Precise estimation of human corticospinal excitability associated with the levels of motor imagery-related EEG desynchronization extracted by a locked-in amplifier algorithm

Physical motor exercise aided by an electroencephalogram (EEG)-based brain-computer interface (BCI) facilitates functional recovery in patients with motor deficits due to stroke [1‐12]. In such a BCI paradigm, event-related desynchronization (ERD) in the alpha (8–13 Hz) and beta (15–30 Hz) bands is extracted from EEG signals recorded over the primary sensorimotor area (SM1), and visual and sensory feedback contingent to the extent of ERD is provided via a motor-driven orthosis or neuromuscular electrical stimulation. The repetition of BCI-aided motor exercise in finger opening or arm reaching paradigms is suggested to help patients learn to activate sensorimotor cortical neurons more efficiently than exercises without a BCI [9, 13].

Previous studies using transcranial magnetic stimulation (TMS) delivered to the primary motor cortex have shown that an increase of corticospinal excitability is associated with ERD [14] and is accompanied with a decrease of GABAergic intracortical inhibition [15], indicating ERD as a biomarker of corticospinal excitability. However, the real-time estimation of ERD trace results has a time delay measured in hundreds of milliseconds. A variety of spectral analyses, such as fast Fourier transformation (FFT) [8, 10, 12, 16‐20], continuous Wavelet transformation [21, 22], an autoregressive model [9, 23, 33], have been used to calculate the frequency spectrum in a given time-sliding window with certain overlaps, but the results were smoothed due to window overlapping, and were delayed due to window length, causing the inevitable limited resolution of ERD.

Recently, we successfully developed a novel online lock-in amplifier (LIA) algorithm to estimate the amplitude modulation of motor imagery-related alpha ERD over SM1 [24]. LIA can reliably extract signal amplitude in a defined frequency band using a point-by point multiplication and filtering algorithms. Using this algorithm, the delay time, accuracy, and stability to estimate motor-related SM1 ERD were significantly improved compared with those calculated by the conventional online FFT, continuous Wavelet transformation, and autoregressive algorithms. These technical improvements to extract an ERD trace imply a potential advantage for a better trace of the excitatory status of SM1 in the context of a BCI.

Therefore, in this paper, we assessed the precision of LIA-based ERD tracking for estimation of corticospinal excitability using a TMS paradigm. We monitored the motor evoked potentials (MEPs) induced by single-pulse TMS over the primary motor cortex depending on the magnitude of SM1 ERD (i.e., 35% and 70%) extracted by the novel online LIA and conventional FFT algorithms during a motor imagery task of wrist extension in healthy participants. Then, we compared the peak-to-peak amplitudes of MEPs and their variabilities triggered by the different magnitudes of ERD extracted by the novel LIA and conventional FFT algorithms to investigate their precision of corticospinal excitability.

In the present study, we found greater MEP amplitude evoked by single-pulse TMS triggered by motor imagery-related alpha SM1 ERDs than that randomly applied at rest. This enhancement was associated with the magnitude of ERD (i.e., 35% and 70%) for both the FFT and LIA algorithms, which is consistent with a previous study [15]. Moreover, we found that the variabilities of peak-to-peak MEP amplitudes at the levels of ERD (i.e., 35% and 70%) calculated by the novel online LIA algorithm were reduced compared to those extracted by the conventional FFT algorithm. Based on the previous findings that LIA-based algorithm can detect motor-related SM1 ERD more accurately and stably than conventional FFT algorithm [24], the present results suggest that the accurate calculation of motor imagery-related SM1 ERDs extracted by the novel online LIA algorithm led to a precise estimation of human corticospinal excitability.

Extensive previous studies have consistently demonstrated that alpha/beta ERDs following kinesthetic motor imagery are a reliable biomarker for increased excitabilities in the sensorimotor cortex and corticospinal tract [28‐31]. Previous TMS studies have shown an inverse correlation between MEP amplitude and alpha band power [32, 33], and an inverse correlation between intracortical inhibition and the alpha ERD level during kinesthetic motor imagery [15]. Indeed, an additional analysis revealed that both algorithms detected targeted values of ERD over SM1 (Additional file 1: Figure S2). In addition, the magnitude of motor imagery-related SM1 ERD during kinesthetic motor imagery was associated with an increase in F-wave persistence, indicating the potentiation of spinal motoneurons during kinesthetic motor imagery accompanied with SM1 ERD [19]. The present findings also reproduced the correlation between the magnitude of motor imagery-related SM1 ERD and MEP amplitude. Moreover, we notably demonstrated the reduced variability of the results when the novel LIA algorithm was used compared to the FFT algorithm. This result suggests that motor imagery-related SM1 ERDs extracted by the LIA algorithm would be a more reliable biomarker to represent increased excitabilities in sensorimotor cortex and corticospinal tract than when an ordinary FFT-based algorithm is used.

A finger movement exercise aided by an ipsilesional SM1 ERD-based BCI stimulates SM1 while attempting paralyzed finger movements, resulting in an improvement of impaired movements [13]. Therefore, the precise estimation of the ERD trace by the newly proposed LIA algorithm and resulting improved correlation between SM1 ERD and peak-to-peak MEP could be potentially more beneficial for neurorehabilitation than the existing algorithm. For instance, delayed visual feedback attenuates error-based sensorimotor learning during prism adaptation [34, 35] and a gradual visuomotor rotation task [36]. Since BCI-mediated movement exercise is characterized by error-based learning through ERD state-dependent visual feedback [13, 20], the LIA-based instantaneous and accurate visual feedback of sensorimotor and corticospinal excitability assures a stable input-output association, resulting in efficient error-based learning. Further studies are awaited to investigate whether BCI learning using the LIA algorithm is more effective for neurorehabilitation than using the conventional method.

In general, lock-in measurements can extract instantaneous amplitude modulation in a defined single frequency by the point-by-point multiplication and integration of an input with a reference trigonometric basis signal [24]. In the present study, the input signal was defined as EEG amplitude modulation over a single frequency, which showed the most reactive ERD during motor imagery. In contrast, previous studies have demonstrated that motor imagery-related ERDs are associated with a broad frequency range including the alpha and beta frequency bands [31, 37, 38]. Although we demonstrated that the magnitudes of ERD calculated over a single frequency are associated with MEP amplitudes, the multi-frequencies calculation by parallel processing of LIAs may allow additional improvements in terms of the precise estimation of corticospinal excitability. In addition, in future study, we need to think to incorporate the other filtering algorithms such as Gaussian, a generalized matched filter to optimize signal noise ratio in real time, or machine learning algorithms to achieve the additional improvements of the LIA-based algorithms [39, 40].

The present study demonstrated that the variability of normalized peak-to-peak MEP amplitudes at different magnitudes of ERD extracted by a novel LIA algorithm were reduced more than those extracted by the conventional FFT algorithm. This result suggests that the LIA algorithm can be incorporated effectively into existing BCI paradigms to estimate human corticospinal excitability more precisely. This finding would be useful for the neurorehabilitation of patients with a movement disorder.