Quantitative Assessment of the Training Improvement in a Motor-Cognitive Task by Using EEG, ECG and EOG Signals

Abstract

Generally, the training evaluation methods consist in experts supervision and qualitative check of the operator’s skills improvement by asking them to perform specific tasks and by verifying the final performance. The aim of this work is to find out if it is possible to obtain quantitative information about the degree of the learning process throughout the training period by analyzing neuro-physiological signals, such as the electroencephalogram, the electrocardiogram and the electrooculogram. In fact, it is well known that such signals correlate with a variety of cognitive processes, e.g. attention, information processing, and working memory. A group of 10 subjects have been asked to train daily with the NASA multi-attribute-task-battery. During such training period the neuro-physiological, behavioral and subjective data have been collected. In particular, the neuro-physiological signals have been recorded on the first (T1), on the third (T3) and on the last training day (T5), while the behavioral and subjective data have been collected every day. Finally, all these data have been compared for a complete overview of the learning process and its relations with the neuro-physiological parameters. It has been shown how the integration of brain activity, in the theta and alpha frequency bands, with the autonomic parameters of heart rate and eyeblink rate could be used as metric for the evaluation of the learning progress, as well as the final training level reached by the subjects, in terms of request of cognitive resources.

Appendix: Estimation of Cortical Source Current Density

Cortical activity has been estimated from EEG scalp recordings by employing the high-resolution EEG technologies (Astolfi et al. 2006, 2007) with the use of the average head model from McGill University (Ding et al. 2005). Electrode positions over the scalp have been obtained individually for each subject by using the Photomodeler software (Eos System Inc). The cortical model consisted of about 8,000 dipoles uniformly disposed on the cortical surface. The estimation of the current density strength for each dipole has been obtained by solving the electromagnetic linear inverse problem (He et al. 2006; Nunez 1995). In particular, the solution of the linear system,

$${\mathbf{Ax}} \, = \, {\mathbf{b}} \, + \, {\mathbf{n}}$$

(2)

at a particular time instant t provides an estimation of the dipole source configuration x at time t that generates the measured EEG potential distribution b in the same instant. The system also includes the measurement noise n, assumed to be normally distributed. A is the lead field matrix, where each j-th column describes the potential distribution generated on the scalp electrodes by the j-th unitary dipole. The current density solution vector ξ of Eq. (2) has been obtained as (Babiloni et al. 2005; Grave de Peralta Menendez and Andino 1999):

$$\xi = \mathop {\arg \hbox{min} }\limits_{x} \left( {\left\| {{\mathbf{A}}x - b} \right\|_{{\mathbf{M}}}^{2} + \lambda^{2} \left\| x \right\|_{{\mathbf{N}}}^{2} } \right)$$

(3)

where M, N are the matrices associated to the metrics of the data and of the source space, respectively, ξ is the regularization parameter and ||x||_M represents the M norm of the vector x. The solution of Eq. (3) is given by the inverse operator G:

$${\varvec{\upxi}}\left( t \right) = {\mathbf{Gb}}\left( t \right),\,\,{\mathbf{G}} = {\mathbf{N}}^{ - 1} {\mathbf{A^{\prime}}}\left( {{\mathbf{AN}}^{ - 1} {\mathbf{A^{\prime}}} + \lambda {\mathbf{M}}^{ - 1} } \right)^{ - 1}$$

(4)

The optimal determination of the regularization term (ξ) of this linear system has been obtained by the L-curve approach (Hansen 1992; He et al. 2006). As a metric in the data space, the identity matrix has been used, while as a norm in the source space, the following metric was adopted:

$$\left( {{\mathbf{N}}^{ - 1} } \right)_{ii} = \left\| {{\mathbf{A}}_{ \cdot i} } \right\|^{ - 2}$$

(5)

where (N ⁻¹)_ii is the i-th element of the inverse of the diagonal matrix N and all the other matrix elements N_ij are set to 0. The L₂ norm of the i-th column of the lead field matrix A is denoted by ||A _.i ||. Using the relations described above, an estimation of the signed magnitude of the dipolar moment for each cortical dipoles has been obtained for each time point. As the orientation of the dipole has been defined to be perpendicular to the local cortical surface in the head model, the estimation process returned a scalar rather than a vector field. The spatial average of the signed magnitude, of all the dipoles belonging to a particular ROI at each time sample, was used to estimate the waveforms of cortical ROI activity in that ROI, indicated as ρ(t) to highlight their time-dependence. Spatial averaging can be expressed in terms of matrix multiplication by a matrix T. This matrix is sparse and has as many rows as ROIs, and as many columns as the number of dipole sources. ROI cortical current density waveforms can then be expressed as:

$${\varvec{\uprho}}\left( t \right) = {\mathbf{Tx}}\left( t \right) = {\mathbf{TGb}}\left( t \right) = {\mathbf{G}}_{\text{ROI}} {\mathbf{b}}\left( t \right),\,\,{\mathbf{G}}_{\text{ROI}} = {\mathbf{TG}}$$

(6)

where b(t) is the array of the waveforms recorded from the scalp electrodes and x(t) is the array of the cortical current density waveforms estimated at the cortical surface. The G _ROI matrix only depends on geometrical factors, and can thus be computed and stored off-line. The matrix multiplication can be interpreted as a spatial filtering of the scalp potential b(t), using the elements of G _ROI as weights. In this way, we could obtain time-varying waveforms at the level of different cortical areas. In the next paragraph, such cortical areas will be described as coincident with particular Brodmann areas for all the subjects involved in the present study.