Pattern Recognition Prosthetic Control (Adaptation)

Efficacy of Control System Adaptation in Improving Upper-Extremity Prosthetic Limb Wear Time in a Real-World Setting, a Randomized Crossover Trial

Many different factors can degrade the performance of an upper limb prosthesis users control with electromyographic (EMG)-based pattern recognition control. Conventional control systems require frequent recalibration in order to achieve consistent performance which can lead to prosthetic users choosing to wear their device less. This study investigates a new adaptive pattern recognition control algorithm that retrains, rather than overwrite, the existing control system each instance users recalibrate. The study hypothesis is that such adaptive control system will lead to more satisfactory prosthesis control thus reducing the need for recalibration and increasing how often users wear their device. Participants will wear their prosthesis as they would normally at-home using each control system (adaptive and non-adaptive) for an 8-week period with an intermittent 1-week washout period (17 weeks total). Prosthetic usage will be monitored during each period in order to compare user wear time and recalibration frequency when using adaptive or non-adaptive control. Participants will also play a set of virtual games on a computer at the start (0-months), mid-point (1-months) and end (2-months) of each period that will test their ability to control prosthesis movement using each control system. Changes in user performance will be evaluated during each period and compared between the two control systems. This study will not only evaluate the effectiveness of adaptive pattern recognition control, but it will be done at-home under typical and realistic prosthetic use conditions.

The study is a randomized crossover home trial consisting of two 8-week periods with an intermittent 1-week washout period (17 weeks total). Participants will use either adaptive control or non-adaptive control during the first 8-week period then switch to using the opposite control style during the second 8-week period.

Participants will not be explicitly informed which type of control they will be using during each 8-week period.

The adaptive control system updates the pattern recognition control algorithm by incorporating new EMG data each instance the prosthetic user recalibrates their device.

The conventional, non-adaptive control systems resets the pattern recognition control algorithm by deleting old EMG data each instance the prosthetic user recalibrate their device.

Using an electromyographic (EMG)-based pattern recognition controller to move an upper limb prosthetic device in a home trial.

We will record each instance participants turn on or off their pattern recognition device throughout the home trial. Prosthetic wear time is defined as the cumulative amount of time participants keep their pattern recognition device turned on during the course of each in-home 8-week period. We will perform a statistical analysis to compare wear time when using each type of pattern recognition control system (adaptive and non-adaptive). We will complete repeated measures analysis of variance with subject as a random factor, order of control system used as a fixed variable, and wear time as a fixed variable.

We will record each instance participants recalibrate their pattern recognition device throughout the home trial. We will perform a statistical analysis to compare the frequency of calibrations when using each control system (adaptive and non-adaptive). We will complete a repeated measures analysis of variance with subject as a random factor, order of control system used as a fixed variable, and wear time as a fixed variable.

Participants will complete two virtual games called Simon Says and In-the-Zone using the Coapt Complete ControlRoom desktop application. Both games will test how well participants control motion of virtual objects using their pattern recognition device. We will measure their overall control performance by computing completion rate, movement time, path efficiency. We will perform a statistical analysis to compare virtual game performance when using each control system. We will complete a repeated measures analysis of variance with subject as a random factor, order of pattern recognition control system used as a fixed variable, and each performance metric as a fixed variable.

Participants will complete the virtual games at the start (0-months), mid-point (1-months) and end (2-months) of each in-home 8-week period.

Participants will complete the Upper Extremity Functional Status module from RIC's Orthotics Prosthetics User Survey (OPUS). The OPUS asks prosthetic users to rate the level of difficulty (from very easy to very difficult) in performing upper arm/hand functions using their pattern recognition device. Survey data will be evaluated using rating scale analysis (Rasch model).

Participants will complete the OPUS at the start (0-months) and end (2-months) of each 8-week period. of each in-home 8-week period.

Participants will complete a survey or phone interview to provide feedback on which control system they prefer between adaptive or non-adaptive. Participants will inform whether they prefer the control system they used in the first or second 8-week period.

Participants will be instructed to use their pattern recognition device to make a set of independent prosthesis motions and hold each motion for 3 seconds. For each motion, we will record the output motion class determined by their pattern recognition classifier every 50 ms. We will measure the performance of their classier when using each control system (adaptive and non-adaptive) by computing the classification accuracy which is defined as the number of correct classifications over the total number of classifications for each motion. We will perform a statistical analysis to compare classification accuracy when using each control system. We will complete a repeated measures analysis of variance with subject as a random factor, order of pattern recognition control system used as a fixed variable, and classification accuracy as a fixed variable.

We will record classification accuracy at the start (0-months), mid-point (1-months) and end (2-months) of each in-home 8-week period.

Inclusion Criteria: Subjects have an upper-limb difference (congenital or acquired) at the transradial (between the wrist and elbow), elbow disarticulation (at the elbow), transhumeral (between the elbow and shoulder), or shoulder disarticulation (at the shoulder) level. Subjects are suitable to be, or already are, a Coapt pattern recognition user (Coapt Complete Control Gen 2). Subjects are between the ages of 18 and 70. Exclusion Criteria: Subjects with significant cognitive deficits or visual impairment that would preclude them from giving informed consent or following instructions during the experiments, or the ability to obtain relevant user feedback discussion. Subjects who are non-English speaking. Subjects who are pregnant.

Only de-identified individual participant data collected during the study may be shared. This includes any experimental data that will underlie results in a publication such as EMG data, prosthesis usage data, virtual game data and surveys and questionnaires.

We expect study data and results to become available at the end of the study upon completing data analysis and publication.

It is at the discretion of authorized study personnel with whom data will be shared or where it may be made available. Only de-identified data will be shared using standard data file formats (.csv or .txt). Data may be shared with the research community at large to advance science and health. Data will be publicly available via an online data sharing website only if required for publication in a scientific journal. Upon data analysis completion, study results may be shared with subjects upon request and will be disseminated to the public in the form of a journal publication. Study results may also be posted on the Coapt website.

Chicoine CL, Simon AM, Hargrove LJ. Prosthesis-guided training of pattern recognition-controlled myoelectric prosthesis. Annu Int Conf IEEE Eng Med Biol Soc. 2012;2012:1876-9. doi: 10.1109/EMBC.2012.6346318.

Scheme E, Englehart K. Electromyogram pattern recognition for control of powered upper-limb prostheses: state of the art and challenges for clinical use. J Rehabil Res Dev. 2011;48(6):643-59. doi: 10.1682/jrrd.2010.09.0177.

Simon AM, Hargrove LJ, Lock BA, Kuiken TA. Target Achievement Control Test: evaluating real-time myoelectric pattern-recognition control of multifunctional upper-limb prostheses. J Rehabil Res Dev. 2011;48(6):619-27. doi: 10.1682/jrrd.2010.08.0149.

Kyranou I, Vijayakumar S, Erden MS. Causes of Performance Degradation in Non-invasive Electromyographic Pattern Recognition in Upper Limb Prostheses. Front Neurorobot. 2018 Sep 21;12:58. doi: 10.3389/fnbot.2018.00058. eCollection 2018.