New Cognitive Neurotechnology Facilitates Studies of Cortical-Subcortical Interactions

Abstract

Most of the studies employing neuroimaging have focused on cortical and subcortical signals individually to obtain neurophysiological signatures of cognitive functions. However, understanding the dynamic communication between the cortex and subcortical structures is essential for unraveling the neural correlates of cognition. In this quest, magnetoencephalography (MEG) and electroencephalography (EEG) are the methods of choice because they are non-invasive electrophysiological recording techniques with high temporal resolution. Sophisticated MEG/EEG source estimation techniques and network analysis methods developed recently can provide a more comprehensive understanding of the neurophysiological mechanisms of fundamental cognitive processes. Used together with noninvasive modulation of cortical-subcortical communication, these approaches may open up new possibilities for expanding the repertoire of non-invasive cognitive neurotechnology.

Significance of cortical and subcortical intercommunication

While the cerebral cortex is an important part of the brain, subcortical structures play a pivotal role in perceptual-motor, cognitive, and affective functions in humans (see Box 1). An integrative view of the dynamic interactions between these two parts of the brain, the cerebral cortex and subcortical structures, is key to understanding brain function, as evidenced by the dense network of overlapping feedforward and feedback anatomical connections linking one another. For example, the thalamus is a key station that relays motor and sensory signals to the cerebral cortex through direct thalamo-cortical connections [1, 2]. While a provincial cortical relay station, the thalamus also processes information through higher-order thalamic nuclei that connect with multiple cortical regions [3, 4], and is believed to have a critical role in regulating sleep, arousal, and consciousness [5, 6]. Beyond the thalamo-cortical network, the hippocampal-cortical network subserves learning and memory functions, the limbic-cortical system mediates emotional processing, and the striatal-cortical network facilitates motor function [7–9]. Thalamo-prefrontal communications have recently been reported to play an essential role in cognitive flexibility [10] and attentional and executive functions [11].

Box 1. Anatomy and function of major cortical-subcortical brain structures.

The human brain comprises a large number of cortical and subcortical areas each subserving distinct perceptual and cognitive functions (Figure I). The thalamus, a key region located at the core of the brain, is believed to primarily act as a relay station, or hub, between different subcortical areas and the cortex. Specifically, every sensory system (with the exception of the olfactory system) includes a thalamic nucleus that receives sensory signals and sends them through direct thalamo-cortical connections to the associated primary cortical area. Thus, for the visual system, information from the retina arrives to the lateral geniculate nucleus (LGN), which then relays it to the primary visual cortex. Similarly, the medial geniculate nucleus (MGN) relays information to the primary auditory cortex; the ventroposterior nucleus to the primary somatosensory cortex, and the ventroanterior nucleus to the primary motor cortex [2]. Beyond relaying information through these first-order nuclei, the thalamus is also believed to process information through higher-order (i.e., more complex in the processing hierarchy) thalamic nuclei [79]. These nuclei have, in general, both reciprocal and nonreciprocal connections with multiple cortical regions enabling them to mediate cortico-cortical information transfer within the cortex. A prominent example is the pulvinar, the largest subdivision of the thalamus with strong connectivity with the visual cortex. The thalamus also plays a critical role in regulating sleep, arousal, awareness, and is the primary candidate for the location of consciousness [5, 6]. The prefrontal cortex regulates goal-directed behavior, and other motor-related cortical and subcortical structures follow to complete the execution of movement. This includes the basal ganglia, the most critical subcortical structure for goal-directed actions [80]. For example, the largest structure of the basal ganglia, the striatum (a structure comprising the caudate nucleus and putamen among others), receives inputs from the substantia nigra that plays an important role in the regulation of movements. The hippocampus is essential for learning and memory function [81]. The limbic system, a set of brain structures comprising the thalamus, basal ganglia, hypothalamus, hippocampus, amygdala, and several other neighboring areas including the cingulate gyrus, is most heavily implicated in emotional processing [82].

Figure I.

Major cortical and subcortical brain structures.

