JMIR Publications

ABSTRACT

Background:

Stroke is a leading cause of long-term upper-limb disability, yet conventional therapy delivers less repetitive practice than the motor-learning literature implies, and the repetitions actually performed are rarely measured. Home-based markerless systems are an accessible alternative, but many require specialized hardware or report movement metrics whose provenance is opaque or, in the worst case, fabricated.

Objective:

This study aimed to design and technically verify Apexa, a markerless, single-webcam, browser-based upper-limb rehabilitation application that computes rehabilitation-relevant movement metrics transparently and is engineered to run on low-specification hardware.

Methods:

Apexa provides three gamified exercise modules mapped to activities of daily living, computing a provenance-tagged range-of-motion estimate, a spectral arc length (SPARC) smoothness metric, and a trunk-compensation flag behind a security-hardened clinician data model. Because no patient data were collected and no clinical claim was made, verification did not use a patient trial. It combined a formative expert content appraisal of the exercise designs by 5 rehabilitation professionals, a source-code audit, synthetic-signal testing of the metric algorithms against inputs of known properties, and a real-time performance measurement on a physically low-specification entry-level device. No end user operated the built system.

Results:

The source audit confirmed that no stored clinical value is fabricated and that uncaptured measurements are stored as null rather than zero, and that the recorded range of motion is provenance-tagged and double-gated against trunk compensation. Synthetic testing showed the SPARC implementation reproduced standard behavior, returning -1.39 on a canonical minimum-jerk reach at the boundary of the reported -1.4 to -1.6 band and behaving correctly in direction and in its designed invariance to duration and amplitude; however, its interpretive anchors did not transfer to continuous movement, where a smooth stir scored about -2.40, outside the discrete-reach band. The formative expert appraisal did not reach conventional content-validity thresholds and is reported as formative. On the entry-level device the system ran all three modules with bounded, non-leaking memory and zero dropped frames, but at a reduced 8 to 12 frames per second, below the 30 frames-per-second target.

Conclusions:

Apexa is a technically verified feasibility artifact, not a clinically validated instrument. Its principal contribution is methodological: a demonstration that a low-cost markerless rehabilitation tool can be built so that the provenance and limits of every figure it reports are explicit. SPARC is delivered as a within-patient relative measure only, and usability and clinical validation remain future work along the staged pathway set out here.