- 59 Downloads
Neuroimaging is an umbrella term that refers to various methods of measuring and imaging the function or structure of the nervous system, in particular the brain.
This chapter focuses on the physical and functional principles of magnetic resonance imaging (MRI) in humans. This includes different technical and conceptual approaches (e.g., structural MRI; functional MRI; connectivity analysis) as well as different statistical techniques (e.g., GLM; MVPA). Major advantages and disadvantages will be identified for each approach, and major conceptual issues concerning the association between brain anatomy or function and individual differences will be provided. Where possible, exemplar studies that make use of particular approaches or techniques are presented. Other imaging techniques (e.g., positron emission tomography) will only be briefly sketched.
KeywordsMultivoxel Pattern Analysis (MVPA) Major Conceptual Issues Representational Similarity Analysis (RSA) Single Positron Emission Computed Tomography (SPECT) Near-infrared Spectroscopy (NIRS)
Neuroimaging is an umbrella term that refers to various methods of measuring and imaging the function or structure of the nervous system, in particular the brain.
This chapter focuses on the principles and analysis of magnetic resonance imaging (MRI) in humans. This includes different technical approaches (e.g., structural MRI; functional MRI) as well as different statistical approaches (e.g., GLM; MVPA). Where possible, exemplar studies that make use of particular approaches or techniques are presented. Major advantages and disadvantages will be identified for each approach, and major conceptual issues concerning the association between brain anatomy or function and individual differences will be provided. A radically synthesized description of the physics underlying magnetic resonance imaging is given. The functional principles of other techniques such as computerized axial tomography (CT or CAT), positron emission tomography (PET), and single-positron emission computed tomography (SPECT), near infrared spectroscopy (NIRS) will only briefly be sketched.
Computerized Axial Tomography (CT or CAT)
Computerized axial tomography (CT) uses a series of two-dimensional X-rays of the head from different angles to reconstruct a three-dimensional image of the structure. It is commonly used to quickly diagnose bone trauma, hemorrhage, or tumors. The main advantage is the elimination of superimposed structures outside of the region of interest (ROI), as common in traditional 2D X-rays. Main disadvantage is the exposure to radiation which can potentially damage cells and DNA.
Positron Emission Tomography (PET)
Positron emission tomography (PET) measures the gamma rays emitted by positron-emitting radionuclides (tracer) that are introduced into the body shortly before the scanning session. Using glucose-analogues such as fludeoxyglucose, PET can be used to image metabolic activity via glucose use. The idea is that highly active brain regions use more glucose than inactive (or less active) regions, which in turn leads to a difference in concentration of the positron-emitting tracer. Seizures will appear as hypometabolic during scanning, allowing the exact determination of a given seizures’ center and extent. PET provides useful functional images but structural resolution is low.
Main disadvantages of PET are the exposure to ionizing radiation and the relatively costly procedure to produce the tracers.
Near Infrared Spectroscopy (NIRS)
Near infrared spectroscopy (NIRS) is a variant of spectroscopy and is based on the excitation of molecules by electromagnetic waves in the near infrared spectrum (760–2500 nm). Changes of hemoglobin can be detected through the skull up to a depth of about 1 cm beneath the inner surface of the skull and serve as a proxy for the level of activity of the underlying brain region. The main advantages of NIRS are its portability and easy and painless application that allows the measurement of more vulnerable samples such as patients or infants and children. It is less cost-intensive compared to PET or MRI and does not (in contrast to PET and MRI) require the participant to remain immobile during measurement.
The main disadvantage is the low spatial resolution.
Magnetic Resonance Imaging (MRI)
In contrast to other imaging techniques that allow investigating the internal structure and function of the human brain such as CT or PET, MRI does not expose the subject to any potentially harmful X-rays or radiation. MRI uses the physical phenomenon that certain atomic nuclei such as water, which is abundant in the human body, can absorb and emit radio frequency energy. To do so, the head is positioned in the center of an extremely strong magnetic field, usually 3 Tesla (3 T). The magnetic field forces all protons to line up with the magnetic field, rather than remaining in their usual random orientation. Next, a short radio frequency impulse (more precisely: oscillating magnetic field in the radio frequency range) is administered which flips the alignment of the protons by a given angular amount (maximally 90°). After this impulse is turned off, the protons fall back to their initial orientation (i.e., they relaxate) and emit a small amount of radio wave energy which can be measured by detectors placed over the subjects’ head. To determine the origin of the emitted signal, field gradients of varying strength along the three dimensions are added to the static magnetic field. Since the precession frequency (the frequency at which particles spin around an external magnetic field) is proportional to the strength of the magnetic field, applying the gradients causes spins at different spatial locations to precess at different rates. This allows measuring their individual contributions to the signal. By using different sequence parameters, different types of tissues can be visualized. For example, certain sequences predominantly visualize fat (e.g., gray and white matter), while others focus on water (e.g., in blood and cerebrospinal fluid).
