Scientists from the Human Movement Research Laboratory (Movi-Lab), based in the Bauru/SP unit of Unesp, used AI (artificial intelligence) to help diagnose and determine the stage of Parkinson’s Disease in volunteers.
The research was published in the journal Gait & Posture and shows that machine learning algorithms can help in recognition by analyzing spatio-temporal parameters in the path of the analyzed.
There are four important characteristics for diagnosis: length, speed, width and consistency of step width, called variability. In determining the stage of the disease, the difference in step width and the time that the person has both feet on the ground (double support) are two prominent factors.
“Our study makes a difference compared to the scientific literature: we use a larger database to make the diagnosis. We chose gait as a parameter because we believe that gait is one of the most compromised factors in patients with with Parkinson’s and does not involve physiological symptoms”, explained Fabio Augusto Barbieri, co-author of the article and professor in the Department of Physical Education in the Faculty of Sciences at Unesp.
Sixty-three patients from Ativa Parkinson – a multidisciplinary physical activity project aimed at diagnosed patients – participated in the study, in addition to another 63 healthy people, all over 50 years old . The data was collected over seven years and sent to the bank which was used in the machine learning process.
Based on information from healthy people, scientists set the so-called baseline, which indicates the expected parameters of gait performance for the age group analyzed. The width, length, duration, speed and rhythm of each individual’s steps were measured, as well as information such as the time each had one foot on the ground and both feet on the ground, the variation in gait and the asymmetry between steps..
The group used the data to create two different models for machine learning – one for diagnosing the disease and the other for determining its stage in patients. At this stage, the researchers had the participation of colleagues from the Faculty of Engineering of the University of Porto (Portugal).
Five algorithms were evaluated: Naïve Baise (NB), Support Vector Machine (SVM), Decision Tree (DT), Random Forest (RF) and Logistic Regression (LR). The NB algorithm achieved an accuracy of 84.6% in diagnosing Parkinson’s disease. For identifying the stage of the disease, the NB and RF algorithms showed the greatest hits.
“Generally, clinical assessments bring an accuracy of around 80%. If we manage to combine the clinic with artificial intelligence, it will be possible to significantly reduce the chance of misdiagnosis,” Barbieri told Agência FAPESP.
With information from UOL