The field of functional neuroimaging is evolving rapidly. Technological advances push the envelope of temporal and spatial resolution, providing windows into the dynamic workings of the brain. The technologies have become more complex, the experimental designs more elaborate, and the data analytic strategies more sophisticated. This primer is intended to introduce the reader to selected topics in functional neuroimaging and to aid understanding of the literature.
Functional neuroimaging refers to a set of procedures that measure the consequences of neuronal activity. Transmembrane currents (axonal action potentials and synaptic and dendritic currents) cause changes in electric and magnetic fields. These can be measured from the scalp surface via electroencephalography (EEG) or magnetoencephalography (MEG). These currents have energy-consuming biochemical causes and effects (e.g., neurotransmitter release, activation of second messengers, and ion pump activity). Changes in local neuronal activity cause local changes in blood flow and glucose metabolism, which are measured by positron emission tomography (PET), and in blood oxygenation levels, which are measured by functional magnetic resonance imaging (fMRI).
These procedures differ in their spatial and temporal resolution, cost, availability, and radiation exposure (see table on this page). All produce a three-dimensional matrix of numbers — an image. Each location in the matrix corresponds to a chunk of brain tissue — a voxel. The numbers in the matrix correlate with the physiologic variable of interest. The image can be manipulated mathematically, or it can be displayed graphically by assigning a color to each number. Even the highest-resolution scanners have voxels billions of times larger than the physiologic compartments of interest. A reliable signal will only be detected if a large number of synapses in a voxel are doing the same thing at the same time. Thus, most of what is happening is missed. When a signal is detected, its mechanistic relevance is often unclear. Nonetheless, by asking the right question, one can occasionally associate a fairly focal brain region with a fairly specific mental process in a meaningful fashion.
In positron emission tomography, radioactive tracers are injected intravenously. These travel to the brain and emit positrons. The PET camera detects the pairs of photons emitted when the positrons annihilate with electrons. Different tracers may be used to index physiological processes (blood flow and metabolism) or biochemical processes (neurotransmitter receptors and transporters). Radioactively labeled water is used to measure cerebral blood flow (CBF). A radioactive analogue of glucose is used to index local cerebral metabolic rate (CMR). A variety of chemicals that bind to receptors are available to image specific neurotransmitter receptor subtypes.
PET captures dynamic changes in brain activity, but at a lower resolution than MRI, making the discrimination of small nuclei difficult (see table). If arterial blood is tested for radioactivity during the procedure, absolute quantification of CBF or CMR is possible. Without quantification, the numbers in the image have only a relative meaning; this limits comparisons among individual subjects.
Single photon emission computed tomography (SPECT) is similar to PET but detects only a single photon per disintegration. This increases scatter and lowers resolution. SPECT is less expensive and more available than PET.
Functional magnetic resonance imaging measures indirect effects of neuronal activation on the flow, volume, and oxygenation of blood. The spatial and temporal resolution of fMRI are superior to those of PET and SPECT, but artifact due to head motion and signal-to-noise ratios present challenges to data analysis. fMRI images are generated from signal differences between activated and nonactivated scans that are alternated in time.
Magnetic resonance spectroscopy (MRS) permits a non-invasive investigation of brain biochemistry. While conventional MRI focuses on protons, MRS can measure a spectrum of other biochemicals that index energy metabolism, membrane turnover, and loss of neurons. These spectra yield information about neurochemistry and neurotransmission that would otherwise only be available via the invasive technique of microdialysis.
The brain activity we image depends on what the brain is doing at the time we 'look through the window.“ Thus experimental design must pay close attention to the subject's mental and emotional state during imaging. One typical design is to acquire images under both ”test“ and ”resting“ conditions. The resting image is subtracted from the test image to generate a ”subtraction image.“ If all factors besides the stimulus conditions are perfectly controlled (l), then the subtraction image reveals which brain regions were ”activated“ (or deactivated) by the test condition. If a third condition is added that only differs from the test condition in one specific way, subtracting this third image from the test image may reveal brain regions involved in performing a particular aspect of a task. Factors that may confound subtraction paradigms include variable compliance with the task, differences in cognitive strategies used to perform the task, and the inability to completely control mental factors irrelevant to the task.
Multiple subjects are needed to form a general conclusion about functional neuroanatomy. These images must be registered with each other before they can be compared. One
Sarah Hollingsworth Lisanby, M.D., is a research psychiatrist at the New York State Psychiatric Institute, a research fellow in the Affective and Anxiety Disorders training program at Columbia University, and a former Fellow of The American. Brett Mensh, M.D., Ph.D., is an assistant clinical professor of psychiatry at Columbia University.
method of handling individual differences in brain anatomy is to transform each image into a standard stereotaxic space (e.g., Talairach atlas), and then use statistical parametric mapping (SPM) to perform a t-test at each voxel, followed by a correction for multiple comparisons.
In the region of interest (ROI) approach, each image is co-registered to the subject's MRI scan and boundaries are drawn around specific brain areas. Activity within each ROI is then compared across groups. The ROI approach increases statistical power by reducing the number of statistical comparisons. A pre-defined model of the neural systems mediating the brain function in question can aid this work. These models predict which brain areas serve as nodes within a functional network.
SPM and ROI analysis are typically univariate approaches in which regions are compared one at a time. However, we know that brain areas do not function in isolation. The cross talk between discrete brain regions is likely to mediate cross-modal mental associations. A variety of statistical approaches can explicate interactions among brain areas and identify functional brain networks that mediate aspects of behavior.
Activation studies reveal which areas of the brain correlate with performance of particular tasks, but they cannot prove that these areas are essential — or even involved — with the performance of the task. Connections between co-varying regions may be symmetric or asymmetric, mono- or polysynaptic, excitatory or inhibitory. Since the brain uses energy to actively inhibit functioning, both excitatory and inhibitory actions can increase CBF and CMR locally. No method can adjust for functional neuroanatomical variability among subjects. A basic limitation of neuroimaging is that it can only reliably locate mental functions whose location is the same in all subjects.
Repetitive transcranial magnetic stimulation (rTMS) is a non-invasive technology using rapidly alternating magnetic fields to induce current flow in the cerebral cortex. rTMS can reversibly excite or inhibit focal brain areas and may have clinical benefit in depression. Functional neuroimaging coupled with rTMS offers a unique paradigm to probe human brain connectivity in vivo. rTMS has the potential to address the causation/correlation question by directly altering activity in areas hypothesized to be involved in particular brain functions and then observing the behavioral effects of that intervention.
While there is good reason to be enthusiastic about functional imaging techniques, it is important to remember what imaging can and cannot tell us. Functional neuroimaging represents a collection of tools to study the brain. The quality of the information we obtain with these tools depends in large part on the sophistication of the questions we ask and the experiments we design to answer them.