Given the central role of cortical-subcortical communication in human cognition, a rigid approach focusing on one aspect of brain function, cortical or subcortical, is insufficient for a comprehensive characterization of human cognitive processes. However, to date, due to technological limitations most studies using non-invasive neuroimaging technology have focused on either cortical or subcortical signals to understand the neural underpinnings of cognitive function. As a result, fast subcortical dynamics and cortical-subcortical interactions have remained mostly inaccessible in humans (an exception is stereotactic EEG (SEEG, see Glossary) studies in patients [12]). Our current knowledge of cortical-subcortical intercommunication remains limited, and is primarily based on invasive studies using intracranial electrodes in animals [13].

However, recent advances in two non-invasive electrophysiological techniques, magnetoencephalography (MEG) and electroencephalography (EEG), and the capability to combine neuromodulation with electrophysiological recordings are promising better access to the dynamics and the interplay between cortical and subcortical structures [14, 15], opening important avenues for future research. Confirmatory recent evidence for these new approaches exists with intracortical recordings [16]. Taken together, it is timely to address the impact of these new approaches and the feasibility of a cortical-subcortical integrative approach in cognitive neurotechnology.

Neuroimaging technology for cortical-subcortical studies

Cortical-subcortical dynamics of human cognition have been probed non-invasively using functional neuroimaging techniques that are sensitive to the electrophysiological signals directly or map the associated hemodynamic responses (Figure 1). EEG [17] and MEG [18, 19] are the two non-invasive electrophysiological techniques. EEG uses an array of electrodes placed on the scalp to record voltage fluctuations, whereas MEG uses sensitive magnetic detectors called superconducting quantum interference devices (SQUIDs) [20] to measure the same primary electrical currents that generate the electric potential distributions recorded in EEG. MEG, in particular, can in the best cases provide localization of cortical sources at the cytoarchitectonic level [21]. Since EEG and MEG capture the electromagnetic fields produced by neuronal currents, they provide a fast and direct index of neuronal activity. Although MEG and EEG arrays’ large number of sensors are used for inferring sources of the underlying activity, the spatial resolutions of MEG and EEG are lower than that of the hemodynamic functional magnetic resonance imaging (fMRI) technique, especially for deep cortical and subcortical sources [16, 22]. This is in part due to the fact that the inverse problem of estimating the sources of extracranial MEG and EEG signals is ill-posed: there are different current distributions that can explain the same data, and the solutions are sensitive to noise. Therefore, anatomically and physiologically meaningful constraints are needed to render the solution unique. All MEG/EEG source estimation techniques rely on the same concept: one assumes a distribution of electrical currents (sources) in the brain and compares the MEG/EEG signals recorded to those predicted by the sources. This comparison is possible with help of an accurate enough solution of the Maxwell’s equations (the forward problem). The ill-posedness of the inverse problem means that there are several candidate source distributions which explain the measured data equally well. Different source estimation techniques are distinguished by the criterion which defines the best candidate. For example, one could require that there is only a limited number of focal sources that are confined to the cortical mantle or that some summarizing property (norm) of the source distribution is minimized. Detailed discussion of different source estimation approaches is beyond the scope of this article but has been previously discussed [18, 19]. As a result, EEG and MEG investigations of subcortical structures have remained scarce and even controversial when no corroborating evidence from other neuroimaging methods exists.

Figure 1. Neuroimaging technologies for non-invasive mapping of brain function.

EEG uses an array of electrodes placed on the scalp to record voltage fluctuations generated by electrical currents flowing through the brain due to active neurons. MEG uses sensitive magnetic sensors surrounding the head to measure the extraordinarily weak magnetic fields generated by active neurons. fMRI measures hemodynamic responses associated with brain activity, and primarily relies on the local BOLD signal, which detects the changes in blood oxygenation due to neural activity. NIRS uses infrared light to probe the concentration and oxygenation of hemoglobin (Hb) in the brain.

Two non-invasive neuroimaging methods that measure hemodynamic changes associated with brain activity are fMRI and near-infrared spectroscopy (NIRS). The most common form of fMRI uses the blood-oxygen-level-dependent (BOLD) contrast to detect changes in blood oxygenation that occur in response to neural activity, and it has become the most common method for in vivo imaging of brain function [23]. NIRS uses infrared lasers to detect light attenuation that depends on blood oxygenation [24]. Since both fMRI and NIRS sample the slow hemodynamic responses that have approximately 10–15 seconds duration and follow the actual electrophysiological activity with a delay [23], they provide an indirect measure of brain activity and offer limited temporal resolution in the order of seconds. The spatial resolution of fMRI is typically a few millimeters and the method can map activations in both cortical and subcortical structures [25]. Unlike fMRI, NIRS cannot detect deep brain activity because light is attenuated by bone and tissue.