Main disadvantage is the exposure to high magnetic field strength which bears the risk of inducing displacement in ferromagnetic implants (e.g., cochlear implant, pace maker) or other ferromagnetic materials that remain inside the body after surgery (e.g., nails after bone fractions). MRI requires the participants to remain immobile during measurement. This poses problems to certain psychiatric samples or children, for example. The confined environment within the MRI scanner may be problematic for persons suffering from claustrophobia.
Structural MRI refers to the analysis of structural aspects of the brain as measured with MRI. Structural MRI provides images of the brain with a very high spatial resolution (usually about 1 × 1 × 1 mm3) and very low temporal resolution (~ 5 min for the whole brain). The high spatial resolution allows the investigation of macrostructural anatomical properties of the brain such as the form of certain gyri and sulci or the volume within circumscribed brain areas, both in cross-sectional and in a longitudinal designs. Brain morphology differs between individuals. When comparing across different individuals, the problem of inter-interindividual variability concerning form, position, and size arises. A number of normalization procedures have been developed that allow registering the recorded brain images to a template that represents the “mean brain image” of, for example, 152 (MNI152 template) or 452 healthy participants (ICBM452 template).
One commonly used analysis approach is to measure the cortical thickness. This requires to additionally determine the border between gray and white matter. Since the cortical thickness varies across different brain areas between 2 and 4 mm, the scanning resolution of 1 mm requires statistical and model-guided procedures to increase subvoxel precision. After the resulting meshes of different subjects have been registered to a common reference, a point-by-point comparison of cortical thickness can be computed and correlated with other variables, such as the Big Five personality traits.
Using this approach, some authors argue for a biological basis of the Big Five (DeYoung et al. 2010). For example, areas that are involved in reward processing correlate with extraversion. Neuroticism, on the other hand, was associated with cortical thickness in areas that are commonly associated with threat and punishment. Others did not observe any correlation between personality measure neuroticism and cortical thickness (Hu et al. 2011), or opposite correlations for male and female participants (Blankstein et al. 2009). Bjornebekk et al. (2013) found thinner cortex in inferior frontal gyrus (IFG) to be associated with higher extraversion and speculate “that a thinner IFG reflects a structural correlate of this tendency for extroverts to be less inhibited in speech and more daring than their introvert opposites” (p. 205).
Voxel-based morphometry (VBM) uses a slightly different approach where the distribution of gray and white matter in different compartments is statistically compared across subjects. That is, for each voxel a binary or probabilistic value is determined that indicates the prevalence of cerebrospinal fluid, white matter, and gray matter, respectively. These maps are entered into a spatial smoothing procedure during which a Gaussian filter is applied to each of the three tissue maps. Smoothing reduces high-frequency noise and thereby increases signal to noise ratio. The resulting values are interpreted as “concentration” and index the local volume ratio of the different tissue classes (Ashburner and Friston 2000; Good et al. 2001). Finally, the concentration maps are entered to a second-level statistical procedure where local differences in the distribution of the three tissue classes can be compared, for example, between groups of subjects.
