Functional Neuroimaging: Fundamental Principles and Clinical Applications

SUMMARY

Functional imaging modalities, such as functional magnetic resonance imaging (fMRI) and diffusion tensor imaging (DTI), are rapidly changing the scope and practice of neuroradiology. While these modalities have long been used in research, they are increasingly being used in clinical practice to enable reliable identification of eloquent cortex and white matter tracts in order to guide treatment planning and to serve as a diagnostic supplement when traditional imaging fails. An understanding of the scientific principles underlying fMRI and DTI is necessary in current radiological practice. fMRI relies on a compensatory hemodynamic response seen in cortical activation and the intrinsic discrepant magnetic properties of deoxy- and oxyhemoglobin. Neuronal activity can be indirectly visualized based on a hemodynamic response, termed neurovascular coupling. fMRI demonstrates utility in identifying areas of cortical activation (i.e., task-based activation) and in discerning areas of neuronal connectivity when used during the resting state, termed resting state fMRI. While fMRI is limited to visualization of gray matter, DTI permits visualization of white matter tracts through diffusion restriction along different axes. We will discuss the physical, statistical and physiological principles underlying these functional imaging modalities and explore new promising clinical applications.

fMRI – Physiologic Principles

Functional MRI (fMRI) allows identification of eloquent areas of the brain for use in treatment planning.¹ The most well developed form of fMRI relies on Blood-Oxygen-Level Dependent, or BOLD, contrast. This is not “contrast” in the traditional Gadolinium sense; rather, BOLD contrast differentiates tissues based on the intrinsic discrepant magnetic properties of oxy- and deoxyhemoglobin. Oxyhemoglobin is a diamagnetic substance, thus repelled by a magnetic field. In contradistinction, deoxyhemoglobin is paramagnetic and tends to align with the static magnetic field supplied by the MRI magnet.^2–4 As neuronal activity increases with a specific task, there is a compensatory increase in cerebral blood flow to that area and oxygen metabolism to the regional tissues. This physiologic response to neuronal activity was initially established by Sherrington et al. in 1890 and is fundamental to all functional imaging that relies on cerebral hemodynamics.⁵ Generally, the increase in blood flow (and oxyhemoglobin) is greater than the metabolic rate of oxygen utilization by the tissue producing a relatively high oxyhemoglobin:deoxyhemoglobin ratio.⁶ The changes in this ratio can be measured utilizing a T2*-weighted sequence, which is very sensitive in detecting alterations in the magnetic field. The brain must be imaged in a dynamic fashion to assess the changes that occur in signal over time. Spatial resolution is sacrificed for temporal resolution, and an ultra-fast sequence such as Echo Planar Imaging (EPI) allows for a whole brain acquisition on the order of two seconds.

Using identical physical principles, resting state fMRI (rs-fMRI) is a relatively novel imaging technique with great promise in discovering neuronal networks. In contrast to task-based fMRI, rs-fMRI relies on detection of low frequency fluctuations in BOLD signal (<0.1 Hz) in the absence of active tasks.⁷ The aim is to identify synchronization between different regions that suggest neuronal connectivity. Connectivity is implied when different areas exhibit similar temporal patterns of activity.⁸ Resting state networks are generally divided into task-positive, which have enhanced connectivity during active tasks (i.e. language, auditory, etc.), and task-negative networks, which are more active during rest (i.e. the default mode network).⁹

fMRI – Data Acquisition and Processing

Data acquisition for BOLD fMRI in the identification of eloquent cortex relies on task-based activation of cortical tissue as part of experiments termed paradigms. There are numerous paradigms that can be used depending on the location of the lesion. These include motor tasks, expressive and receptive language tasks, hearing tasks and visual tasks. Often more than one paradigm will be required for reliable identification of eloquent cortex.¹⁰ As discussed earlier, the hemodynamic response to the activation of corresponding cortical regions can then be visualized. We will discuss a typical paradigm used for the identification of cortical motor tissue.

