Robotic Ankle Assessment Post-Stroke: Reliability, Comparison to Therapists, and Benchmark Dataset Development

In rehabilitation research, coarse clinical outcome measures may limit the sensitivity of trials by failing to detect meaningful intervention-induced changes. Sensorized robotic platforms can potentially improve the evaluation of motor function, yet their reliability compared to skilled therapists remains unclear. This study utilized a robotic device to measure two fundamental impairments that are critical to ankle function: range of motion (ROM, active and passive) and dorsiflexion maximum voluntary contraction (MVC). In 34 chronic hemiparetic post-stroke individuals, we assessed test-retest reliability over two days for the robot and experienced therapists, who used a goniometer and manual muscle testing (MMT). We also evaluated robotic test-retest reliability in 36 young and 26 older unimpaired adults. Reliability for robotic and therapist-based AROM and MVC measures was high (ICC > 0.86), consistent across all groups. Robotic AROM and PROM measure-ments correlated strongly with therapist assessments (r > 0.60, p < 0.001) but were 120% and 37% larger than therapist assessments (p < 0.001), respectively. For the MVC measurement, the therapist assigned 85% of participants a score of 1 on the MMT, but their MVC torque was distributed from 0 to ~20 Nm. Measurement differences between methods likely arose from the robot's constrained setup, allowing compensatory muscle activation. The increased granularity provided by robotic MVC measurements could enable more precise tracking of motor recovery and facilitate tailored rehabilitation strategies. These results support the clinical utility of robotic platforms for ankle assessment, offering detailed, objective measurements that can augment traditional evaluations. A supplementary dataset for benchmarking future research is provided.