A Haemodynamic Brain–Computer Interface Based on Real-Time Classification of near Infrared Spectroscopy Signals during Motor Imagery and Mental Arithmetic

Neural activity in the brain is accompanied by changes in cerebral blood flow (CBF) and blood oxygenation that are detectable with functional magnetic resonance imaging (fMRI) techniques. In this paper, recent mathematical models of this hemodynamic response are reviewed and integrated. Models are described for: (1) the blood oxygenation level dependent (BOLD) signal as a function of changes in cerebral oxygen extraction fraction (E) and cerebral blood volume (CBV); (2) the balloon model, proposed to describe the transient dynamics of CBV and deoxy-hemoglobin (Hb) and how they affect the BOLD signal; (3) neurovascular coupling, relating the responses in CBF and cerebral metabolic rate of oxygen (CMRO(2)) to the neural activity response; and (4) a simple model for the temporal nonlinearity of the neural response itself. These models are integrated into a mathematical framework describing the steps linking a stimulus to the measured BOLD and CBF responses. Experimental results examining transient features of the BOLD response (post-stimulus undershoot and initial dip), nonlinearities of the hemodynamic response, and the role of the physiologic baseline state in altering the BOLD signal are discussed in the context of the proposed models. Quantitative modeling of the hemodynamic response, when combined with experimental data measuring both the BOLD and CBF responses, makes possible a more specific and quantitative assessment of brain physiology than is possible with standard BOLD imaging alone. This approach has the potential to enhance numerous studies of brain function in development, health, and disease.

We employed CT scanning to correlate scalp markers placed according to the international 10-20 system with underlying cerebral structures. Subjects were 12 normal volunteers. Measurements included assessment for cranial asymmetry to determine the effect of skull asymmetry on cortical location of electrodes. Results were correlated with the cortical histological map of Brodmann. Primary cortical locations agree well with previously published data and provide cortical localization in greater detail than previous studies. Variability of cortical electrode location was substantial in some cases and not related to cranial asymmetry. The results indicate that CT scanning or other neuroimaging techniques which reveal detailed cerebral anatomy would be potentially highly useful in defining the generators of electrocerebral potentials recorded from the scalp.

The registration of functional brain data to the common brain space offers great advantages for inter-modal data integration and sharing. However, this is difficult to achieve in functional near-infrared spectroscopy (fNIRS) because fNIRS data are primary obtained from the head surface and lack structural information of the measured brain. Therefore, in our previous articles, we presented a method for probabilistic registration of fNIRS data to the standard Montreal Neurological Institute (MNI) template through international 10-20 system without using the subject's magnetic resonance image (MRI). In the current study, we demonstrate our method with a new statistical model to facilitate group studies and provide information on different components of variability. We adopt an analysis similar to the single-factor one-way classification analysis of variance based on random effects model to examine the variability involved in our improvised method of probabilistic registration of fNIRS data. We tested this method by registering head surface data of twelve subjects to seventeen reference MRI data sets and found that the standard deviation in probabilistic registration thus performed for given head surface points is approximately within the range of 4.7 to 7.0 mm. This means that, if the spatial registration error is within an acceptable tolerance limit, it is possible to perform multi-subject fNIRS analysis to make inference at the population level and to provide information on positional variability in the population, even when subjects' MRIs are not available. In essence, the current method enables the multi-subject fNIRS data to be presented in the MNI space with clear description of associated positional variability. Such data presentation on a common platform, will not only strengthen the validity of the population analysis of fNIRS studies, but will also facilitate both intra- and inter-modal data sharing among the neuroimaging community.