Neuronal activity results in increased cerebral blood flow, hemoglobin oxygenation, and relative T2* signal. The hemodynamic response curve plots the T2* signal (y axis) in relation to time (x axis). Figure 1 depicts a typical HRC. The hemodynamic response, and thus T2* weighted signal, lags behind neural activation. While neural activation occurs on the order of milliseconds, the corresponding hemodynamic response occurs on the order of seconds.¹¹ The HRC dictates the temporal structure of the motor paradigm. As the T2* weighted signal change caused by oxygenated blood is only on the order of 2%, many trials of activation and rest have to be performed for a single task in order to allow robust detection of activated cortical tissue.¹²

Figure 1.

The hemodynamic response curve, shown above, demonstrates the temporal lag of T2* signal (generated by increased Hb/dHb ratio due to increased CBF) relative to neuronal activity.

A block design paradigm is the most commonly used (Figure 2). A typical block design may comprise 20 seconds of activation (i.e. finger tapping) followed by 20 seconds of rest, which may be repeated up to eight times. A total study time of 320 seconds would allow for 160 separate volumetric brain acquisitions, and the T2* weighted signal for each voxel can then be plotted over time. Voxels activated by the prescribed task exhibit an MRI signal that coincides with the expected shape of the hemodynamic response curve. Uninvolved regions exhibit the random fluctuations of background noise. The strength of correlation between the fMRI data set and the expected hemodynamic response curve is quantified mathematically by the correlation coefficient (CC). Voxels that demonstrate a large CC are displayed as regions of positive activity on the fMRI examination. It should be noted that an fMRI examination is a statistical map of activity and that the examination may be displayed with multiple different CC thresholds (Figure 3). Accurate determination of functional tissue is dependent on a correlation coefficient threshold that captures neuronal activity representative of task performance, thus post-processing requires some experience. A threshold that is set too low will include too much background activity, while one that is too stringent will fail to elicit enough activations to reliably identify neural activity. The post-processed BOLD signal is then superimposed on a separately acquired volumetric anatomic sequence such as post-contrast T1, allowing localization of the activated tissue.

Figure 2.

A typical paradigm for identification of the primary motor cortex. Alternating finger tapping is performed for 20 seconds, during which 10 T2* acquisitions are acquired. This is followed by 20 seconds of rest and 10 additional T2* acquisitions. This is repeated as necessary to generate reliable data.

Figure 3.

Determining the statistical threshold is an essential part of post processing. Setting the statistical threshold too high obscures eloquent cortex, while setting it too low results in excessive background noise and limits specificity.

Resting state fMRI paradigms differ in both data acquisition and analysis. Acquisition is performed in the resting state, without a required stimulus or task, to identify resting state networks. These are networks of related parenchymal tissue that may be involved in either attention-based networks or the default mode network but are active in the absence of performing an active task.¹³ The default mode network is an example of a network of neural tissue that functions in the passive state but demonstrates enormous connectivity with different neural systems and is one of the most evolutionarily conserved neural networks in mammals.¹⁴ The default mode network was first identified using fMRI.^15–17 Multiple methods of analysis can be employed which generally fall into two categories: seed-based (also known as model-based) and data-driven. Seed-based analysis relies on a priori placement of regions of interest (ROIs), or seeds, and comparing data between ROIs. This method offers a practical means to demonstrate functional connectivity to a specific voxel or ROI.^18,19 The limitation of this method is that the interpretation and power of the study is influenced to a large degree by placement of the ROIs. It obviously also offers limited utility in the determination of connectivity outside the ROIs. More recently, rs-fMRI has tended towards data-driven analyses. There are numerous methods of data-driven analysis. One of the more popular forms is independent component analysis (ICA). ICA does not require a priori selection of ROIs, rather it relies on a temporal map of the entire brain based on independent components. Temporal patterns of activity are compared between the independent components to assess for correlations that suggest connectivity.^16,20 There are multiple other data-driven forms of analysis that allow for parcellation of the brain, including clustering analysis,²¹ k-means²² and edge-finding methods.²³ While increasing in popularity, data-driven analysis of fMRI data presents potential obstacles. Background physiologic activity in the form of respiratory or cardiovascular noise may generate a signal that suggests spurious neuronal connectivity, generating false positive correlations. Specific software packages are often required to mitigate the confounding signal and increase the specificity of temporospatial correlations.²⁴