A comprehensive understanding of the brain’s neural circuits requires a combined investigation of both its function and structure. Advanced processing of diffusion-weighted MRI (dwMRI), which measures the restricted diffusion of water in tissue, has opened a new non-invasive window to the anatomical connectivity of white matter fibers in the brain [26]. The dwMRI data can be processed to estimate and visualize the courses of axonal tracts in the brain using fiber tractography (FT) methods [27] (Figure 3A).

Figure 3. Brain connectivity.

(A) Structural connectivity delineates anatomical interconnections across different brain areas. Shown here are nerve tracts estimated with fiber tractography. The structural connectivity between subcortical and cortical structures is indicated in different colors: thalamo-frontal (red), thalamo-precentral (blue), thalamo-postcentral (green), thalamo-parietal–occipital (yellow), hippocampal-cortical (purple), and amygdalo-cortical (brown) connectivity. (B) Functional connectivity assesses the statistical dependence of neuroimaging signals from different brain regions. (C) Effective connectivity assesses the cause-effect relationships in neuroimaging signals from different brain regions (e.g., activation in one area directly causes a change of activation in another area).

In summary, anatomical MRI and dwMRI yield detailed anatomical images of brain structures and their connectivity. The current functional neuroimaging measures brain function either directly or indirectly. The direct methods (MEG/EEG) have a high temporal resolution, but their spatial resolution, especially for deep sources, is lower than the prevailing hemodynamic technique, fMRI. Consequently, investigations of cortical-subcortical interactions in humans have been difficult, producing equivocal results.

Novel approaches for integrative cortical and subcortical studies

Non-invasive neuroimaging techniques with a higher temporal resolution such as MEG and, its electric counterpart, EEG provide a temporal resolution suitable for studying brain dynamics associated with human cognition. Mapping both cortical and subcortical activation simultaneously in high temporal resolution will lead to a more comprehensive understanding of brain function. However, there are crucial limitations in current non-invasive neuroimaging, which have constrained the ways we can study the brain function. EEG and MEG offer the necessary temporal resolution, but sources deep beneath the brain easily escape detection. In addition to the increased distance from sources to the sensors, the targeted deep structures are small and may lack organized geometrical arrangement of neurons present in the cortex; both of these factors lead to a poor signal-to-noise ratio. Furthermore, a more fundamental problem stems from the fact that the subcortical structures are surrounded by the cortical mantle: measurements arising from the activity of deep structures can in principle be explained by a surrogate distribution of currents on the cortical surface. However, recent advances in EEG and MEG source imaging offer compelling evidence that localizing brain signals is, under favorable circumstances, possible even from deep brain structures. Specifically, MEG/EEG fields from subcortical sources can be distinguished from those generated by the cortex when the underlying salient cortical activity is sparse, i.e., is confined to a small number of cortical locations. This is a condition that is often encountered in neuroscience investigations [14]. A new technique linking MEG/EEG recordings to fMRI recordings using representational similarity analysis offers an alternate way to access whole-brain dynamics with combined high spatial and temporal resolution [15]. This method relies on the assumption that the same similarity structure is present in MEG/EEG and fMRI, at specific times in the former and at specific sites in the brain for the latter. Even though these new methods open new possibilities for the use of MEG/EEG independently, together, or in combination with fMRI, it must be emphasized that they do not provide a generally applicable “prescription” for the analysis. For example, the method described in [14] relies on the assumption of sparsity of the cortical activity while the representational similarity analysis predicates presentation of many different types of stimuli.

Experimentally, MEG recordings were recently shown to contain contributions from amygdala and hippocampal activity, which could be disentangled from activity arising from neocortical networks using independent component analysis [28] (Figure 2AB). Similarly, EEG recordings detected and correctly localized signals from the centromedial thalamus and the nucleus accumbens [16] (Figure 2CD). Importantly, both investigations corroborated their findings by analyzing simultaneously recorded SEEG data. In addition, MEG can localize deep thalamic sources and propagating activity from subcortical to cortical structures if a high number of responses is averaged to increase the signal-to-noise ratio [29, 30]. This has been shown in both healthy individuals as well as people with congenital blindness [31].

