Brain–Computer Interface Using Functional Near-Infrared Spectroscopy for Post-Stroke Motor Rehabilitation: Case Series

Full Text

Introduction

Brain–computer interfaces (BCIs) are systems that translate brain activity signals into commands for the output devices. Non-invasive BCIs enable feedback motor imagery (MI) training in neurological patients to support their motor rehabilitation. At present, several systematic reviews comprise the evidence base for employing BCIs, which utilize sensory and motor EEG rhythms (EEG BCIs), in post-stroke rehabilitation [1–4]. We also conducted a randomized controlled trial (RCT) to evaluate effectiveness of using EEG BCI with visual and exoskeleton-mediated kynesthetic feedback in the post-stroke population [5, 6]. A series of training sessions for MI, controlled with this technology, resulted in improved grasping and pinching. However, the EEG-BCI use is complicated due to the necessity of applying EEG gel onto the scalp, signal artifacts in patient motion and muscle contraction during training, and the low spatial resolution of signal source detection.

Functional near-infrared spectroscopy BCIs (fNIRS-BCIs) is a non-invasive BCI to conduct feedback MI training. The sources of brain activity may involve alterations in concentrations of oxy-, deoxy-, and total hemoglobin up to 4 cm below the head surface. This technology does not require the use of any electrode gel, and patient movements during training do not result in significant signal distortion. At present, fNIRS-BCI remains largely unexplored approach to post-stroke rehabilitation. The latest review [7] of biofeedback methods mentions only one fNIRS-BCI study in post-stroke rehabilitation [8]. The study protocol implied the presentation of the signal amplitude as the color and the height of the bars on the monitor rather than signal classification.

The exploratory study of this technology with assessment of the attained level of online signal detection was necessary as an initial test prior to the development of the fNIRS BCI effectiveness RCT protocol.

The study objective is to evaluate the potential fNIRS BCI use for customized hand MI training in comprehensive post-stroke rehabilitation.

Materials and methods

Study design

This pilot study is a case series of post-stroke rehabilitation integrating standard management with additional fNIRS BCI mental training.

The study was conducted at the Institute of Neuro-Rehabilitation and Recovery Technologies of the Research Center of Neurology from June to October 2022. The study included inpatients receiving scheduled rehabilitation. Each patient participated for a total of 12 days including five training days followed by two days off, and then five more training days. fNIRS BCI training supplemented the standard clinical rehabilitation program. Prior to the first fNIRS BCI training session and post the last one, hand motion was assessed and scored using international validated scales.

The study protocol was approved by the Local Ethics Committee of the Research Center of Neurology (Protocol No. 5-4/22, 1 June 2022). Patient participation in this study was entirely voluntary, and all participants provided informed consent. The study protocol was pre-registered in the local research project protocol database of the Institute of Neuro-Rehabilitation and Recovery Technologies (ID 210).

Inclusion criteria

Inclusion criteria: primary or recurrent stroke with a supratentorial CT or MRI-confirmed focus, mildto-moderate clinical paresis of the distal upper limb, the stroke time of one day to 12 months, clinically stable state without any life-threatening conditions, and informed consent to participate in the study.

Exclusion criteria: severe speech, vision, and/or cognitive function impairment and/or hand tissue contractures.

Standard rehabilitation course

All patients underwent a two-week rehabilitation course that included personal instructor-led exercise therapy, neuro-muscular electric stimulation of lower limb muscles, therapeutic massage, robotic biofeedback mechanotherapy to recover hand fine movements, and stationary cycling exercises. The above activities were performed on daily basis, except for the days-off (10 sessions each).

fNIRS BCI training

MI fNIRS BCI training was performed on a daily basis except for the days-off in addition to the standard rehabilitation course. Each patient underwent 10 training sessions.

The study used a non-invasive BCI based on the recognition of a BOLD cortical signal in the hand MI expressed as changes in the relative concentrations of oxy- and deoxyhemoglobin assessed by near-infrared spectroscopy. The NIRScout system (NIRxMe- dicalTechnologies), with 8 detectors and 16 sources, was used for fNIRS. The study protocol, the filtration method, and the classification of brain activity signals and the software used were described previously [9]. The patient received visual feedback on the 22'' computer monitor. The flow chart is presented in Fig. 1.

 

Fig. 1. The fNIRS BCI and post-priming training flow chart

Hb, deoxyhemoglobin; HbO, oxyhemoglobin.

 

The investigator customized movement types for MI during BCI training and selected the most difficult (for which, accordingly, the lowest score on this test was assigned) movement based on the patient's ARAT results. Before each training session, the investigator asked the patient to repeat the target movement several times until they confirmed their readiness to perform it mentally (priming). If the movement implied manipulating an item (ARAT), the item was provided to the patient during priming.

During the procedure, a BCI cap equipped with sources and detectors was put on the patient's head. During the session, the patients sat by the computer monitor at the desk with their hands comfortably positioned on either the armrests or the desk. There was a fixation point consisting of a circular shape in the center of the dark screen to stabilize the patient's gaze. Additionally, three arrows surrounded the fixation point, which provided commands with colors that changed dynamically. Following one of three commands, the patient either kynesthetically imagined slow motion of their left or right hand (with the corresponding arrow changing color) or relaxed and directed their gaze towards the center of the screen (with the upper arrow changing color). When MI task was successfully detected by the classifier, a gaze fixation point in the center of the screen was gradiently coloring green. Following the command to relax, the point color did not change in any classifier response.

