5. ESTUDIO ECONOMICO Y FINANCIERO
5.7 PROYECCIÓN DE GASTOS Y ADMINISTRACIÓN DE VENTAS
Functional MRI is a powerful tool employed in neuroimaging which allows to capture dynamic metabolic changes in brain tissue and offers an in-vivo method to measure fluctuations in brain activity. Estimating neural and glial activity is made possible by the haemodynamic response that is initiated in case of increased brain activity. In order to meet the metabolic demands of the activated brain region, an oversupply of oxygenated blood is delivered to the respective region. Oxygenated and deoxygenated haemoglobin in the blood differ in their magnetic susceptibility, therefore, the haemodynamic response induces a change in the magnetic properties. These blood oxygenated level dependent (BOLD) changes lead to a difference in the signal on T2*weighted MR images and allows to differentiate between regions of different metabolic demands (Chen & Glover, 2015; Huettel, Song, & McCarthy, 2014).
Fluctuations of brain activity occur during specific tasks but also in phases of rest such as relaxation or sleep (Buckner et al., 2013). While the investigation of task-related activity provides information on which brain regions are involved in a specific cognitive process, analyses of functional connectivity (FC) offer insight into the functional coupling of brain regions. These two differential approaches are presented next.
3.1.1 Analyses of task-related activity
Analyses of task-related brain activity stem from the basic principle to evoke and compare different mental states in the subject. Based on the observed differences between the two states, it is possible to derive the relative involvement of brain areas in the evoked states. Paradigms in fMRI studies differ in many aspects such as computational strategies to compare the BOLD signal, stimulus presentation, image acquisition and image analysis (Huettel, Song & McCarty, 2014). The presentation of stimuli is central for the first empirical study of this thesis; thus, the different strategies are outlined in short.
Stimulus presentation strategies can be divided into three main strategies: block and event-related designs as well as mixed designs. In a block design, several stimuli of the same condition are presented in a row, and different blocks are presented in an alternating order. This leads to a BOLD response of high magnitude and enables the comparisons of the evoked states (e.g. task states and task-free states). In contrast, stimuli are presented separately in event- related paradigms and the individual haemodynamic response to stimuli is investigated. Mixed designs also rely on blocks of different conditions, yet stimuli are not presented continuously but as distinct events within a block (Amaro & Barker, 2006). Although the proportion of event- related studies is growing increasingly, block design fMRI is still a powerful tool with many areas of application (Huettel, Song & McCarty, 2014).
The first empirical study of this thesis relies on an analysis of task-related activity during a block design fMRI experiment to derive areas involved in the processing of dynamic fearful faces compared to dynamic landscapes. The resulting brain regions were forwarded into an analysis of FC with the aim to investigate the functional coupling of the identified relevant brain regions. Since the analysis of FC represents the core aspect of the first empirical study, the background and application of FC is subsequently presented in more detail.
3.1.2 Functional connectivity
Research on brain activity during tasks has greatly advanced our understanding of human brain functions regarding the involvement of distinct brain regions in tasks. Conceptual as well as methodological advances in the last two decades of neuroimaging have led to the shift from functional segregation to the new paradigm of functional integration (Friston, 2011). Adopting the perspective of interconnected large-scale brain networks instead of focusing on single areas offers a new access to investigate the functional architecture of the brain (van den Heuvel & Hulshoff Pol, 2010).
Interest in the functional coupling of brain regions took off after the seminal discovery of Biswal and colleagues (1995), who were the first to demonstrate that seemingly spontaneous low-frequency (< 0.1 Hz) fluctuations in the BOLD signal of the human motor cortices showed in fact highly coherent time courses of activity (see figure 6, panel A). Although participants lay still while their brain activity was recorded, a significant correlation was observed between the left and right motor cortex, yet not between motor cortices and visual areas. Notably, this correlation occurred at rest and was not triggered by an external event. This led to the assumption that this functional coupling reflects ongoing information processing within a functional neural circuit (Biswal et al., 1995). Ever since, FC was defined as the statistical
dependency among neurophysiological events in spatially distributed brain regions (Biswal et al., 1995; Friston, Frith, Liddle, & Frackowiak, 1993; Friston, 2011).
Figure 6. Origin of research on functional connectivity and currently discussed intrinsic
functional networks. A Exemplary visualisation of Biswal and colleagues (1995) results of correlation between the BOLD time series of the left and right motor cortices but independence of BOLD time series of visual and motor cortices. Figure adapted from van Dijk et al. (2010); MOT, primary motor cortex; VIS, primary visual cortex; L, left; R, right. B Intrinsic functional networks observed in task-free states. Figure adapted from Buckner et al. (2013).