Figure 2. Consistency between MEG/EEG and simultaneous SEEG deep source estimation.

(A) Visualization of two SEEG electrodes (A and TBA) on the right hemisphere of a patient with drug-resistant epilepsy. Each electrode has multiple contacts, and the arrow indicates the contact with the highest correlation with an independent component of MEG data. (B) Source localization of the MEG independent component overlaid on patient MRI, showing a confidence interval that includes regions sampled by the mesial contacts of electrode A (highlighted in yellow and blue). (C) Visualization of an SEEG electrode on the nucleus accumbens of another patient with drug-resistant epilepsy, with contacts indicated in blue dots. (D) Localization of EEG sources with the highest correlation between the EEG and SEEG time courses in the alpha band. Sources were located in close proximity to the nucleus accumbens implantation region (warm colors). Figure adapted from [28] (AB) and [16] (CD) under a Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/).

Beyond methodological advances, the advent of new types of MEG sensors may also help to resolve cortical-subcortical networks in higher resolution. Optically-pumped magnetometers (OPMs), magnetic field sensors that monitor the perturbation of spin-polarized alkali metal vapor, do not need cooling [32, 33], unlike SQUIDs, which operate at liquid helium or liquid nitrogen temperatures. This allows the OPMs to be placed within millimeters from the scalp, and the sensor array can be adapted to the size and shape of the subject’s head, substantially increasing spatial resolution [34, 35]. While the benefits are greater for cortical than for subcortical sources, one can anticipate that a finer spatial specificity in resolving cortical signals would help isolate deep sources and lead to improved estimation of distributed brain networks. Notwithstanding, it has been recently demonstrated that subcortical signals from deep brain structures such as the hippocampus [36] and cerebellum [37] were measured using OPMs.

Previous studies have demonstrated the use of EEG in combination with neuromodulatory technology, such as transcranial magnetic stimulation (TMS) and transcranial current stimulation (TCS), which can selectively stimulate or suppress different brain areas. Unlike SQUID sensors, the OPM sensors recover quickly from transient strong magnetic fields [32, 33] and help to resolve cortical-subcortical networks in higher resolution [34, 35]. Therefore, OPMs can be used to record MEG in combination with TMS and TCS. The sophistication of such studies can be further enhanced with the introduction of dedicated multichannel stimulation arrays that enable rapid shifting of the target by means of electronic control [38].

Taken together, recent advances in MEG/EEG reconstruction algorithms and the ability to combine stimulation with MEG/EEG enable complex task-based paradigms in which neuromodulation is combined with neuroimaging to study brain circuits in an altered state and assess causal mechanisms in cognition. By using high-resolution dwMRI and large-scale tractography models, it is possible to target TMS selectively [39] and modulate cortical-subcortical networks in novel ways, and the effects of brain stimulation can be followed with both EEG and MEG.

Assessing functional interactions

The recent methodological and technological advances in non-invasive human brain imaging and stimulation allow neuroscientists to better characterize the fast neural interactions between cortical and subcortical structures and to construct large-scale brain connectivity networks. The taxonomy of functional connectivity metrics is extensive [40]. One important division is whether a metric is non-directional, in which case we simply seek to establish statistical dependence between brain regions, e.g., correlation, or directional, e.g., Granger causality. Also, the connectivity can be computed from broadband or narrowband signals, which is important, since cognitive processes can be associated with neural rhythms, such as attention (alpha band, 8–12 Hz) [41], motor control (beta band, 12–30 Hz) [42], and memory processes (theta band, 4–8 Hz [43]; gamma band, >30 Hz [44, 45]); for more detailed information on the relationship between cognitive functions and neural oscillations, please see [46, 47]. Furthermore, in many cases coupling of different oscillatory bands is involved. Narrowband signals can be effectively represented by their amplitude (envelope) and phase, giving rise to metrics that quantify amplitude-amplitude or phase-phase couplings. Phase-amplitude coupling across frequencies is also possible (cross-frequency coupling, CFC) [48, 49], which in the case of cortical and subcortical signals can also contribute to the assessment of cortical-subcortical modulatory communications.