Each nine-minute training session consisted of four units, with two commands randomly given to the left and right hand for 15 seconds, interspersed with rest intervals. Two-second preparatory instruction preceded the main commands.

Signal acquisition and processing

Sources and detectors were nested on the EEG cap. Signal recording frequency was 15.625 Hz. Recorded radiation intensities were calculated as oxyhemoglobin (HbO) and deoxyhemoglobin (Hb) using the modified Beer–Lambert law. To classify active states, the signal was filtered with a first order Chebyshev type I filter with 1 dB passband ripple and 0.005 Hz cutoff frequency. To classify rest vs active state, the signal was filtered with a second order Chebyshev type I filter with 1 dB passband ripple. The cutoff frequencies were selected for zero drift at the command frequency. For classification, we used linear discriminant analysis with additional learning based on the previous units of the current session and the participant's previous test sessions. In gradual classification, we first classified resting vs. active state and left hand vs. right hand in the following classification of active states. Per-second record intervals were classified.

Endpoints

First of all, this pilot study evaluated the quality of the detection of patient's mental states during training by the classifier. This indicator is measured as the percentage of correctly detected intervals. Over 33% of the total amount of the classified intervals is considered to be higher than random because the patients performed three mental tasks (100% / 3 ≈ 33%).

For the pre- and post-rehabilitation assessment of the hand function, we used the ARAT scale (with a maximum possible score of 57 points and a clinically significant increase in 6 points in the chronic stroke period and 12–17 points in the acute stroke period) [10, 11] and the Fugl-Meyer Assessment of the Upper Extre- mity (with a maximum possible score of 126 points and a clinically significant increase in 5 points) [10, 12]. The trained physician performed the blinded movement assessment.

Statistical data processing

The resulting percentage of the detection of brain signal activity is presented as the median and the 25^th and 75^th quartiles. The MatLab R2019b package (MathWorks, Natick, MA, USA) was used for analysis.

Because the study was a non-comparative exploratory case series, we did not use other statistical methods.

Results

Population

While eight patients were successfully screened, five were included in the study and three were not included. Table 1 shows the profiles of the participants. All five participants completed the study with no withdrawals.

 

Table 1. Patient demographics and baseline characteristics

Patient	Sex	Age, years	Stroke type	Stroke lesion side	Stroke time, months ago	APAT score	FMA score (upper extremity)
P1	Male	71	Primary	Left	12	44	107
P2	Male	58	Primary	Left	12	39	104
P3	Male	58	Secondary	Right	8 and 2	35	114
P4	Male	49	Primary	Right	< 1	35	116
P5	Female	43	Primary	Left	1	52	122

 

In order to train MI, we selected the following movements: 'to pinch and hold a 6-mm ball with their digit 1 and digit 4 for P1, P3, and P4; 'to pinch and hold a 6-mm ball with their digit 1 and digit 2 for P2; 'to pinch and hold a 1.5-mm ball with their digit 1 and 4 for P5.

Motor control and improvement

All patients were able to control MI fNIRS-BCI with better than random classification accuracy rate (>33%; Table 2).

 

Table 2. Resulting BCI control and motor score improvement

Patient	Classifier accuracy, %	ARAT score improvement	FMA score improvement
P1	68 [57; 73]	6*	1
P2	41 [37; 47]	4	3
P3	45 [41; 47]	6*	0
P4	45 [34; 50]	20*	9*
P5	49 [46; 59]	5	3

Note. *Clinically significant improvement as adjusted by the stroke time [10].

 

Following comprehensive inpatient rehabilitation and additional MI fNIRS-BCI training, all patients showed improved motor scores (Fig. 2). While three patients demonstrated the clinically significant improvement in ARAT score, one of them also achieved significant improvement in FMA score (Fig. 2).

 

Fig. 2. Changes in motor scores during rehabilitation and additional fNIRS BCI training.

 

The MI fNIRS BCI training was not associated with any adverse events. However, all participants reported experiencing drowsiness during training, which impaired concentration and task performance by the end of the session.

Discussion

In this study, we demonstrated the potential fNIRS BCI use for customized hand MI training in comprehensive post-stroke rehabilitation. All study participants achieved BCI control rates higher than random.

In the study, the total MI training exposure was 1.5 h (10 nine-minute sessions), which is significantly lower than the total training exposure of 5–27 h, with the session duration of 0.5–1.5 h, in other non-invasive BCI studies [13]. Nevertheless, patients developed drowsiness even during the shorter sessions. Thus, we find it reasonable to change the feedback scenario for play or various visualizations and to break the session units with pauses. Notably, MI drowsiness was mentioned in other studies [13].