fMRI – Clinical Applications and Limitations

Task-based fMRI has demonstrated significant value in the pre-surgical evaluation of patients undergoing neurosurgical treatment where preservation of function is essential. Localization of eloquent cortex such as the visual cortex or primary motor cortex can be particularly helpful in the setting of pathologic processes that distort native anatomy which limits the use of traditional anatomic landmarks in its identification (Figure 4). fMRI is also a powerful, non-invasive tool both in determining hemispheric language dominance, which is critical in patients undergoing frontal or temporal lobe resections, and in accurately locating major receptive and expressive language areas which tend to be quite variable in anatomic location.^25,26 Pre-operative localization of these regions and their anatomical relationship with the intended surgical target may alter the surgical approach or the extent of the resection. Ultimately, its use has led to a decrease in the risk of post-operative deficits and an improvement in clinical outcomes, albeit in a specific patient population.²⁷ Presurgical planning using task-based fMRI necessitates patient cooperation in performing the tasks during the specific paradigms. Alteration in mental status, seizure-like activity, patient age, sedation or limited cognitive function at baseline may limit or preclude accurate identification of eloquent cortex, severely limiting the generalizability of fMRI in neurosurgical planning. Recent evidence suggests that rs-fMRI may be useful in identifying eloquent cortex in patients who are unable to participate in task-based paradigms.^28–30 Specifically through use of a seed-based analysis, Zhang et al. demonstrated that ROI placement on the spared motor cortex allowed reliable identification of the contralateral distorted motor cortex for neurosurgical planning.³¹

Figure 4.

The primary motor cortex may be readily identifiable in normal patients, however its location can be easily obscured by malignancy and other pathologic processes. Here, a large glioma distorts the right cerebral hemisphere and limits reliable determination of the primary motor cortex. Utilization of a bilateral finger tapping motor allows identification of functional motor cortex.

In the case of mesial temporal lobe epilepsy, recent resting state fMRI studies demonstrate reproducible differences in basal functional connectivity within the temporal lobes. The data suggest an asymmetric disruption of certain neuronal networks^32–35 and an overall decrease in basal functional connectivity within the epileptogenic temporal lobe and increased basal functional connectivity within the contralateral temporal lobe.^36–38 Indirect localization of the epileptogenic focus via determination of increased basal functional connectivity in the non-epileptogenic temporal lobe appears to be the most accurate measure of localization of the epileptogenic zone, with a specificity of up to 91% demonstrated by Bettus et al. Importantly, the laterality of increased basal functional connectivity within the non-epileptogenic temporal lobe was evident in patients without evidence of structural insult on MRI, suggesting a unique role of rs-fMRI in localization of the epileptogenic zone in cases of drug resistant epilepsy for pre-surgical planning.³⁹

Several functional connectivity studies have demonstrated characteristic patterns of cortical activations and deactivations in different forms of dementia, suggesting diagnostic value of resting state fMRI in its diagnosis.^40–45 Specifically, Alzheimer’s dementia consistently demonstrates decreased resting state activity within the default mode network, even early in its clinical course.⁴³ More extensive functional connectivity studies have also demonstrated decreased functional connectivity within the medial temporal lobe and posterior cingulate cortex, as well as increased activity within the salience network (a neural network within the anterior insular cortex and anterior cingulate cortex involved in monitoring external inputs and internal brain activity) compared to healthy controls.⁴⁶ In contradistinction, frontotemporal dementia may exhibit increased connectivity within the default network and decreased connectivity within the salient network, allowing for reliable diagnosis in cases where there may be clinical ambiguity.⁴¹

Since its inception, rs-fMRI has primarily been used in research to identify resting state networks and determine functional connectivity. Clinical use of rs-fMRI is still in its infancy and it has not yet achieved the widespread acceptance of task-based fMRI.⁴⁷ As previously discussed, several recent studies have demonstrated promising diagnostic applications of rs-fMRI in a variety of pathologies. Resting state fMRI also offers the unique benefit of interrogating neuronal activity in an uncooperative or sedated patient. With further clinical-based research and development of processing tools, the role of rs-fMRI in the clinical realm is destined to become more well-established.

