Automatic classification of upper-limb motor function after stroke

Study overview

Reliable, repeatable assessment is essential in stroke rehabilitation. Clinicians need to understand how motor function is changing over time, but traditional assessments can be time-intensive and difficult to repeat frequently in real-world care settings.

This exploratory study evaluated a portable motion capture system designed to automatically classify upper-limb motor function after stroke. The system used three-dimensional kinematic data from wearable sensors and an automated decision-tree classifier based on features from the Functional Ability Score of the Wolf Motor Function Test.

Five stroke patients were tested on both sides across five selected tasks. The system’s classifications were compared with assessments from a trained clinician who evaluated the movements simultaneously and was blinded to the system output.  

Key findings

Performance time measurement closely matched clinician assessment

The automated system measured performance time with a mean difference of 0.17 seconds compared with clinician assessment.

This suggests the system could measure task timing with a high level of agreement in this small exploratory sample.  

Functional ability scores aligned in most patients

For Functional Ability Score evaluation, the system and clinician agreed in 4 out of 5 patients across the two tasks evaluated.

While the sample was small, this early result suggested that automated classification could support more consistent assessment of upper-limb motor function after stroke.  

The system was designed for portable, lower-cost assessment

The study described the system as portable and lower-cost, with potential use in ambulatory clinical settings and research trials.

That matters because stroke rehabilitation often depends on repeated assessment, and tools that are easier to deploy may help clinicians and researchers measure recovery more consistently over time.  

Why this study matters

This study reflects Sword’s early work in making rehabilitation more measurable, scalable, and clinically grounded.

The core question was not only whether technology could guide therapy, but whether it could help assess movement more objectively. By combining wearable motion sensing with automated classification, the study explored how motor assessment after stroke could become more repeatable and less dependent on manual observation alone.

For clinical and research audiences, the study offers early evidence that sensor-based assessment may support more consistent measurement of upper-limb motor function. For organizations evaluating Sword’s clinical model, it shows an early foundation in motion analysis, rehabilitation science, and technology-enabled care delivery.

The study should be understood as exploratory. It tested a prototype in a small group of patients, and further validation would be needed before drawing broader conclusions about clinical use at scale.

Read the full study