Liu and colleagues (2013) assessed the correlations of VMB-based distribution of gray and white matter with personality traits as measured by the complete version NEO Five Factor Inventory (McCrae and Costa 1987) in a sample of 227 participants. No associations were observed for gray matter concentration. In contrast, for white matter a negative correlation was observed between Conscientiousness and (a) right insula/rolandic operculum, and (b) left fusiform gyrus/parahippocampus. Authors interpret these correlations in terms of compensatory behavior. Since the described areas play an important role in “different complex sensory functions,” participants may “compensate for impaired information processing […] with higher conscientiousness” (p. 380). Coutinho et al. (2013) found extraversion to be negatively correlated with grey matter density in the middle frontal and orbitofrontal gyri while agreeableness was negatively correlated with grey matter density in the inferior parietal, middle occipital, and posterior cingulate gyri. Hu and colleagues (Hu et al. 2011) applied structural equational modeling to VBM measures to demonstrate the impact of various nuisance covariates such as age and gender on the subtle neuroanatomical correlations with personality traits.
Issues when Correlating Structural Information with Personality Measures
A number of criticisms can be raised when structural characteristics of the brain are correlated with personality measures. First, there is an inherent bigger-is-better or bigger-is-more assumption. Yet, whether or not larger brain areas signify that more cognitive resources are devoted to a given function is questionable. It is conceivable that functional refining is realized via adding and pruning connections between neurons rather than by merely adding more neurons. Moreover, the impact a given region may exert on a cognitive function also depends on other nonstructural parameters of brain activity such as synchronicity. Hence, the association between function and structure may be more subtle than equating bigger areas with more computational power for a certain function.
While the particular association between structural and functional brain characteristics and circumscribed personality traits remain elusive, personality changes following brain damage in patient groups (e.g., frontotemporal lobe degeneration) generally support the overall idea (Mahoney et al. 2011).
Diffusion Tensor Imaging
Diffusion Tensor Imaging (DTI) can be used to visualize the structure of white matter fibers in the brain. DTI measures the displacement of molecules per unit time. By exploiting the phenomenon that in the brain free displacement of water molecules is restricted by bounding fibers (i.e., it is anisotropic), DTI provides a measure of directional diffusion. DTI uses a pulsed sequence, where two magnetic field gradients are applied with brief temporal distance to each other. The spin magnetization at each position is specifically labeled by these pulses. A displacement of the molecules in the delay between two pulses causes a signal loss that is proportional to the amount of displacement. Especially in white matter, the orientation of the gradients influences contrast and signal decay. Signal loss is maximal in directions perpendicular to the fiber orientation. In combination with a multitude of gradients in different directions (e.g., 64 directions), a reconstruction of white matter architecture is achieved with DTI. That is, for each voxel a three-dimensional ellipsoid is computed that determines the principle direction of diffusion (parallel to the largest axis of the ellipsoid). From the ellipsoid, different DTI indices can be computed such as fractional anisotropy (FA), which is the normalized variance of the ellipsoid’s three eigenvalues. Combinations of eigenvalues can be used to demonstrate white matter pathology such as myelination (radial diffusivity) or axonal degeneration (axial diffusivity) (Alexander et al. 2007). This is helpful in detecting atypical white matter development or degenerative diseases (e.g., multiple sclerosis). To visualize white matter structure, left-to-right oriented fibers are coded in red, fibers in anterior-posterior direction coded in green, and fibers in inferior to superior direction in blue. Visualization of white matter tracts plays an important role in neurosurgical preoperative planning.
Main advantage of DTI is the painless and radiation-free possibility to measure and analyze structure of white matter in vivo. Main disadvantages concern some inherent methodological problems and the relatively long acquisition times (usually around 5–10 min) during which participants should remain stable. The algorithms underlying DTI are problematic for voxels containing different tissue types (e.g., at the boundary between CFS and white matter) and have difficulties determining DTI indices for voxels containing intersections between two nerve fibers.
Functional MRI (fMRI)
Functional MRI (fMRI) seeks to establish a relation between a given region’s change of neural activity and an experimentally induced change of stimulus and/or mental state. The idea is to examine the neurofunctional consequences of a given change in experimental condition in order to understand how a given ROI contributes to a cognitive mechanism at hand. For example, in a box-car design, participants change every 30 seconds between (1) repeatedly subtracting 7 from a starting number (e.g., 93 ➔86 ➔79 ➔72 ➔…) and (2) sentence reading (“The old man is reading a book.”). Comparing the activity between calculation and reading reveals areas that are more involved in mental arithmetic than during sentence reading, such as bilateral areas along the intraparietal sulcus and dorsolateral prefrontal cortex. Since two conditions are subtracted from each other, this is referred to as subtraction logic (Donders 1969; Posner et al. 1988). More recent designs do not require the block-wise presentation of stimuli but operate on an event-related schedule with fast and intermixed presentation of different conditions in an experimental design. It has been proven advantageous to include more than two expressions of the independent variable in a so-called parametric approach. For example, rather than dividing numerical distance between numbers in a numerical magnitude comparison task into categories small (<5) and large (>5), the numerical distance may serve as a parametric regressor in a given experiment.
