September 18, 2022
A Unesp study in Bauru compares the digital tools used in the tests and identifies the most effective. The development of technology should improve the accuracy of diagnosis and the recognition of the stage of pathology.
Malena Stariolo – Unesp Journal
Parkinson’s disease is known to cause impairments in motor functions, which manifests in symptoms such as tremors, slow movements, difficulty walking and imbalance. But the pathology also impairs several non-motor functions, causing sleep or wakefulness disorders, body aches and sensory disturbances. According to estimates by the World Health Organization (WHO), in 2019 more than 8.5 million people suffered from this disease worldwide. And, worryingly, functional disabilities and deaths from Parkinson’s disease have increased faster than those from other neurological diseases, according to the WHO itself.
Despite the global reach of the disease, there are no precise data on the number of people suffering from this disease in Brazil. The lack of data makes it difficult to analyze its real impact.
One of the probable causes of the lack of data is the difficulty of diagnosis. Even today, the most common method is the clinical evaluation of the patient, through the analysis of the symptoms that affect his motricity, and the use of magnetic resonance imaging. This approach, however, makes it difficult to distinguish between Parkinson’s disease and other degenerative neurological conditions that generate similar symptoms, and often leads to misdiagnosis.
According to Fabio Augusto Barbieri, professor and researcher at the Department of Physical Education of the Faculty of Sciences of Unesp, Bauru campus, and coordinator of the Human Movement Research Laboratory (MOVI-LAB), the error rate of diagnosis is about 15%, a number considered high for an increasingly common disease. “The error rate is high because it is a clinical assessment. Although neurologists use validated criteria and follow a checklist to make the diagnosis, there is a subjective component to the process,” says Barbieri.
Moreover, in cases where the diagnosis is positive, current methods are also inaccurate in identifying the stage of development of Parkinson’s disease. And this is reflected in the indication for treatment. “If the MRI is not directed at the point where the degeneration of the disease occurs, it will also show an error. Normally, the degenerative process caused by Parkinson’s disease is thought to occur in the basal ganglia , but it can also affect other regions,” he says.
Given the need to find more accurate methods of diagnosing and monitoring the disease, researchers around the world have turned to using techniques of machine learning.
Also known as machine learning, the machine learning is a branch of the application of Artificial Intelligence (AI). In these systems, the algorithms are fed by data from observations, with variable quantities and characteristics. Algorithms use statistical methods to classify and analyze incoming data to find patterns and identify new information. Applied to medicine, the use of machine learning can mean more accurate diagnoses, allowing for more effective treatments.
In August of this year, Barbieri was one of the authors of the article “Machine learning models for Parkinson’s disease detection and stage classification based on spatiotemporal gait parameters», published in the journal Gait and posture. The objective of the study was to test several algorithmic languages to identify greater accuracy in the diagnosis of Parkinson’s disease and in the identification of its stages. The research was carried out with researchers from the University of Porto, Portugal, and graduate students from the Movement Sciences program who participate in the MOVI-LAB, coordinated by Barbieri.
“Our main goal was to use patient gait variables to try to more accurately determine the diagnosis of Parkinson’s disease, i.e. to differentiate who has the disease from those who do not. . In addition, we also wanted to control the progression of the disease,” explains the researcher.
Seven years of data to assess
The differential of the new study lies in the observation of variables more related to the so-called gait deficit, such as gait variability and the so-called asymmetry. Gait variability refers to differences in stride width. The asymmetry is related to the spread of the disease. It is estimated that in 95% of cases, the problem only occurs on one side of the body, which then begins to show symptoms.
The data supporting the study was gathered from a laboratory database built up over seven years of research with volunteer patients. These people were supported in the Ativa Parkinson extension project, also coordinated by Babieri. For the article, the variables of interest from the different studies were grouped together in order to obtain a large sample.
A total of 126 people were analyzed, both patients diagnosed with Parkinson’s disease and neurologically healthy people. These formed the control group. “Thanks to the control group, we can check what is expected for this age in terms of walking performance, the so-called baseline. Thus, we can compare sick patients with those in the control group and verify the effects, both of the disease and of the interventions,” explains the professor.
Other data measured included the width, length, duration, speed and cadence of each individual’s steps, as well as information such as how long each person had one foot on the ground and how long the individuals had one foot on the ground. on the floor. The information was then applied to two different models of machine learning: the first focused on the diagnosis of Parkinson’s disease and the second on the identification of the stage of the disease.
From these two models, five algorithms were analyzed, called Naïve Baise (NB), Support Vector Machine (SVM), Decision Tree (DT), Random Forest (RF), Logistic Regression (LR) and Multilayer Perceptron (MLP) , which were used the most in previous similar tests.
Each algorithm was able to select, among the various parameters retained, those which could provide a more precise analysis. The Naïve Bayes algorithm selected only four pieces of information: step length, speed, width and step width variation. With this data, the algorithmic language achieved an accuracy of 84.6% in correctly diagnosing people with Parkinson’s disease. In comparison, Random Forest chose twelve pieces of information, with step length, step speed variation, and step speed being the most important.
On the other hand, when identifying the stage of the disease, Naïve Bayes and Random Forest stood out for having the highest AUC, which indicates the ability to classify language, and the highest accuracy, respectively. . The two most important pieces of information for analyzing disease stage were the variability in step width and the variation in the time people had both feet on the ground.
Despite the good clues, the use of machine learning it will not imply the abandonment of clinical analysis techniques. It is an additional tool to increase the accuracy of diagnoses and facilitate the definition of an appropriate treatment. In addition to diagnosis, digital resources should provide new insights into disease mechanisms and help identify previously unknown gait patterns.
Access to state-of-the-art treatments at university
Barbieri is also coordinator of the Unesp extension project Activates Parkinson’s disease, in partnership with the Department of Psychology of Unesp, Bauru, under the responsibility of Professor Marianne Ramos Feijó. In the project, which in addition to Barbieri and Feijó involves other professors, researchers and students from Unesp, participants have access to physical activities and psychological counseling, in addition to other activities designed to promote the good – be people with Parkinson’s disease. Participation is free and meetings take place on Tuesdays and Thursdays from 8:30 a.m. to 9:30 a.m. Barbieri stresses that all the activities carried out with patients are based on current scientific studies, which, in addition to the multidisciplinary nature of the care offered by the team, places the treatments at the forefront of knowledge on Parkinson’s disease.
“Many participants in the studies we have developed belong to this extension project. In this way, they also help other patients, because the knowledge we produce is open to other groups and thus improves the quality of life,” says Barbieri. “It’s also a way for the university to give back to society. Our patients can have access to some of the best treatments, as well as the opportunity to learn a little more about research. Who knows, in the future we may find a cure for the disease or at least develop increasingly effective treatments,” he says.
share this news