DTI – Physiologic Principles

Evaluation of white matter has been revolutionized by diffusion tensor imaging (DTI). DTI allows interrogation of the brain and visualization of neuronal tracts using the same principles as diffusion weighted imaging–the Brownian motion of water molecules. Complex cellular membranes affect the diffusion of water molecules; hence the diffusion profile is reflective of the underlying tissue microstructure.⁴⁸ Collecting diffusion measurements in at least six non-collinear directions allows for a directional understanding of diffusion and the resultant data can be displayed as a three-dimensional ellipsoid known as the “diffusion tensor.” When diffusion is heavily skewed in a particular direction, or anisotropic, the resultant tensor becomes elongated and exhibits a specific orientation.^49,50 Within the highly organized white matter tracts there is relatively free diffusion along the axis of the tract, while diffusion perpendicular to the tracts tends to be limited by the cell membranes and myelin sheaths.⁴⁹ The clinical applications for DTI are varied and far-reaching throughout neuroradiology. One of the most visually appealing applications is tractography, or the ability to model major white matter tracts within the brain.⁵¹

DTI – Data Acquisition and Processing

DTI measures the three-dimensional diffusion profile of water molecules. Detection of the direction of preferential diffusion depends on applying different diffusion-weighted gradients along the x, y, and z axes and measuring the corresponding diffusion coefficients.⁵² Diffusion measurements along a minimum of six non-collinear directions [e.g. (x,y,z) = (1,0,0), (1,1,0), (0,1,0), (0,1,1), (1,0,1), (0,0,1)] are needed to derive the tensor matrix, D, which has six components that represent diffusion coefficient along different directions: D_xx, D_xy, D_yy, D_yz, D_xz, D_zz. The resultant tensor takes the shape of an ellipsoid with its three principle axes aligning with the tensor eigenvectors (v₁, v₂, v₃), each with its own radius or eigenvalue (λ₁, λ₂, λ₃) representing the diffusivity along the three axes (Figure 5). The eigenvalues are typically sorted such that λ₁ > λ₂ > λ_3, with the preferential diffusion direction aligning with v₁. As the number of diffusion measurements increases, the accuracy of the tensor data improves. Generally, in the clinical setting, 30 direction measurements uniformly oriented around a sphere are acquired^53–55. Fractional anisotropy (FA) is a quantitative scalar measurement indicating the extent to which the diffusivity differs from the expected spheroid distribution (a Gaussian diffusion profile). A FA value of 0 implies a spheroid distribution, whereas 1 implies maximum anisotropy.⁵⁶ The FA value does not contain any information regarding the direction of anisotropy. Post processing software will assign a color for the orientation of the white matter tract, with the intensity of the color signifying the FA value of the voxel (Figure 6).

Figure 5.

Diffusion tensor imaging uses diffusion measurements in at least 6 non-collinear directions to determine a “diffusion ellipsoid” for each voxel. The ellipsoid is described by 3 eigenvectors (v1,v2, v3) each with its own eigenvalue (λ1, λ2, λ3). This information can be used to determine the extent and orientation of diffusion.

Figure 6.

Post processing software assigns colors to the orientation of the white matter tracts. Traditionally, and in this case, blue = cranio-caudad, red = transverse, green = anterior-posterior. The intensity of the color is directly proportional to the FA value. In this case the DTI data is co-registered to a 3D T1-weighted image.