Functional Principle and Preprocessing
Functional MRI uses the phenomenon that the brain uses glucose which is supplied via the blood. The regional cerebral blood flow (rCBF) flexibly adapts to the amount of energy needed in a given area of the brain. More active areas require more glucose and hence increased rCBF is observed. Glucose consumption burns oxygen. As a consequence, the ratio of deoxygenated to oxygenated hemoglobin changes locally in response to cortical activity, with oxygenated blood supply usually overshooting the actual demand. Since deoxygenated hemoglobin (paramagnetic) has different magnetic properties compared to oxygenated hemoglobin (diamagnetic), the compensatory change of their ratio in response to cortical activity causes an improved MR signal. The blood-oxygen-level dependent (BOLD) contrast hence provides an indirect measure of cortical activity at a local scale. The BOLD signal changes at a relatively slow temporal scale, usually reaching a maximum after 4–6 s after stimulation. The peak latency depends on the brain area under investigation.
During fMRI scanning, a number of 2D brain slices (e.g., 30 slices) are repeatedly acquired, often covering the entire brain. The acquisition of each entire volume usually takes between 1.5 and 3 s. Each slice comprises a matrix of subunits, known as voxels. Compared to structural MRI, fMRI has a lower spatial resolution with voxel dimensions around 2–4 mm per side (e.g., 2 × 2 × 2 mm3). Since each slice of a given volume is acquired at a different point in time, the temporal differences in acquisition require a correction, known as slice-time correction. A second major problem in fMRI are artifacts due to head motion. Each voxel is associated with a certain position in real space (i.e., the participant’s head inside the scanner) and numerically reflects the magnetic properties at that position. If, for example, two neighboring voxels cover white and gray matter, respectively, the numerical values differ by a large margin (e.g., 120 and 55). A change in nervous activity predominantly affects gray matter voxels. Consequently, in the next recorded volume, the corresponding values may change to 120 and 58, for example. If, however, the participant moves in-between the acquisition of two volumes, the voxel previously covering gray matter may now cover white matter. If undetected, we would erroneously conclude that activity has changed from initial 55 to 120 – a huge change in activity. During motion correction, the time-series of brain volumes are aligned via rigid body transformations along six dimensions (three spatial axes plus three rotations). After motion correction, nonlinear transformations are applied to coregister the individual brain scans with a given template (using information from the anatomical MRI). Finally, high-frequency noise is reduced via spatial smoothing.
Mass Univariate Statistics
During the mass univariate analysis approach, a general linear model (GLM) is applied to each voxel. The model parameters contain the experimental conditions as predictors (e.g., calculation & reading) and the head motion parameters as covariates (nuisance parameters). This model is convolved with the typical hemodynamic response function to predict the expected time-course for each voxel. As a result of model estimation, each of the regressors in the model (calculation, reading, head motion) is associated with a given parameter estimate (ß). The parameter estimates for each voxel is tested for significance (e.g., larger than zero; calculation > reading; …) across participants at the second level. Different approaches exist to protect the resulting statistical parametrical map (SPM) against the risk of multiple testing (a typical brain volume comprises ~60,000 voxels). In the resulting maps, each voxel is associated with a statistical significance value for a given contrast in the GLM (e.g., calculation > reading). The thresholded values can be projected on a brain template using a color code where significance ranges are associated with different colors (e.g., from red over orange and yellow to white).
