Single-trial classification of motor imagery differing in task complexity: a functional near-infrared spectroscopy study

Abstract

Background: For brain computer interfaces (BCIs), which may be valuable in neurorehabilitation, brain signals derived from mental activation can be monitored by non-invasive methods, such as functional near-infrared spectroscopy (fNIRS). Single-trial classification is important for this purpose and this was the aim of the presented study. In particular, we aimed to investigate a combined approach: 1) offline single-trial classification of brain signals derived from a novel wireless fNIRS instrument; 2) to use motor imagery (MI) as mental task thereby discriminating between MI signals in response to different tasks complexities, i.e. simple and complex MI tasks.

Methods: 12 subjects were asked to imagine either a simple finger-tapping task using their right thumb or a complex sequential finger-tapping task using all fingers of their right hand. fNIRS was recorded over secondary motor areas of the contralateral hemisphere. Using Fisher's linear discriminant analysis (FLDA) and cross validation, we selected for each subject a best-performing feature combination consisting of 1) one out of three channel, 2) an analysis time interval ranging from 5-15 s after stimulation onset and 3) up to four Δ[O2Hb] signal features (Δ[O2Hb] mean signal amplitudes, variance, skewness and kurtosis).

Results: The results of our single-trial classification showed that using the simple combination set of channels, time intervals and up to four Δ[O2Hb] signal features comprising Δ[O2Hb] mean signal amplitudes, variance, skewness and kurtosis, it was possible to discriminate single-trials of MI tasks differing in complexity, i.e. simple versus complex tasks (inter-task paired t-test p ≤ 0.001), over secondary motor areas with an average classification accuracy of 81%.

Conclusions: Although the classification accuracies look promising they are nevertheless subject of considerable subject-to-subject variability. In the discussion we address each of these aspects, their limitations for future approaches in single-trial classification and their relevance for neurorehabilitation.

Figure 1

Experimental design. An example of the trial layout showing the stimulation periods (15 s) alternating with the rest periods (20 s) during which subjects had to either execute or imagine finger tapping on a keyboard. Start of the stimulation was indicated by the word GO.

Figure 2

Wireless fNIRS sensor. a) Top-view: schematic of light sources (L1, L2, L3, and L4) and detectors (D1, D2, D3, and D4) on the sensor. b) Wireless fNIRS sensor with casing; (red) light sources, (blue) detectors, (1) analog and wireless communications and power-supply electronics, (2) optical probe [3]. The centre of the sensor was positioned presumably covering position F3 according to the 10-20 system [30]. Three channels were considered for analysis. D1-L1 was positioned in cranial direction, D4-L4 in caudal direction.

Figure 3

Mean Δ[O₂Hb] and Δ[HHb] profile. Mean Δ[O₂Hb] and Δ[HHb] (mean ± SE μmol/l) on the overall-subject-level averaged over 12 trials for each channel separately (channel 1 [black], channel 2 [dark gray], channel 3 [light gray]), of the contralateral (right) hemispheres during performance of MI-simple and MI-complex. Shown are also relevant significances of paired t-test (CI 95%, p-values) of Δ[O₂Hb] between the two tasks. The second y-axis (green) represents the Δ[O₂Hb] signal-to-noise ratio (SNR, defined as the ratio of the mean signal to its standard deviation) for each channel; the values of each SNR are shown below.

Figure 4

Analysis time intervals. Results of the analysis time intervals across subjects ranked by classification accuracy (%). Shown are the ranges of individual analysis intervals used for classification.

Figure 5

Sample subjects Δ[O₂Hb] and Δ[HHb] profile. Averaged Δ[O₂Hb] (red) and Δ[HHb] (blue) responses in two sample subjects (subject 1 and 2) corresponding to the classification defined in Table 2. After the rest period (20 s) the on- and offset of the stimulation period (15 s) are indicated by dashed lines from time = 0 - 15 s. The regions highlighted with a box correspond to the time intervals selected for the classification as specified in Table 2.

Figure 6

Correlations between classification accuracy and feature value. Scatter plots illustrating the correlations between the classification accuracies (%) and the averaged feature values over all trials for each subject (each dot represents one subject, only those subjects are shown for whom the feature was selected for classification). Separate plots are shown for the significant findings in two of the four feature: (Left) Δ[O₂Hb] variance was negatively correlated with classification accuracy in both conditions (MI-simple: r = -0.688*, p = 0.028; MI-complex: r = -0.701*, p = 0.024); (Right) Δ[O₂Hb] skewness was negatively correlated with classification accuracy in MI-simple (r = -0.850*, p = 0.032) and positively correlated in MI-complex (r = 0.854*, p = 0.031).