Since anatomical connections between brain regions provide the structural basis for functional connectivity, fiber tractography is a valuable tool to find plausible functional interactions [50]. In particular, white matter fiber trajectories can inform functional brain networks by restricting direct neuronal interactions between pairs of regions that are anatomically connected. In functional networks, there is a distinction between functional connectivity, which assesses statistical dependencies between neural systems, and effective connectivity (Figure 3), which refers to the influence that one neural system exerts over another and is embodied, etc., in dynamic causal modelling (DCM) [51].

Finally, topological properties of brain networks offer the possibility of characterizing whole-brain organization. Graph theory provides a wide variety of tools for analyzing complex brain networks and opens the possibility to characterize the organization of the brain on a general level [52]. This is exemplified by a key finding that structural and functional brain networks are usually “small-world” networks, meaning that all brain regions are linked through relatively few intermediate steps.

Optimizing experimental design for cortical-subcortical interactions

In addition to measurement technologies and data analysis methods, experimental design has a key role in untangling the role of cortical-subcortical interaction processes. One important consideration is to design experimental paradigms that control for nuisance variables, such as strong low-level signals, in order to increase the chances of distinguishing the weak subcortical signals. For instance, using mismatching sequences of auditory stimuli for prediction-error generation, cortical-subcortical networks were recently revealed across prefrontal, auditory, and hippocampal cortices in theta-band (4–8 Hz) oscillations [53].

Another consideration is to leverage the stimulation techniques to optimally modulate cortico-subcortical communication. This includes manipulations of task performance by transcranial electromagnetic stimulation or sonication [54], timed critically and/or aimed at specific brain regions. For example, TMS can be used to establish which brain regions and at what processing stage are critical for a task. This can be achieved by systematically manipulating TMS pulse delivery times in a given brain area and monitoring the impact in disrupting performance [55–58].

Careful selection of tasks can also highlight different cortical-subcortical networks. Since the thalamus is involved in conscious perception and attentional control [59], different tasks could recruit different thalamo-cortical dynamics [60]. In addition, as the fundamental function of the thalamus is inhibitory control over sensory, motor, and other cognitive processing [61, 62], an experimental paradigm evoking inhibitory control can also elicit a considerable activation linked to thalamo-cortical intercommunications. Since a perception-action coupling task may drive anterior-posterior coupled cortical-subcortical interactions, this could be another possible task paradigm to employ. This is because information is processed in a looping fashion through a series of hierarchically organized brain regions and interactive connections constituting the perception-action cycle [63]. Learning and memory tasks are also appropriate paradigms for an integrative investigation, because the hippocampus, a deep subcortical structure, engages in the learning and memory processes and subsequently communicates with task-relevant cortical regions for the consolidation of memory [9]. Similarly, emotional processes, the neural correlates of which are distributed over the cortical-subcortical network, e.g., the limbic system, would benefit from the integrative approach; for example, activity changes of the amygdala, a small subcortical structure essential for emotional assessment, could not be effectively detected by MEG/EEG and fully interpreted in the whole-brain context otherwise [64–66]. Since almost all fundamental cognitive functions are based on cortical-subcortical interactive communications, the integrative approach is imperative for a better understanding of brain processes.

