|
Scientific Understanding of Consciousness |
fMRI Description
(paraphrase of Haxby, Functional Magnetic Resonance Imaging, 123ff) In the early 1990s fortuitous insights and advances in the physics of magnetic resonance imaging (MRI) led to the discovery that MRI could be used to measure hemodynamic processes noninvasively in the human brain. Unlike positron emission tomography (PET), new methods for functional MRI, (fMRI) required no exposure to ionizing radiation, no injections of tracers, and no sampling of blood. Technical difficulties imposed by working in a powerful ambient magnetic field have been overcome, and fMRI studies of the neural basis of human cognition have proliferated. fMRI is now a powerful tool for investigating cognitive processes. PHYSICAL BASIS OF MRI MRI uses the radio frequency (RF) electromagnetic waves emitted by the nuclei of hydrogen atoms with single-proton nuclei to construct detailed images of the brain and other organs. In an MRI scanner, the magnetic dipoles of individual protons become aligned with the strong magnetic field of the scanner. The direction of the magnetic dipoles is perturbed by an RF pulse generated by gradient coils in the scanner. After the perturbation, the protons wobble, or precess, back to their original alignment. The precession has a frequency, called the Larmor or resonance frequency, that is specific to the type of nucleus, in this case single protons, and for the strength of the magnetic field. For example, single protons in a 1.5 tesla magnetic field, a strength typical of most scanners used for fMRI, has a resonance frequency of 63.84 MHz. A detectable radio signal at the resonance frequency is generated by protons that are precessing in phase with each other. The receiving coils in the scanner detect the radio signal emitted by the tiny senders. Spatial information about the location of the protons is afforded by slightly altering the strength of the magnetic field over the volume being imaged, and thereby altering the frequency and phase of signals emitted by protons at different locations. Blood Oxygen Level Dependent fMRI Broadly defined, fMRI refers to numerous methods for obtaining information about hemodynamic processes. This review focuses on the subset of those methods that is currently used for most fMRI studies of human cognition. This subset relies on blood oxygenation level dependent (BOLD) changes in the intensity of the magnetic resonance (MR) signal. MR imaging of BOLD changes is based on a difference in the magnetic property of oxygenated versus deoxygenated hemoglobin. Changes in local neural activity cause a change in local blood flow. The result is an increase in blood oxygenation when blood flow increases and a decrease in blood oxygenation when blood flow decreases. A brightening of the image reflects an increase in blood oxygenation and neural activity. A darkening of the image reflects a decrease in blood oxygenation and neural activity. Fast Imaging Methods The final advance in MRI physics that made modern fMRI possible was the development of fast imaging techniques. The most widely used fast imaging method is echo-planar imaging (EPI), which allows the acquisition of a complete cross-sectional image from one excitation pulse. A second rapid imaging method, spiral imaging, uses a more efficient search of frequency space than EPI and will probably be used more with improvements in gradient coil design, which are necessary to take full advantage of it. With those fast imaging methods and current standard echo-speed gradients, it is possible to obtain multiple cross-sectional images every second. A volume of cross-sectional images that contains most or all of the brain can be obtained every 2 to 6 seconds, depending on the spatial resolution of the images, the method being used, and the performance characteristics of the scanner system. The Hemodynamic Response Function The signal change seen with fMRI typically lags behind the onset of stimulation or motor activity. The offset between sensory stimulation or motor activity and the associated change in neural activity is presumably on the order of tens or hundreds of milliseconds. The hemodynamic change measured by fMRI, on the other hand, does not reach its maximum for 4 to 8 seconds. The function relating a change in neural activity to a hemodynamic change is called the hemodynamic response function. Most attempts to describe the transfer function have been atheoretical searches for mathematical functions that best fit the observed change in MR signal intensity. Those mathematical descriptions generally assume that the change in neural activity is a square wave, representing an essentially instantaneous change in neural activity to a new steady state that is coincident with the sensory, cognitive, or motor change. The delayed and smoothed hemodynamic change elicited by the change in neural activity has been modeled as a linear ramp or as a nonlinear curve with the shape of half of a Poisson or a Gaussian distribution. A mathematical