Machine Learning Catches Multiple Sclerosis-Related Gait Issues

Machine learning could help providers monitor and predict multiple sclerosis disease progression in individual patients, according to a study published in Institute of Electrical and Electronics Engineers Transactions on Biomedical Engineering.

Multiple sclerosis (MS) is one of the most common neurological conditions worldwide, and is especially prevalent in among people 50 to 60 years old, researchers stated. MS can present itself in many ways, but walking problems are a common symptom. About half of patients need walking assistance within 15 years of disease onset.

The team set out to better understand the disability-related changes that can come with MS, and evaluate the effectiveness of a gait data-based machine learning tool for MS prediction.

“We wanted to get a sense of the interactions between aging and concurrent MS disease-related changes, and whether we can also differentiate between the two in older adults with MS,” said Manuel Hernandez, kinesiology and community health professor at University of Illinois.

“Machine learning techniques seem to work particularly well at spotting complex hidden changes in performance. We hypothesized that these analysis techniques might also be useful in predicting sudden gait changes in persons with MS.”

Using an instrumental treadmill, researchers collected gait data that was normalized for body size and demographics from 20 adults with MS, as well as 20 age-, weight-, height-, and gender-matched older adults without MS.

Participants walked at a comfortable pace for up to 75 seconds while specialized software captured gait events, corresponding ground reaction forces, and center-of-pressure positions during each walk. The team then extracted each participant’s characteristic spatial, temporal, and kinetic features in their strides to examine variations in gait during each trial.

Changes in various gait features, including a data feature called the butterfly diagram, enabled the team to identify differences in gait patterns between participants. The diagram gets its name from the butterfly-shaped curve created from the repeated center-of-pressure trajectory for multiple continuous strides during a subject’s walk, and is associated with critical neurological functions.

“We study the effectiveness of a gait dynamics-based machine-learning framework to classify strides of older persons with MS from healthy controls to generalize across different walking tasks and over new subjects,” said Rachneet Kaur, a graduate student at University of Illinois Urbana Champaign.

“This proposed methodology is an advancement toward developing an assessment marker for medical professionals to predict older people with MS who are likely to have a worsening of symptoms in the near term.”

Future studies could provide more thorough examinations to manage the study’s small cohort size, researchers noted.

“Biomechanical systems, such as walking, are poorly modeled systems, making it difficult to spot problems in a clinical setting,” said Richard Sowers, mathematics professor at the University of Illinois.

“In this study, we are trying to extract conclusions from data sets that include many measurements of each individual, but a small number of individuals. The results of this study make significant headway in the area of clinical machine learning-based disease-prediction strategies.”

Advanced analytics technologies have increasingly emerged as viable tools to improve the management and diagnosis of neurological conditions. A team from the University of Florida recently announced that it would use a $5 million grant from NIH to test a new artificial intelligence tool that aims to enhance the diagnosis of Parkinson’s and related conditions.

The AI tool could help providers distinguish between three distinct neurodegenerative disorders: Parkinson’s disease, multiple system atrophy Parkinsonian variant (MSAp), and progressive supranuclear palsy (PSP).

“What is new is the use of artificial intelligence for predicting the type of Parkinsonism,” said Angelos Barmpoutis, PhD, an associate professor and coordinator of research and technology at UF’s Digital Worlds Institute.

“In order to train a computer to identify Parkinsonism, we need to teach it using a lot of data. One solution for that is crowd sourcing — going around to different institutes that have patients and asking them to contribute to this big project. We try to collect as many data points by creating what I believe is one of the largest databases for this particular type of diagnosis.”