Decoding hand and wrist movement intention from chronic stroke survivors with hemiparesis using a user-friendly, wearable EMG-based neural interface

Abstract

Objective: Seventy-five percent of stroke survivors, caregivers, and health care professionals (HCP) believe current therapy practices are insufficient, specifically calling out the upper extremity as an area where innovation is needed to develop highly usable prosthetics/orthotics for the stroke population. A promising method for controlling upper extremity technologies is to infer movement intention non-invasively from surface electromyography (EMG). However, existing technologies are often limited to research settings and struggle to meet user needs.

Approach: To address these limitations, we have developed the NeuroLife^® EMG System, an investigational device which consists of a wearable forearm sleeve with 150 embedded electrodes and associated hardware and software to record and decode surface EMG. Here, we demonstrate accurate decoding of 12 functional hand, wrist, and forearm movements in chronic stroke survivors, including multiple types of grasps from participants with varying levels of impairment. We also collected usability data to assess how the system meets user needs to inform future design considerations.

Main results: Our decoding algorithm trained on historical- and within-session data produced an overall accuracy of 77.1 ± 5.6% across 12 movements and rest in stroke participants. For individuals with severe hand impairment, we demonstrate the ability to decode a subset of two fundamental movements and rest at 85.4 ± 6.4% accuracy. In online scenarios, two stroke survivors achieved 91.34 ± 1.53% across three movements and rest, highlighting the potential as a control mechanism for assistive technologies. Feedback from stroke survivors who tested the system indicates that the sleeve's design meets various user needs, including being comfortable, portable, and lightweight. The sleeve is in a form factor such that it can be used at home without an expert technician and can be worn for multiple hours without discomfort.

Significance: The NeuroLife EMG System represents a platform technology to record and decode high-resolution EMG for the real-time control of assistive devices in a form factor designed to meet user needs. The NeuroLife EMG System is currently limited by U.S. federal law to investigational use.

Fig. 1

Illustration of experimental data collection procedure. Subjects were seated in front of a computer monitor with the sleeve on their impaired arm, and their arms placed on the table. The sleeve was connected to a custom-built EMG signal acquisition module, which then connected to a laptop computer. Images of hand postures were shown on the monitor and the subject followed along to the best of their ability. Each recording block was approximately 2–3 min in length and involved hand posture cues interleaved with rest periods. The recording block began with an 8-s lead in rest period. Each cue and rest period presentation time were randomly selected between 4 and 6 s for subjects with stroke. An operator ran the data collection software and observed EMG signals during data collection to ensure proper recording of data

Fig. 2

Representative EMG data recorded from subject with stroke. A Filtered EMG data recorded from 3 separate channels on the NeuroLife Sleeve during 3 movements: Hand Open (HO), Forearm Supination (FS), and Hand Close (HC). B Heatmap of normalized RMS activity, with the channel number on the y-axis and time on the x-axis. Note the activity across clusters of electrodes for each of the 3 separate movements. C Normalized RMS activity mapped to the sleeve orientation, with a legend showing the orientation of the sleeve mapping (flex. = flexors, ext. = extensors). Note the location of EMG activity is spatially located near the related musculature for each of the 3 movements

Fig. 3

Decoding hand and wrist movements using the NeuroLife EMG System. A Illustration depicting the data used for training and testing the decoder. The presentation of the cue is shown as a black bar on the top of the plot, and the middle 2.5 s of the cue presentation is used for analysis. B Heatmaps of various movements from a subject with stroke. C Decoding performance comparing 3 models: LR (Logistic Regression), SVM (Support Vector Machine), and NN (Neural Network). The NN outperforms both the LR and SVM models (paired t-test NN vs. SVM, p = 9.3 × 10^–3; NN vs. LR, p = 9.1 × 10^–4). D Association between the observed movement score and decoder performance of the neural network (One-way ANOVA, Accuracy (%): F[3, 80] = 13.38, p = 3.7 × 10^–7). The decoder struggles learning to predict movement attempts in which there was no observable movement (movement score = 0), and performs similarly when there is observable movement (movement score ≥ 1). E Confusion matrix for a subject with stroke detailing the decoding performance across all movements

Fig. 4

Decoding hand and wrist movements in subjects with severe hand impairment (UEFM-HS < 3). A Left: Comparison of severe (UEFM-HS < 3) and mild (UEFM-HS ≥ 3) subject impairment average movement scores (Average movement score: unpaired t-test UEFM-HS < 3 vs. UEFM-HS ≥ 3, p = 0.02). Right: Comparison of NN decoding performance for severe and mild subject impairments (Decoding accuracy: unpaired t-test UEFM-HS < 3 vs. UEFM-HS ≥ 3, p = 0.006). B Decoding performance of NN binary classifier for UEFM-HS < 3 subjects comparing Rest and Move in which Move is made up of combining all 12 movements into a single class. Confusion matrix of subject 61,204 for the two-class problem. The observed movement score is the average of all movements observed movement scores. The two-class decoder can reliably distinguish the difference between a resting and moving state. C Decoding performance of NN model when restricting classes to Rest, Hand Close, and Hand Open. Confusion matrix of lowest performing subject (61,204) for the three-class problem. The three-class decoder is not sufficient to distinguish the movements reliably. D Decoding performance of NN model when restricting classes to Rest and the top 2 movements for each subject for a total of three classes. Confusion matrix of subject 61,204 for the three-class problem. Focusing on movements specific to subjects increases the robustness of decoder performance

Fig. 5

Decoding hand and wrist movements in a continuous EMG dataset. A Dynamic cue shifting significantly improved accuracy compared to no cue shift (Cue shift: paired t-test Dynamic vs. None, p = 0.020). There was no significant difference between a dynamic cue shift and static 800-ms cue shift (approximately the average cue shift across subjects) (Cue shift: paired t-test Dynamic vs. Static 800 ms, p = 0.22). B Confusion matrix detailing performance from one subject in the continuous dataset. C Time series plot depicting decoder class probabilities across time. The presented cue is shown in above the time series plot as a rectangular colored bar with the color corresponding to the movement class

Fig. 6

Online decoding of hand and wrist movements. A Confusion matrix of subject 30,458 from the online decoding session. B Online decoding performance for both subjects on the real-time demonstration dataset. C Time series plot depicting decoder class probabilities across time for subject 30,458. The presented cue is shown above the time series plot as a rectangular colored bar with the color corresponding to the movement class

Fig. 7

Summary of the NeuroLife Sleeve usability data from subjects with stroke. Each subject with stroke ranked the NeuroLife Sleeve based on 6 usability domains. Group data is presented for each of the 6 domains