model of the hemodynamic response function is necessary for the analysis of fMRI data sets because it provides the basis for deriving the predicted fMRI time series so that the fit of the predicted response to the obtained response can be tested. FMRI EXPERIMENTAL DESIGN AND DATA ANALYSIS As compared to previous functional brain imaging methods, such as PET and other nuclear medicine procedures for measuring cerebral blood flow, fMRI offers the cognitive neuroscientist much greater freedom in experimental design. Unlike nuclear medicine procedures, which integrate activity over durations measured in tens of seconds, fMRI measures are virtually instantaneous, making the hemodynamic response the only factor that limits temporal resolution. A second advantage is the ability to obtain a virtually unlimited number of measures because fMRI is not limited by radiation dose restrictions. Consequently, enough data from one individual can be collected to perform massive signal averaging, increasing sensitivity and precision sufficiently to obtain detailed maps of responses in an individual brain. Removing the limit on data set size also allows the cognitive neuroscientist to test more experimental conditions in an individual subject and to test changes in neural response that occur over days, weeks, or months with various forms of learning. Fundamentals of Experimental Design An fMRI experiment consists of a series of images obtained over a period of time that typically lasts from one to 20 minutes. One such series of images shall be referred to as an fMRI time series. The length of the time series can be limited by scanner performance characteristics and scanner system memory capacity. Additional time series can be obtained, without removing the subject from the scanner, to increase the data set size and increase the signal-to-noise ratio. For every doubling of data set size, the standard error of the noise is reduced by Ö2. Consequently, obtaining four time series instead of one doubles the sensitivity of the experiment and affords detection of signal changes that are one half the magnitude of changes detected with a single series. Likewise, obtaining eight time series affords detection of signal changes that are 35% (1/Ö8) the size of changes detected with a single series. Contrasts between conditions are best made within time series to minimize the confounding of BOLD changes with intensity changes due to head movement. Head movement is a major source of data degradation in fMRI experiments. A small head movement can cause a large change in signal intensity in an image volume element that is unrelated to blood oxygenation and neural activity. This signal intensity change is due to the change in the type of tissue or substance (gray matter, white matter, or cerebrospinal fluid) contained in that volume element. Signal intensities for those tissue types vary more than the differences caused by changes in blood oxygenation. Misalignment of scans caused by between-scan head movements can be partially corrected with software, but such corrections can be inadequate for correcting large movements. Experimental Designs Based on Between-Task Contrasts: Subtraction and Parametric Variation Most functional brain-imaging research is based on the assumption that the pattern of activity measured during the performance of a task reflects all of the sensory, cognitive, and motor components of that task. The functional image is seen as a measure of the total integrated neural activity associated with the complete task. That assumption makes sense for nuclear medicine procedures for measuring hemodynamics that integrate activity over periods measured in tens of seconds. It is not a necessary assumption for fMRI, however, as will be made clear in the next section. Under that assumption, the isolation of the activity related to a single cognitive component of a task requires the comparison of tasks that are matched on all other components except the one of interest. If the comparison is between a pair of tasks, then the comparison is a subtraction. If the comparison is across a series of tasks that systematically vary the component of interest in a graded fashion, then the comparison is a correlation. Task subtraction is by far the most common experimental design in functional brain imaging research. In attention studies, tasks can be contrasted that involve attention to different stimulus locations (e.g., right or left field) or attributes (e.g., motion or color). Although used less often, parametric variation is a potentially more powerful experimental design because it has the potential to reveal the quantitative relationship between a cognitive parameter, such as attention or difficulty, and neural activity. In a typical fMRI experiment using between-task contrasts, multiple scans are obtained over a period of 10 to 60 seconds while a subject performs each task. Each task block, therefore, has to consist of a homogeneous set of trials. (end of paraphrase)
Return to — Brain Imaging |