Predicting Multiple Sclerosis From Gait Dynamics Using an Instrumented Treadmill: A Machine Learning Approach.

Area of Science:

• Neurology
• Biomedical Engineering
• Data Science

Background:

• Multiple Sclerosis (MS) is a heterogeneous neurological condition often presenting with mobility limitations.
• Gait disturbances are a common symptom in persons with MS (PwMS), impacting quality of life.
• Objective identification of MS through gait analysis can aid in patient monitoring.

Purpose of the Study:

• To develop and evaluate a machine learning (ML) framework for identifying MS using spatiotemporal and kinetic gait features.
• To compare data normalization strategies for optimizing ML-based gait analysis in MS.
• To assess the generalizability of the ML model across different walking tasks and subjects.

Main Methods:

• Collected gait data from 20 PwMS and 20 healthy older adults (HOA) using an instrumented treadmill.
• Explored size-normalization and regress-normalization techniques for gait data.
• Developed an ML methodology to classify individual strides, employing gradient boosting machine and multilayer perceptron models.

Main Results:

• Regress-normalization significantly improved ML accuracy for pathological gait identification compared to size-normalization.
• Gradient boosting achieved 94.3% accuracy and 1.0 AUC for classifying walking tasks.
• Multilayer perceptron achieved 80% accuracy and 0.86 AUC for subject generalization using regression-normalized data.

Conclusions:

• Integrating gait data and ML offers a patient-centric approach for MS monitoring.
• Regression-normalized gait features show promise for ML-based disease prediction and progression monitoring in PwMS.