Concluding remarks and future perspectives

With sophisticated neuroimaging and neuromodulatory techniques it has become easier to study the dynamic interactions between the cortex and subcortical structures. A deeper understanding of the fundamental cognitive processes subserved by the cortical-subcortical network could open new avenues for applying this cognitive neurotechnology to therapeutics and neural rehabilitation. For instance, since higher-order thalamic nuclei are involved in pathophysiology at the early stages of schizophrenia [67], and are heavily interconnected with the prefrontal cortex believed to play a role in signal transmission and control, specific portions of the abnormal thalamo-cortical network are promising candidates for treatment by medical drugs or electromagnetic stimulation. Similarly, dopaminergic degeneration of substantia nigra inputs to the striatum leads to a compromised cortical-subcortical motor loop in Parkinson’s disease, highlighting the critical role of cortical-subcortical networks in brain disease in general. For healthy individuals, a cortical-subcortical integrative approach could manipulate the level of consciousness, the degree of memory consolidation, the amount of emotional assessment, or the rate of the sleep-wake cycle [9, 66, 68–70]. Thus, a future challenge is to design a cortical-subcortical neuromodulatory interfacing system that is appropriate for enhancing specific cognitive functions or treating specific psychiatric/neurological pathologies (see Outstanding Questions). Such a system could be driven by real-time brain activity to selectively modulate cortical-subcortical neurodynamics using TCS [71, 72], TMS [73, 74], or focused ultrasound (FUS) [75, 76], forming a closed-loop neuromodulatory interfacing system (Figure 4) that aims to augment human cognitive abilities or rehabilitate cognitive deficits. Finally, the integrative approach highlighted in this article should be extended to the cerebellum and the parts of the thalamus it communicates with to better understand the role of the cerebellum in motor and associative functions [77, 78].

Outstanding Questions.

• Which non-invasive neuromodulatory technology will be proven optimal to manipulate cortical-subcortical neurodynamics?

• What are the best combinations of different neuroimaging methods to investigate cortical-subcortical neuronal communications?

• Which subcortical structures will be more amenable to imaging and modulating with our current technology, and under which neurocognitive experiments?

• What cognitive functions or psychiatric/neurological pathologies are the most promising candidates for augmentation and rehabilitation with a cortical-subcortical neuromodulatory interface?

Figure 4. Schematic diagram conceptualizing a closed-loop cortical-subcortical neuromodulatory interface.

Different neuromodulatory techniques would selectively regulate cortical-subcortical interactions, aiming to improve a specific cognitive function. Examples of possible non-invasive neuromodulatory techniques are transcranial current stimulation (TCS), transcranial magnetic stimulation (TMS), focused ultrasound (FUS) sonication, or their combination. Through direct feedback from brain signals reflecting task-related neural activity, the neuromodulatory parameters would then be adjusted in real time within a close-loop neuromodulatory interfacing system. Repeated neuromodulatory sessions would aim to augment cognitive capabilities or rehabilitate patients with cognitive impairments.

Highlights.

• Advances in neuroimaging methods, combined with new brain magnetic sensors and neuromodulatory technology, open important new avenues for future research.

• Recent studies in brain electrophysiological imaging offer compelling evidence that under favorable circumstances localizing brain signals is possible even from deep brain structures.

• Methodological and technological developments have opened new possibilities for comprehensive investigation of cortical-subcortical temporal dynamics.

• A cortical-subcortical integrative approach is essential for understanding human cognition and the mind-brain relationship.

Glossary

Cross-frequency coupling (CFC)

A connectivity metric that measures the interaction between brain oscillations at two different frequency bands. Phase-amplitude coupling (PAC) is the most common type of CFC, where the phase of the lower frequency rhythm drives the power of the coupled higher frequency rhythm. Theta (4–8 Hz) to high-gamma (>50 Hz) CFC is critical for understanding effective local computations and large-scale inter-regional communication during cognitive processing in humans

Diffusion-weighted MRI (dwMRI)

An MRI-based technique that generates images sensitive to the restricted diffusion of water molecules in tissue. Since water molecules tend to diffuse more freely along the direction of axonal fascicles of neurons than across them, this information is used to estimate the organization of the white matter, i.e., the part of the brain made up of neuronal axons

Dynamic causal modelling (DCM)

A Bayesian framework for specifying causal neural interaction models and selecting the model with the highest evidence given the neural data. DCM embodies the interactions between neural populations observed directly or indirectly through functional neuroimaging data in differential equations

Electroencephalography (EEG)

An electrophysiological technique that uses an array of electrodes placed on the scalp to record voltage fluctuations generated by electrical currents in active neurons. EEG is typically measured using arrays of 32–256 electrodes. Since EEG measures neuronal electrical signals directly, EEG is useful for measuring the relative timing of cognitive events reflected in brain activity.