Combining psychometric personality measures with fMRI has revealed that a wide range of functional responses in different regions are modulated by personality measures. In a recent review, Kennis and colleagues (2013) report numerous correlations with different scales of Gray’s reinforcement sensitivity theory (McCrae and Costa 1987; McNaughton and Corr 2004; McNaughton and Gray 2000). Activity in the amygdala, ventral prefrontal cortex (vPFC), and the basal ganglia (i.e., striatum) has been associated with the behavioral approach system (BAS), in particular, in response to reward and expectance of reward. Reward expectancy was also associated with activity in anterior cingulate cortex (ACC). Less clear was the picture concerning the neural correlates of the behavioral inhibition system (BIS) and the fight-flight-or-freeze system (FFFS). Both appear to be correlated with activity in PFC during cognitive tasks and with negative stimuli such as punishment expectancy (Bruhl et al. 2011) or learning of negative emotional associations (Hooker et al. 2008).
Another approach exploits the fact that rCBF decreases upon repeated presentation of a given piece of information (adaptation or repetition suppression). Using such adaptation paradigms, several studies found ventral mPFC to be associated with the encoding of a person’s traits (Heleven and Van Overwalle 2016; Ma et al. 2014).
Issues when Correlating Functional Measures with Personality Measures
Many of the issues when correlating structural information with personality measures hold when it comes to associations with functional MRI. That is, associating a given trait with a given ROI rests on the assumption that more activation in that ROI in a given task means that more intensive or efficient cognitive processing results from increased rCBF. For example, when reporting a correlation of amygdala activity during a fear learning task with high neuroticism, the authors assume that an “increased sensitivity in the neural mechanism for fear learning […] leads to enhanced encoding of fear associations,” which is, in turn, reflected on the neural level by a stronger BOLD response (p. 2709; Hooker et al. 2008). Yet, this is not necessarily the case and theoretically the opposite may be true; higher cortical efficiency may express via a reduced fMRI response, rather than an increased fMRI response.
Another problem arises from the idea of associating one region of the brain with one cognitive function (e.g., when stating that areas that are involved in reward processing correlate with extraversion). While a given brain area may contribute to the function at hand, it usually also contributes to a number of other functions which may or may not be relevant for the personality measure at hand. That is, picking the one function of an area that may be conceptually associated with a personality concept over-accentuates the area’s functional selectivity.
Further, small sample sizes (e.g., below 20 participants in Brühl et al. 2011 and Hooker et al. 2008) and the association of broad high-level personality measures (e.g., neuroticism) with a single cognitive task (i.e., one learning experiment) that measures a given cognitive function may be problematic in terms of generalizability. Convergent evidence from multiple studies may alleviate this issue.
Vul and colleagues (2012) pointed to another theoretical issue arising from the notion that, theoretically, the correlation between any brain measure and any personality measure cannot exceed the product of their respective reliabilities. Vul and colleagues argue that a “non-independence error” is responsible for the fact that many empirical studies report values that exceeded this theoretical threshold. That is, voxels are selected based on whether or not their correlation with an external parameter exceeds a given statistical threshold. In a second step, a mean correlation is computed from the selected voxels only, leading to artificially inflated correlations because all nonsignificant voxels had been eliminated before.
Instead of statistically testing the significance of a given contrast in GLM for each voxel, decoding techniques such as multivoxel pattern analysis (MVPA) rely on supervised learning algorithms that operate on spatial patterns across an array of voxels (i.e., the features). On a trial-by-trial basis where each trial represents one exemplar, a statistical algorithm (e.g., support vector machines) is trained to differentiate between two (or more) experimental conditions (e.g., calculation vs. reading), such that the classifier learns to associate each pattern with one of the conditions. This is equivalent to searching for a hyperplane in feature space (where each feature represents one dimension) that ideally separates the experimental conditions. After training, the classifier is cross-validated with an unknown (i.e., independent) portion of the data for which it is asked to “predict” the true label (e.g., reading or calculation). This procedure is repeated while leaving out different parts of the data during training. Accuracy is averaged across repetitions and tested against chance level of classification (e.g., 50% for two conditions). Ideally, the accuracy reaches 100% correct. Here, voxel activity is used to predict experimental information (e.g., task). Decoding allows responding to a slightly different type of questions concerning the function of a given region of interest (ROI) in the brain (Haxby et al. 2014). While mass univariate analysis asks what a given ROI’s function is, decoding allows analyzing what information is represented in a given ROI and how it is organized. For example, a given ROI may be equally and indistinguishably active both for addition and subtraction compared to reading. Yet, the representational patterns may still differ fundamentally and systematically between these arithmetic operations (Knops et al. 2009; Knops and Willmes 2014), affording decoding of arithmetic operations.