DTI – Clinical Applications and Limitations

DTI has dramatically changed the ability to probe and visualize white matter. Numerous pathologic processes, including inflammation, edema, tumor, gliosis and demyelination all demonstrate increased signal intensity on T2 weighted imaging, whereas DTI can offer quantitative data regarding the extent white matter involvement and compromise (in the form of ADC and FA values). Among the clinical uses for DTI is its diagnostic utility in characterizing multiple sclerosis (MS). Conventionally, T2-weighted/FLAIR MRI has been used to diagnosis and monitor disease progression in MS. While this imaging protocol remains an irreplaceable diagnostic tool, it offers limited histopathologic insight into the disease process. Normal appearing white matter on MRI may demonstrate MS-related white matter disease on the cellular level. Evaluation with advanced diffusion metrics can detect local alterations in diffusion anisotropy, identifying foci of white matter disease which fall below the soft tissue resolution of MRI.^57–66

DTI has also shown value in diagnosing epilepsy. While conventional MR imaging is a critical component of the diagnostic protocol in the work-up of epilepsy, some of the more typical findings on conventional MR (i.e. hippocampal volume loss and T2 hyperintensity) may be too subtle for reliable diagnosis.⁶⁷ DTI in the evaluation of epilepsy relies on the distortion of cytoarchitecture from epileptogenic activity which causes increased diffusivity and decreased anisotropy.^68–70 Distortion of cytoarchitecture is also seen in neoplastic disease, which can be detected by DTI.^71–78 Determination of white matter tract involvement in this setting can be critical in neurosurgical planning.⁷⁹ The 3D modeling of white matter tracts based on DTI data, known as tractography, has become increasingly utilized in the clinical arena (Figure 7).

Figure 7.

A case of a high-grade glioma involving the right cerebral hemisphere. On the left, a T1 weighted post contrast image demonstrates the tumor but with limited evaluation of involvement of underlying white matter tracts. The middle image demonstrates DTI representation of the white matter tracts. On the right, a 3D representation shows sparing of the right corticospinal tract which drapes over the anterosuperior aspect of the tumor.

While DTI has transformed the visualization of white matter, the diffusion metrics used are based on a Gaussian diffusion model, which assume that the Brownian motion of water molecules will demonstrate a normal spatial distribution. This is not an entirely accurate model of the diffusion profile as several experiments have demonstrated non-Gaussian diffusion characteristics within neurologic tissues.^80–82 The resolution of DTI is also limited in that each voxel in a typical DTI examination may measure 2 × 2 × 2 mm, for which a single diffusion tensor is generated. On the cellular level there are tens of thousands of axons and glial cells within the voxel region which likely comprise multiple white matter tracts. The data generated are ultimately an average of the diffusion characteristics of that voxel and does not offer enough spatial resolution to reliably identify multiple axonal tracts within a single voxel.^83–85 Additionally, in the evaluation of normal appearing white matter on MRI in patients with known white matter pathology (i.e. multiple sclerosis), local aberrations in diffusion metrics may be so subtle as to only be appreciable in group analysis, limiting diffusion-based white matter interrogation on an individual basis. In the case of abnormal appearing white matter, the utility of DTI is primarily in the identification and spatial characterization of involved white matter tracts in cases of anatomic distortion.⁸⁶

Conclusion

Functional MRI and DTI have drastically changed the ability to interrogate and visualize the brain. Functional modalities have offered insight into the inner workings and connectivity of the healthy brain and have summoned a new age of brain-mapping that offers great promise. Since their widespread clinical acceptance, there has also been a steady flow of new uses and data that further our understanding of neuropathology. The full potential of functional imaging has not yet been realized as more pre- and post- data processing techniques are continually being developed. A basic understanding of the more common functional imaging modalities is necessary as they have and will continue to shape our understanding of neuropathology and interconnectivity.

Funding

This research received no specific grant from any funding agency in the public, commercial or not-for-profit sectors.

Conflict of interest

The authors declare no conflict of interest.