Following the fNIRS BCI, all study participants showed improvement in motor function on at least one scale, including clinically significant improvement in three patients. However, the study design did not include a control group and the range of inclusion criteria was quite broad, which limits this exploratory study. Therefore, it is impossible to prove that the resulting improvement was caused by fNIRS BCI training. Nevertheless, the effectiveness of MI fNIRS BCI training has been demonstrated in a number of meta-analyses and systematic reviews based on multiple RCTs [1–4, 13]. Nowadays, professionals are searching for the most convenient and practical BCI systems and feedback scenarios [14].

Although the fNIRS BCI is a more convenient method than EEG BCI, it is the EEG BCI that has been evaluated in the vast majority of studies the present systematic reviews are based on. М. Mihara et al. showed fNIRS BCI effectiveness for rehabilitation of patients with subcortical stroke in the only randomized trial involving 20 patients: six 20-minute training sessions contributed to a better improvement of motor function measured as FMA score in the active group than in the sham fNIRS BCI group [8]. Interestingly, unlike the system we used in our study, the fNIRS BCI technology used by M. Mihara et al. did not involve any online signal classification. Therefore, additional, unobvious actions are needed to translate the recorded signals into commands to start exoskeleton operation or electrostimulation, if these methods are used for additional sensorimotor feedback.

The hallmark of our study is customisation of the movement type for further in-training MI based on the ARAT. This approach corresponds to the modern concept of customized rehabilitation and allows the use of goal attainment scaling [10].

A recent meta-analysis showed a better effect of BCI systems with functional electrostimulation in rehabilitation setting, as compared to those with the exoskeleton for kinesthetic feedback or with visual feedback only [13]. Therefore, we recommend to supplement the fNIRS BCI hardware with functional electromyosti- mulation controlled by brain activity signals in the MI during further rework.

Conclusion

In this pilot study, we demonstrated the potential fNIRS BCI use for customized hand MI training in comprehensive post-stroke rehabilitation and identified the ways to improve this technology and the training protocol.

Ethics approval. The study was conducted non-invasively in accordance with the ethics of the Declaration of Helsinki. The study protocol was approved by the Local Ethics Committee of the Research Center of Neurology (Protocol No. 5-4/22, 1 June 2022). All the participants signed informed consent.

Source of funding. The study was conducted by R.Kh. Lyukmanov, E.S. Ikonnikova, A.N. Cherkasova, and N.A. Suponeva on state assignment by the Ministry of Science and Higher Education of the Russian Federation for the Research Center of Neurology for 2021–2023 (research project No. 210). The study was conducted by P.D. Bobrov, M.R. Isaev, and O.A. Mokienko on state assignment by the Ministry of Science and Higher Education of the Russian Federation for the Institute of Higher Nervous Activity and Neurophysiology of RAS for 2021–2023.

Conflict of interest. The authors declare no apparent or potential conflicts of interest related to the publication of this article.

About the authors

Roman Kh. Lyukmanov

Research Center of Neurology

Email: xarisovich@gmail.com
ORCID iD: 0000-0002-8671-5861

Cand. Sci. (Med.), Researcher, Head, Brain–Computer Interface Group, Institute of Neurorehabilitation

Russian Federation, Moscow

Mikhail R. Isaev

Institute of Higher Nervous Activity and Neurophysiology of the Russian Academy of Sciences

Email: shycmympuk@yandex.ru
ORCID iD: 0000-0002-3907-5056

Junior Researcher, Laboratory of Mathematical Neurobiology of Learning Department

Russian Federation, Moscow

Olesya A. Mokienko

Research Center of Neurology; Institute of Higher Nervous Activity and Neurophysiology of the Russian Academy of Sciences

Email: lesya.md@yandex.ru
ORCID iD: 0000-0002-7826-5135

Cand. Sci. (Med.), Researcher, Brain–Computer Interface Group, Institute of Neurorehabilitation, Research Center of Neurology; Senior Researcher, Institute of Higher Nervous Activity and Neurophysiology of the Russian Academy of Sciences

Russian Federation, Moscow; Moscow

Pavel D. Bobrov

Institute of Higher Nervous Activity and Neurophysiology of the Russian Academy of Sciences

Author for correspondence.
Email: bobrov_pd@mail.ru
ORCID iD: 0000-0003-2566-1043

Cand. Sci. (Biol.), Head, Laboratory of Mathematical Neurobiology of Learning Department

Russian Federation, Moscow

Ekaterina S. Ikonnikova

Research Center of Neurology

Email: xarisovich@gmail.com
ORCID iD: 0000-0001-6836-4386

Junior Researcher, Brain–Computer Interface Group, Institute of Neurorehabilitation

Russian Federation, Moscow

Anastasiia N. Cherkasova

Research Center of Neurology

Email: lesya.md@yandex.ru
ORCID iD: 0000-0002-7019-474X

Junior Researcher, Brain–Computer Interface Group, Institute of Neurorehabilitation

Russian Federation, Moscow

Natalia A. Suponeva

Research Center of Neurology

Email: xarisovich@gmail.com
ORCID iD: 0000-0003-3956-6362

D. Sci. (Med.), Corresponding Member of the Russian Academy of Sciences, Director, Institute of Neurorehabilitation

Russian Federation, Moscow

References

Supplementary files