Decoding has been successfully applied to predict which of four characters participants were imagining based on activity patterns from medial prefrontal cortex (Hassabis et al. 2014). Characters’ personalities differed along two traits (agreeableness and extraversion). Activity patterns in lateral temporal cortex and posterior cingulate cortex (pCC) allowed for predicting the degree of agreeableness and extraversion, respectively. Authors proposed that medial PFC encodes a complex personality model for others while lateral temporal cortex and pCC code for more specified personality traits (Hassabis et al. 2014).
Representational Similarity Analysis (RSA)
Representational similarity analysis (Kriegeskorte et al. 2008) seeks to analyze the state of the brain in response to a given stimulus, plan, decision, or planned action. It understands the ROI’s representation as a point in multidimensional space. Voxels or neurons are the dimensions. Given two or more points in this space (representing two or more stimuli, plans, etc.) one may compute the representational distance between them as a proxy for the representational dissimilarity between them. When extending the stimulus space, one may compute a representational dissimilarity matrix (RDM), reflecting the pair-wise dissimilarity values between different stimuli (plans, decisions, etc.). For example, the RDM for six objects (apple, pear, melon, mobile phone, screwdriver, and tennis ball) may reveal a higher-level semantic distinction between man-made objects versus natural objects.
This recent approach has been exploited to illustrate that similarities in subjective social space are mirrored on the neural level in left temporoparietal junction (TPJ), the left fusiform gyrus, and the subcallosal ventromedial prefrontal cortex (vmPFC; Dziura and Thompson 2017). Hence, learning about differences, similarities, and relations between others may be supported by these areas, authors conclude (Dziura and Thompson 2017).
Most cognitive functions rely on the joint use of several interconnected brain regions. The domain of network neuroscience (Sporns and Betzel 2016) focuses on the analysis of the relation between brain regions that can be represented as a set of nodes and edges that form a network. Most approaches use the Pearson correlation of the time course between two (or more) foci in the brain in response to a given stimulus or task (functional connectivity or effective connectivity) or during a state of rest (resting state connectivity). Several approaches have been proposed.
Psycho-Physiological Interactions (PPI)
PPI (Friston et al. 1997) tests connectivity changes as a function of task condition. In other words, PPI analysis demonstrates functionally significant brain-wide connectivity changes in response to experimentally induced changes in a given seed ROI. The term “seed region” refers to any set of voxels that has been selected for conceptual reasons and for which the time course is extracted and entered to PPI analysis. Technically, the PPI regressor represents the product of the deconvolved ROI time course and the task regressor (Gitelman et al. 2003). For each voxel outside the ROI, the correlation with the PPI regressor is computed individually for each participant. The resulting maps are entered into a second level model (participant as random factor) to assess statistical significance of the observed connectivity changes on a group level.
PPI has been used to unravel the functional network that subserves the recognition of other persons based on information about these persons’ bodies and traits (e.g., “He cut in front of the man in line” for an inconsiderate person; Greven et al. 2016). Authors argue for a “who” system that combines these pieces of information and comprises of fusiform gyrus (bodily appearance), and temporal poles and temporal-parietal junction as part of a network that makes inferences about other’s thoughts and traits (Greven et al. 2016; Mitchell et al. 2002).
Dynamic Causal Modeling (DCM)
Dynamic causal modeling (DCM) aims at reconstructing how an observed pattern of activity has been produced by the interaction of distributed dynamical systems. To this aim, differential equations are used to define dynamic causal models, which, in turn, model latent states of nodes in a probabilistic graphic model. This requires the researcher to define a priori realistic models of brain regions that contribute to the task at hand and how these are interconnected. The observed data is used to test these models, that is, how the hidden states of each node map onto the measured responses. Bayesian model selection is used to select the best fitting and most parsimonious model that corresponds best to the observed data (Stephan et al. 2009). In contrast to PPI, DCM makes directional predictions (e.g., region A drives/influences region B).
Dima and colleagues (2015) used DCM to investigate the impact of the big five personality traits on effective connectivity in a working memory network (bilateral parietal cortex, anterior cingulate cortex, and dorsolateral prefrontal cortex). While neuroticism reduced short-term plasticity in this network, conscientiousness increased modulation of connectivity in this network (Dima et al. 2015).
