A personalized methodology based on continuous gait monitoring and OC-SVM technique detects significant changes in functional state of assistive-device users, achieving accuracy of 70-97% in healthy individuals with simulated functional states and an average accuracy of 78% in a one-year longitudinal study with post-stroke individuals.
Key Findings
Results
The proposed OC-SVM-based methodology achieved accuracy in the range of 70-97% when validated in nine healthy individuals with simulated functional states.
Nine healthy participants were used for the validation with simulated functional states.
The accuracy range of 70-97% reflects variability across different individuals and simulated conditions.
The methodology generates personalized models for each individual rather than a generalized model.
One-Class Support Vector Machine (OC-SVM) was the core machine learning technique used.
Results
In a one-year longitudinal study with three post-stroke individuals, the methodology achieved an average accuracy of 78% in detecting changes in functional state.
Three post-stroke individuals participated in the real-case validation study.
The study duration was one year, providing longitudinal gait monitoring data.
All three participants required an assistive device for walking.
The 78% average accuracy was obtained under real-world conditions as opposed to controlled simulated conditions.
Methods
The methodology addresses inter-individual gait variability by generating personalized models for each patient using the OC-SVM technique.
The individualized approach generates a separate model for each individual rather than a population-level model.
OC-SVM was selected to handle the variability that may exist among different individuals.
The personalized modeling approach is designed to be applicable to patients who require assistive devices for walking.
This individualized framework distinguishes the methodology from general, non-personalized approaches.
Background
Continuous gait monitoring is proposed as a tool for early detection of functional-state changes in individuals with lower limb mobility impairments.
Lower limb mobility impairments considered include those resulting from neurological diseases, trauma injuries, or aging.
The rationale is that gait is 'a reflection of the physical and mental states of each individual.'
Early detection of functional-state changes enables adjustment of therapies based on the current condition of the patient.
Continuous specialist assessment was identified as unfeasible given existing limited resources, motivating the automated approach.
Results
The methodology was validated in both a controlled experimental setting with healthy participants and a real-world longitudinal clinical setting with post-stroke patients.
Healthy participants (n=9) had different simulated functional states to provide controlled validation conditions.
Post-stroke participants (n=3) were monitored over one year in a real clinical context.
The two-phase validation strategy allowed assessment of the methodology under both idealized and realistic conditions.
Results showed higher accuracy in simulated (70-97%) versus real post-stroke (78% average) conditions.
What This Means
This research suggests that it is possible to automatically detect meaningful changes in a person's physical condition by continuously monitoring how they walk. The study focused on people who need assistive devices (such as canes or walkers) due to conditions like stroke, and developed a computer-based system that learns each person's unique walking pattern and then flags when that pattern changes significantly — potentially indicating a change in their health or rehabilitation status. Because each person walks differently, the system was designed to build a customized profile for each individual rather than comparing everyone to a single standard.
The system was tested in two ways: first with nine healthy volunteers who intentionally altered their gait to simulate different functional states, where the system correctly identified changes with 70-97% accuracy; and second with three stroke survivors monitored over an entire year, where the system achieved an average accuracy of 78% in detecting real changes in their condition. Both tests suggest the approach works reasonably well across different types of users and conditions.
This research matters because rehabilitation specialists currently have limited time and resources to continuously monitor patients between clinic visits, meaning important changes in a patient's condition can go undetected for extended periods. A tool like this could help bridge that gap by providing ongoing, automated monitoring in everyday life, allowing therapists to be alerted when a patient's walking patterns change and to adjust treatments more promptly. While the study was conducted on a small number of participants and further validation is needed, the results indicate that personalized gait monitoring is a promising direction for supporting rehabilitation and maintaining the autonomy of people with mobility impairments.
Otamendi J, Zubizarreta A, Torre I, Sesma C. (2026). Personalized Detection of Functional-State Changes Through Continuous Gait Monitoring: A Methodology for Assistive-Device Users.. Sensors (Basel, Switzerland). https://doi.org/10.3390/s26165234