Focused ultrasound (FUS)

A novel non-invasive technology that uses ultrasonic waves for the spatially-accurate transcranial delivery of thermal or mechanical energy in a focused (in the order of millimeters) brain region. For example, low-intensity FUS modulates the function of the nervous system through the creation of reversible and transient functional interruptions that produce no pathological changes on histological examination

Fiber tractography (FT)

A modelling technique that estimates and visualizes nerve tracts using data collected by diffusion-weighted MRI. FT allows for both the identification and quantification of neuronal pathways (anatomical connectivity) non-invasively

Functional magnetic resonance imaging (fMRI)

A non-invasive neuroimaging technique that measures hemodynamic changes associated with brain activity. It primarily relies on the blood-oxygenation-level-dependent (BOLD) contrast, which detects the changes in blood oxygenation that occur in response to neural activity. Typical fMRI images have a 3–4 mm pixel size, but their temporal resolution is in the order of seconds owing to the inherently slow metabolic brain responses

Granger causality

A method for measuring causal effects between two variables. It assesses the ability to predict the future values of a time series using past values of another time series, and its mathematical formulation is based on linear regression modeling. Brain networks are often constructed by evaluating the Granger causality between every pair of brain regions

Graph theory

The study of graphs, which are structures that consist of nodes (also called vertices) connected by edges (also called links). In the context of brain networks, nodes are brain regions and edges are connectivities (anatomical or functional) between pairs of brain regions. Graph theory provides several measures, such as node degree, clustering coefficient, hubs, and modularity, to quantify different topological properties of brain networks

Magnetoencephalography (MEG)

A non-invasive neuroimaging technique that uses sensitive magnetic sensors surrounding the head to measure the extraordinarily weak magnetic fields generated by neuronal activity. MEG is due to the same current sources as the EEG. The magnetic sensors, called superconducting quantum interference devices (SQUIDs), measure signals in the 10–2000 fT (1 femto-Tesla = 10⁻¹⁵ Tesla) range. Modern MEG devices typically employ helmet-shaped sensor arrays with ~200 – 300 SQUIDs

Near-infrared spectroscopy (NIRS)

A non-invasive technology that uses infrared light to probe the concentration and oxygenation of hemoglobin in the brain. Near-infrared light at the approximate wavelength of 630–1300 nm has low absorption and can diffuse several centimeters through the tissue before being detected. NIRS uses multiple probes (optodes) arranged in a cap in arbitrary or standard EEG layout positions. The advancement in this field yields the development of a useful method for neuroimaging studies called “functional near-infrared spectroscopy (fNIRS)” that enables us to assess functional insights of cortical activities.

Optically pumped magnetometers (OPMs)

Magnetic field sensors that use lasers to polarize the spin of alkali metal vapor and then monitor spin perturbations caused by an external magnetic field. OPMs do not need cryogenics to operate, a major advantage compared to the current liquid helium-cooled magnetic sensors (SQUIDs)

Stereo-EEG (SEEG)

An intracranial EEG (iEEG) technology that uses stereotactic (three-dimensional) placement of electrodes in the human brain to invasively measure neural activity of predefined deeper structures of the brain. It is primarily used to record ictal and interictal activity from epileptic patients. By comparison, electrocorticography (ECoG) is another type of iEEG by using electrodes implanted in the subdural space

Transcranial current stimulation (TCS)

A non-invasive neuromodulatory technique that uses weak electrical currents to stimulate specific parts of the brain. Subcategories include transcranial direct current stimulation (tDCS) that uses direct currents, transcranial alternating current stimulation (tACS) that uses alternating currents, and transcranial random noise stimulation (tRNS) that uses randomly generated current levels. In all variants, a low-intensity current (~1 – 2 mA) is passed through two electrodes (anode and cathode) placed over the head to modulate neuronal activity. Typically, anodal stimulation acts to excite neuronal activity while cathodal stimulation inhibits or reduces neuronal activity

Transcranial magnetic stimulation (TMS)

A non-invasive neuromodulatory technique that uses coils placed outside the scalp to deliver strong magnetic pulses that induce currents inside the brain. Compared to transcranial current stimulation (TCS), TMS is spatially more precise, and neurons are excited directly by each TMS pulse. To induce long-term neurophysiological changes, many TMS pulses can be delivered in sequences, called repetitive TMS (rTMS); rTMS is useful for treating clinical depression and possibly other brain disorders