Motor improvement estimation and task adaptation for personalized robot-aided therapy: a feasibility study

Abstract

Background: In the past years, robotic systems have become increasingly popular in upper limb rehabilitation. Nevertheless, clinical studies have so far not been able to confirm superior efficacy of robotic therapy over conventional methods. The personalization of robot-aided therapy according to the patients' individual motor deficits has been suggested as a pivotal step to improve the clinical outcome of such approaches.

Methods: Here, we present a model-based approach to personalize robot-aided rehabilitation therapy within training sessions. The proposed method combines the information from different motor performance measures recorded from the robot to continuously estimate patients' motor improvement for a series of point-to-point reaching movements in different directions. Additionally, it comprises a personalization routine to automatically adapt the rehabilitation training. We engineered our approach using an upper-limb exoskeleton. The implementation was tested with 17 healthy subjects, who underwent a motor-adaptation paradigm, and two subacute stroke patients, exhibiting different degrees of motor impairment, who participated in a pilot test undergoing rehabilitative motor training.

Results: The results of the exploratory study with healthy subjects showed that the participants divided into fast and slow adapters. The model was able to correctly estimate distinct motor improvement progressions between the two groups of participants while proposing individual training protocols. For the two pilot patients, an analysis of the selected motor performance measures showed that both patients were able to retain the improvements gained during training when reaching movements were reintroduced at a later stage. These results suggest that the automated training adaptation was appropriately timed and specifically tailored to the abilities of each individual.

Conclusions: The results of our exploratory study demonstrated the feasibility of the proposed model-based approach for the personalization of robot-aided rehabilitation therapy. The pilot test with two subacute stroke patients further supported our approach, while providing encouraging results for the applicability in clinical settings. Trial registration This study is registered in ClinicalTrials.gov (NCT02770300, registered 30 March 2016, https://ichgcp.net/clinical-trials-registry/NCT02770300).

Keywords: Personalized therapy; Rehabilitation robotics; Stroke rehabilitation.

Conflict of interest statement

The authors declare that there is no conflict of interest.

Figures

Fig. 1

Experimental setup and protocols. a Schematic overview of experimental setup. b Design of the three-dimensional point-to-point reaching task. Eighteen targets (representing the different subtasks) are positioned over a sphere of 19 cm of radius (equally distributed on the three planes). The empty circle represents the center of the workspace (starting position). c Experimental protocol for healthy participants. Experiments were completed in a single session and were divided into blocks (one initial assessment block AI,1–3, five inversion blocks B1–5, one final assessment block AF,1–3). The assessment blocks consisted of three runs, each composed of 18 reaching movements (one towards each target). The inversion blocks consisted of five runs, each composed of eight reaching movements. The training targets for the inversion blocks were automatically selected by the implemented personalization routine. Breaks were allowed between the blocks to prevent fatigue. d Experimental protocol for the patient. During the initial (AI,1–2) and final (AF,1–2) assessment sessions, all 18 targets were presented to the patient. For each treatment session eight training targets were selected by the implemented personalization routine. The total number of repetitions performed in each session was determined by the physical therapist. e Schematic overview of the different steps performed for the adaptive scheduling of the reaching task with vision inversion for healthy participants and the reaching task without vision inversion for patients

Fig. 2

Analysis of performance measures for the experiment with healthy participants. Average values of mean velocity (MV, panel a), spectral arc length (SAL, panel b) and rate of success (%SUCC, panel c) for each run (eight reaching movements) of fast (red) and slow (grey) adapters. Measures were averaged for all targets presented during a run and for all subjects of a group. Shaded areas depict standard error of the mean (sem). Vertical bars (panel d) depict the percentage of subjects in each group for which a target was replaced in B3–5 or was not replaced at all. No targets were replaced in and B1–2 due to lack of data needed for proper estimation of motor improvement

Fig. 3

Examples of MI estimates and performance measures at subtask level. Data are presented for a fast adapter and a slow adapter for the same two targets. Repetitions for each target are concatenated for all inversion blocks and presented in chronological order. Data for mean velocity (MV), spectral arc length (SAL) and MI were low pass filtered for visualization purposes (raw data shown in light red/grey). Dotted lines depict one of the necessary conditions (MI > 0) for triggering a target replacement. Green areas indicate the time span where the model detected a performance plateau and triggered a target replacement. Estimated model parameters (αj, βj) for each target and subject are presented next to the corresponding MI curves (a summary and analysis on the model parameters can be found in Additional file 1)

Fig. 4

Summary of the results from the pilot test with two subacute stroke patients. a The first three rows show the mean values for mean velocity (MV), spectral arc length (SAL) and rate of success (%SUCC) for each assessment and treatment session of both patients. Measures were averaged for all targets presented during a session, shaded areas depict standard error of the mean (sem). The fourth row shows number of movements performed by the patients in each session. The fifth row shows the scores on the Fugl-Meyer scale for upper extremities (FMA-UE) for initial (AI,1–2) and final (AF,1–2) assessment sessions. The dotted line indicates the maximum achievable score for FMA-UE (66 points). b Summary of the training targets presented to the patients in each treatment session. Targets are listed by the order as presented to the patients (first eight targets from the top are the initial training set). c Analysis of performance measures for two different time points (before replacement and after reinsertion). Values are compared between the last four movements towards a training target before its replacement and the first four movements towards the target after it has been reinserted for training. The data show the mean improvement for MV, SAL and %SUCC averaged for all targets at both time points. Improvements were calculated with respect to the mean values obtained from the first four movements towards each target in AI,2. Error bars depict standard error of the mean (sem). P-values of Wilcoxon signed-rank tests are reported above the bars