Resting State fMRI
The brain is not always engaged in cognitive processing. A set of brain region has been shown to be most active when the participant is in a state of mind wandering (Greicius et al. 2003; Raichle et al. 2001). This network comprises of posterior cingulate cortex, medial prefrontal cortex, angular gyrus, and retrosplenial cortex/precuneus. Two main approaches emerged and are continuously updated and enriched, independent component analysis (ICA) and seed-based analyses. Individual variation in resting state connectivity has been proposed as a potential biomarker for psychiatric disorders and response to treatment in a context of personalized psychiatry (Finn et al. 2015; Finn and Todd Constable 2016).
Seed-based analysis is a model-based approach in which the time-course from an a-priori selected ROI (seed) is extracted and correlated with the time-course of all remaining voxels in the brain. As a result, a network of brain regions can be defined for which the time-course is highly correlated with each other. Major disadvantage of this approach is the highly subjective selection of a seed region that strongly influences the results.
This method was used to investigate the relation between DMN activity and stressor-evoked cardiovascular reactivity as a proxy for the emotional reactions to conflict which is, in turn, associated with agreeableness (Ryan et al. 2011). Ryan et al. (2011) found that more positive functional connectivity between posterior and anterior cingulate cortex correlated positively with agreeableness and mediated the relation between agreeableness and stressor-evoked cardiovascular reactivity. Connectivity between OFC and putamen was positively correlated with trait impulsivity, a concept from Gray’s model of personality and addiction (Angelides et al. 2017; Franken et al. 2006).
Independent Component Analysis (ICA)
ICA is a mathematical technique to decompose a given (time-resolved) data set into a number of maximally independent components. In the context of fMRI, ICA is used to spatially separate regions of joint activity modulations that are maximally distinct from other regions. Major disadvantage is that the user needs to differentiate noise-based components from artifacts and actual functional components, which requires some experience and mathematical understanding. Major advantage is the data-driven and objective nature of the resulting componential structure.
Aberrant connectivity in the DMN is associated with psychiatric disorders such as borderline personality disorder and impulsivity (Wolf et al. 2011).
- Blankstein, U., Chen, J. Y., Mincic, A. M., McGrath, P. A., & Davis, K. D. (2009). The complex minds of teenagers: Neuroanatomy of personality differs between sexes. Neuropsychologia, 47(2), 599–603. https://doi.org/10.1016/j.neuropsychologia.2008.10.014.CrossRefPubMedGoogle Scholar
- Bjornebekk, A., Fjell, A. M., Walhovd, K. B., Grydeland, H., Torgersen, S., & Westlye, L. T. (2013). Neuronal correlates of the five factor model (FFM) of human personality: Multimodal imaging in a large healthy sample. Neuroimage, 65, 194–208. https://doi.org/10.1016/j.neuroimage.2012.10.009.CrossRefPubMedGoogle Scholar
- Bruhl, A. B., Viebke, M. C., Baumgartner, T., Kaffenberger, T., & Herwig, U. (2011). Neural correlates of personality dimensions and affective measures during the anticipation of emotional stimuli. Brain Imaging and Behavior, 5(2), 86–96. https://doi.org/10.1007/s11682-011-9114-7.CrossRefPubMedGoogle Scholar
- DeYoung, C. G., Hirsh, J. B., Shane, M. S., Papademetris, X., Rajeevan, N., & Gray, J. R. (2010). Testing predictions from personality neuroscience. Brain structure and the big five. Psychological Science, 21(6), 820–828. https://doi.org/10.1177/0956797610370159.CrossRefPubMedPubMedCentralGoogle Scholar
- Dima, D., Friston, K. J., Stephan, K. E., & Frangou, S. (2015). Neuroticism and conscientiousness respectively constrain and facilitate short-term plasticity within the working memory neural network. Human Brain Mapping, 36(10), 4158–4163. https://doi.org/10.1002/hbm.22906.CrossRefPubMedPubMedCentralGoogle Scholar
- Dziura, S. L., & Thompson, J. C. (2017). The neural representational space of social memory. bioRxiv. https://doi.org/10.1101/130351.
