et al., Machine Learning Models for Parkinson’s Disease Detection and Scoring Based on Spatial and Temporal Gait Parameters, Gait and posture (2022). DOI: 10.1016/j.gaitpost.2022.08.014
Elisa de Carvalho Costa et al., Multi-domain postural control assessment in people with Parkinson’s disease: conventional, non-linear and disturbance and tremor trajectory analysis, Gait and posture (2022). DOI: 10.1016/j.gaitpost.2022.07.250
Quotation: Artificial intelligence helps detect gait changes and diagnose Parkinson’s disease (2022, November 23) Retrieved November 23, 2022 from https://medicalxpress.com/news/2022-11-artificial-intelligence-gait-parkinson-disease. html
This document is subject to copyright. Except for reasonable communications for the purpose of private research or investigation, no part may be reproduced without written permission. The material is provided for informational purposes only.