Building on the initial work of Biswal et al. (1995), a vast number of studies have replicated this first finding and have further uncovered numerous networks that show synchronous activity over time (Van Dijk et al., 2010). Large-scale networks overlapping with central domains of brain functioning such as language, vision, auditory processing or attention as well as the so-called default mode network (Greicius, Krasnow, Reiss, & Menon, 2003) have been documented consistently (Buckner et al., 2013; see figure 6, panel B). Early studies investigated FC while participants did not engage in any task, therefore these functional networks were termed resting state networks (Damoiseaux et al., 2006). However, this term can be misleading for two reasons. First, the existence of correlated activity in many functional networks already indicates that even in a presumed state of rest, the brain is not idle (van den Heuvel & Hulshoff Pol, 2010). Additionally, only little differences in metabolic demands of
approximately 5% were observed between task states and task-free states (Raichle & Mintun, 2006). Secondly, these networks cannot only be identified during task-free periods but are also present when cognitive tasks are performed (Gonzalez-Castillo & Bandettini, 2017). Studies comparing functional networks at rest and during task states have yielded highly overlapping results with only subtle differences in the observed functional networks (Cole, Bassett, Power, Braver, & Petersen, 2014; S. M. Smith et al., 2009). Furthermore, functional networks can also be observed under different states of consciousness such as sleep (Larson-Prior et al., 2009) or anaesthesia (Hutchison, Womelsdorf, Gati, Everling, & Menon, 2013). These findings led to the conclusion that the observed networks reflect the intrinsic functional architecture of the brain into functionally different yet interconnected networks, and therefore the term resting state was abandoned in favour of intrinsic FC within intrinsic functional networks (Buckner et al., 2013).
Various methodological approaches have been used to investigate intrinsic functional networks. While an extensive discussion of the different methods goes beyond the scope of this thesis, two main approaches are outlined briefly (see Lee, Smyser, & Shimony, 2013 and van den Heuvel & Hulshoff Pol, 2010 for an overview)3
. Frequently, model-dependent methods have been applied to analyse FC of one specific region of interest (ROI, also called seed) with either the whole brain (seed-to-voxel FC) or previously defined target regions (e.g. ROI-to-ROI analysis). By correlating the time course of the seed with further ROIs or all brain voxels, FC of the seed area can be explored in a straightforward fashion. However, this approach depends strongly on the a-prior selection of the ROIs. Alternatively, model-free tools such as the Independent Component Analysis (ICA) allow data-driven extraction of whole-brain functional networks. In short, ICA algorithms extract statistically independent components from the fMRI data in order to explain the fluctuations of activity over time. The extracted components not only depict the different intrinsic functional networks, they can further be used to remove physiological noise from the data since noise is also subsumed in components. Despite the differences between the approaches, they yield highly comparable results and produce largely overlapping intrinsic functional networks (van den Heuvel & Hulshoff Pol, 2010).
The explanatory power of FC analyses largely depend on the appropriate handling of confounds and noise. Major sources of noise arise from the participants themselves, such as respiration, cardiac activity and head movement (Chang & Glover, 2009). Most importantly,
3
Aside from FC, analyses of effective connectivity offer a valuable, complementary approach in the analysis of functional networks. Effective connectivity among brain regions is investigated in model-based
cyclical physiological processes can induce spurious correlation in the data, and need to be controlled for carefully (van Dijk et al., 2010). Various methods for corrections such as global signal regression, regression of the time courses of white matter masks and masks of cerebrospinal fluid or component based methods (e.g. CompCor, Behzadi, Restom, Liau, & Liu, 2007, see van Dijk et al., 2010 for regression-based methods) have been introduced for this matter. A further concern is the reliability of analyses of FC, which has not been determined conclusively. Studies on intra- and interindividual reliability vary from low to moderate reliability especially for long-distance connections (Honey et al., 2009) to high reliability for the strongest and positive intrinsic FC but weaker for negative intrinsic FC (Shehzad et al., 2009). Thorough de-noising of the data and appropriate choice of pre-processing steps are thus of utmost importance in order to yield reliable estimates of intrinsic functional networks (Van Dijk et al., 2010).
In sum, spontaneous fluctuations of neural activity offer a window into the intrinsic functional architecture of the brain. Analyses of intrinsic FC can be a powerful tool to gain information about the functional circuits of the brain, yet require careful pre-processing and thoughtful choice of methodological approach.
In the first empirical study of this thesis, FC was investigated among regions involved in the processing of facial emotional expression. Although fMRI data was recorded during a task involving the presentation of dynamic fearful faces and dynamic landscapes in a block design, regressing out task-related activity allowed us to investigate intrinsic FC (Gonzalez- Castillo & Bandettini, 2017). Task states exert a rather small yet not negligible influence on intrinsic functional networks (Cole et al., 2014), thus the findings of this study should only be interpreted in the context of emotional face processing. However, our approach offers an opportunity to make best use of fMRI data from patients, which are usually a scarce yet highly informative resource.