- Finn, E. S., Shen, X., Scheinost, D., Rosenberg, M. D., Huang, J., Chun, M. M., . . . Constable, R. T. (2015). Functional connectome fingerprinting: Identifying individuals using patterns of brain connectivity. Nature Neuroscience, 18(11), 1664–1671. doi: https://doi.org/10.1038/nn.4135.CrossRefPubMedPubMedCentralGoogle Scholar
- Greicius, M. D., Krasnow, B., Reiss, A. L., & Menon, V. (2003). Functional connectivity in the resting brain: A network analysis of the default mode hypothesis. Proceedings of the National Academy of Sciences of the United States of America, 100(1), 253–258. https://doi.org/10.1073/pnas.0135058100.CrossRefPubMedGoogle Scholar
- Haxby, J. V., Connolly, A. C., & Guntupalli, J. S. (2014). Decoding neural representational spaces using multivariate pattern analysis. Annual Review of Neuroscience, 37, 435–456. https://doi.org/10.1146/annurev-neuro-062012-170325.CrossRefPubMedGoogle Scholar
- Hooker, C. I., Verosky, S. C., Miyakawa, A., Knight, R. T., & D'Esposito, M. (2008). The influence of personality on neural mechanisms of observational fear and reward learning. Neuropsychologia, 46(11), 2709–2724. https://doi.org/10.1016/j.neuropsychologia.2008.05.005.CrossRefPubMedPubMedCentralGoogle Scholar
- Hu, X., Erb, M., Ackermann, H., Martin, J. A., Grodd, W., & Reiterer, S. M. (2011). Voxel-based morphometry studies of personality: Issue of statistical model specification – effect of nuisance covariates. NeuroImage, 54(3), 1994–2005. https://doi.org/10.1016/j.neuroimage.2010.10.024.CrossRefPubMedGoogle Scholar
- Ma, N., Baetens, K., Vandekerckhove, M., Kestemont, J., Fias, W., & Van Overwalle, F. (2014). Traits are represented in the medial prefrontal cortex: An fMRI adaptation study. Social Cognitive and Affective Neuroscience, 9(8), 1185–1192. https://doi.org/10.1093/scan/nst098.CrossRefPubMedGoogle Scholar
- Mahoney, C. J., Rohrer, J. D., Omar, R., Rossor, M. N., & Warren, J. D. (2011). Neuroanatomical profiles of personality change in frontotemporal lobar degeneration. The British Journal of Psychiatry, 198(5), 365–372. https://doi.org/10.1192/bjp.bp.110.082677.CrossRefPubMedPubMedCentralGoogle Scholar
- Mitchell, J. P., Heatherton, T. F., & Macrae, C. N. (2002). Distinct neural systems subserve person and object knowledge. Proceedings of the National Academy of Sciences of the United States of America, 99(23), 15238–15243. https://doi.org/10.1073/pnas.232395699.CrossRefPubMedPubMedCentralGoogle Scholar
- Raichle, M. E., MacLeod, A. M., Snyder, A. Z., Powers, W. J., Gusnard, D. A., & Shulman, G. L. (2001). A default mode of brain function. Proceedings of the National Academy of Sciences of the United States of America, 98(2), 676–682. https://doi.org/10.1073/pnas.98.2.676.CrossRefPubMedPubMedCentralGoogle Scholar
- Ryan, J. P., Sheu, L. K., & Gianaros, P. J. (2011). Resting state functional connectivity within the cingulate cortex jointly predicts agreeableness and stressor-evoked cardiovascular reactivity. NeuroImage, 55(1), 363–370. https://doi.org/10.1016/j.neuroimage.2010.11.064.CrossRefPubMedGoogle Scholar
- Sporns, O., & Betzel, R. F. (2016). Modular brain networks. Annual Review of Psychology, 67, 613–640. https://doi.org/10.1146/annurev-psych-122414-033634.CrossRefPubMedGoogle Scholar
- Wolf, R. C., Sambataro, F., Vasic, N., Schmid, M., Thomann, P. A., Bienentreu, S. D., & Wolf, N. D. (2011). Aberrant connectivity of resting-state networks in borderline personality disorder. Journal of Psychiatry & Neuroscience, 36(6), 402–411. https://doi.org/10.1503/jpn.100150.CrossRefGoogle Scholar