An EEG Tool for Monitoring Patient Engagement during Stroke Rehabilitation: A Feasibility Study

Abstract

Objective: Patient engagement is of major significance in neural rehabilitation. We developed a real-time EEG marker for attention, the Brain Engagement Index (BEI). In this work we investigate the relation between the BEI and temporary functional change during a rehabilitation session.

Methods: First part: 13 unimpaired controls underwent BEI monitoring during motor exercise of varying levels of difficulty. Second part: 18 subacute stroke patients underwent standard motor rehabilitation with and without use of real-time BEI feedback regarding their level of engagement. Single-session temporary functional changes were evaluated based on videos taken before and after training on a given task. Two assessors, blinded to feedback use, assessed the change following single-session treatments.

Results: First part: a relation between difficulty of exercise and BEI was identified. Second part: temporary functional change was associated with BEI level regardless of the use of feedback.

Conclusions: This study provides preliminary evidence that when BEI is higher, the temporary functional change induced by the treatment session is better. Further work is required to expand this preliminary study and to evaluate whether such temporary functional change can be harnessed to improve clinical outcome.

Clinical trial registration: Registered with clinicaltrials.gov, unique identifier: NCT02603718 (retrospectively registered 10/14/2015).

Figures

Figure 1

Algorithm for use of the BEI during treatment sessions. If the BEI level was stable, the current exercise continued. When BEI level dropped consistently below average for at least 30 seconds, the patient was first encouraged to concentrate on the exercise, and if this did not help, the therapist evaluated the exercise level. If it was too easy, the therapist intensified the exercise. If it was too difficult, the therapist reduced the intensity of the exercise. If this did not improve the BEI, the therapist suggested rest or used supportive and passive exercises for a few minutes. ∗Note that both BEI level and functional level are evaluated relatively for each patient.

Figure 2

Demonstration of component template matching. The component template is emphasized in black in the top inset. The new sample in the bottom of the figure is scanned with a moving window, following normalization to the [−1,1] range. Whenever a match is found (in black rectangles), it is counted. The BEI is a normalization of this count to the [0,1] range.

Figure 3

Computation of BEI session. For the sake of clarifying the computation of the BEI session from the basic 10 seconds BEI values, we show an example from one patient of the basic BEI values from the two sessions he underwent. BEI values were computed every 10 seconds for both sessions. The mean BEI (dashed line) and +1 standard deviation (thick line) of all sampled values obtained from both sessions were computed. For each session, the number of samples above the mean + 1 standard deviation was counted and was divided by the total number of samples for this session. This value was used as the BEI session, and it is shown for each of the two sessions for demonstration. The same BEI computation session was followed for all patients.

Figure 4

BEI association with exercise difficulty and practice. (a) Dynamics of BEI as a function of exercise difficulty. At each level, the BEI is averaged for both exercises over all participants (±SD). The arrows mark the tendency of BEI change between 3 exercise levels: an average of levels 1 and 2 (owing to functional similarity), level 3, and level 4. The inset shows the percent of success, reported by ArmTutor. The success rate reduced in the 4th level. (b) Dynamics of BEI between start and end of the exercises. The figure shows the decrease of the index between start-of-exercise and end-of-exercise. The start and end columns present averages (±SD) from all 4 levels of exercise (2 exercises in each level) over all 13 controls.

Figure 5

Comparison of the temporary functional change between the sessions in which the BEI was higher for each patient and the sessions in which the BEI was lower for each patient. As each patient participated in two sessions, one of them had by definition higher BEI than the other and the data in the figure aggregates all sessions with higher BEI and all sessions with lower BEI over patients. For some patients the session with higher BEI was the feedback session, while for other patients the session with higher BEI was the no-feedback session. The y-axis shows the percentage of patients with a session temporary functional change index above thresholds, which are presented in the x-axis. This BEI-based comparison revealed a significant difference in temporary functional change (p < 0.05).

Figure 6

Accumulative histogram comparisons between patients who started with a feedback session and those who started with a no-feedback session. The y-axis shows the percentage of sessions with temporary functional change indices above the thresholds, which are presented in the x-axis. For participants who started with a feedback+ session, both the first feedback+ and the second feedback− session were included in the count. For participants who started with a feedback− session, both the first feedback− and the second feedback+ session were included in the count. Post hoc analysis of the highest possible temporary functional change (≥+2) revealed a significant difference (p < 0.01) between patients who started with feedback in the first session and patients who started with no feedback in the first session. This difference is emphasized with the dashed rectangle.