Imaging: From Plain X-ray to MRI
In 1901, Wilhelm Roentgen was awarded the first Nobel Prize in medicine for the discovery of X-rays. More than a 100 years later, in 2003, Paul Lauterbauer and Sir Peter Mansfield were awarded the Nobel Prize jointly for their work on magnetic resonance imaging (MRI). The 20th century saw an explosion of knowledge in medical imaging technology. The journey from plain X-ray to MRI is a fascinating story: initially, structural imaging improved in spatial resolution, and then, in the 1990s, functional imaging matured, adding a totally new dimension to imaging.
Early on, X-rays could image only bony structures (and regions with density similar to bony structures), but this by itself was a major revolution in diagnostic techniques in medicine. Subsequently, when contrast dyes (such as barium) became available, many visceral organs could also be imaged. Ultrasound imaging brought in another level of improvement in imaging technology. Ultrasound has many advantages. It is noninvasive, it has no associated radiation hazards, and the ultrasound device is portable (the imaging can be done at the patient’s bedside if required). Ultrasound devices have been developed for specific purposes: cardiac imaging (echocardiography); imaging of peripheral vasculature, brain, and abdominal organs; and guidance for nerve blocks.
However, the human brain was considered an imaging black box, and technology for imaging of the brain took a longer time to develop. For almost 75 years, cerebral angiography was the only technique available for imaging the interior of the brain. Visualization of internal structures of the brain using cerebral angiography was far inferior to the imaging available with present-day technology. Furthermore, techniques like ventriculography and pneumoencephalography were good for visualizing structures around or inside the ventricular system of the brain, but they were only marginally superior to, or complementary to, cerebral angiography.
Computerized axial tomography (which today is called simply CT) was invented in the 1970s by Sir Godfrey Hounsfield (he was awarded the Nobel Prize in 1979 for his invention). Hounsfield envisioned CT as an imaging technique based on the differential absorption of X-ray energy by the various structures being imaged, and CT scanners were made possible because of the coinciding rapid development in microprocessors and computing technology. CT was a major revolution in the imaging of the interior of the brain. For the first time, interior structures of the brain could be imaged in detail noninvasively and in a short period of time. With further development in computing technology, next-generation CT scanners were faster and their image resolution was far superior to that of early-generation CT scanners. CT soon became the standard workhorse modality for diagnosis in clinical neuroscience and it continues to be so today in emergency medicine. CT scan resolution has improved significantly because of continuing advances in computer technology and the use of multiple processors.
MRI was developed in the 1980s. MRI has imaging capabilities far superior to those of CT (although imaging time is longer) and the technology applied in MRI had never been used before in medical science. MRI is based on the magnetic resonance property of protons (H+ ions) and it exploits the ubiquitous presence of protons in the human body, including the brain. MRI has the additional advantage of being free of radiation hazards (although it entails safety issues related to the use of an ultra-high-power magnetic field). The spatial resolution of MR images is far superior to that of CT images. Developments in computer technology contributed significantly to MRI in data collection, storage, analysis, and the creation of real-time images. Allan Cormack, who shared the Nobel Prize with Godfrey Hounsfield, derived the mathematical calculation applicable in CT for development and display of reconstructed images. This part of image reconstruction was applicable to MRI as well.
Functional Imaging Before Functional MRI
The primary inspiration for functional MRI (fMRI) came from functional imaging studies done with positron emission tomography (PET). In 1978, Lassen captured cerebral blood flow (CBF) activation images of the human brain using the technique developed by Kety and Schmidt that utilized intracarotid injection of xenon 133 for measuring CBF (Figure 1.1). Lassen published the first images of brain activation (during a word-processing task) in Scientific American in 1978.1 Earlier, Sokoloff had laid the foundation for PET with the introduction of autoradiographic measurement of cerebral metabolism using fluorodeoxyglucose (FDG) injection in rats.2 Peterson used PET to image CBF in humans during a word-processing task by injecting 15O-labeled water.3 With PET, CBF and cerebral metabolic rate of oxygen (CMRO2) could be measured by injecting isotope-tagged tracers. The isotope emitted positrons in the target organ (brain) and the positrons could be imaged. Thus, changes in CBF and CMRO2 could be measured quantitatively with PET and could be assessed as a reflection of neuronal activity in the brain. For the first time in history, functional brain activity during simple tasks (such as visual, auditory, and motor activation) could be visualized in the respective regions of the brain and recorded. The advent of CT and MRI allowed recording of high-resolution anatomical images of the brain, and then functional activity on PET scans could be overlaid on the anatomical images. Imaging of functional activation of the brain and overlaying the images on a high-resolution anatomical image soon became the standard technique for functional imaging of the brain.
From MRI to fMRI
The progress from MRI to fMRI was surprisingly rapid. MRI was approved by the U.S. Food and Drug Administration (FDA) in 1985, and the first fMRI study was presented at the Society of Magnetic Resonance Medicine meeting in 1991 by Jack Belliveau from Massachusetts General Hospital in Boston. Some of the landmark discoveries that led to fMRI are:
1931—Linus Pauling demonstrated that oxygenation altered the magnetic susceptibility of the hemoglobin in blood.4 He was able to show that oxygenated hemoglobin is diamagnetic (it has no unpaired electrons and as a result has no magnetic susceptibility). As a corollary, deoxygenated hemoglobin is paramagnetic (has magnetic susceptibility). Totally deoxygenated hemoglobin has a 20% greater magnetic susceptibility than oxygenated hemoglobin. The significance of this observation reported by Pauling subsequently caught the attention of other leading researchers in the field—paramagnetic property leads to a distortion of the surrounding magnetic field, which will induce a rapid decay of the transverse magnetization—the T2 signal. (T1 and T2 are the standard signals in MRI.) This subsequently was recognized as the origin of the BOLD (blood oxygen level dependent) signal.
1980—Fifty years after the description of the paramagnetic property of deoxyhemoglobin, Thulborn verified the observation reported by Pauling in 1931.5 Thulborn showed that the decay of the transverse relaxation of blood (T2) changed with the oxygenation of blood. He also showed that T2 is proportional to the square of the strength of the magnetic field. Tesla (T) is the preferred (SI) unit for measuring magnetic field strength (1 T = 10,000 gauss, gauss (G) being the older magnetic unit). For the sake of comparison the strength of the magnetic field at the Earth’s core is 2.3 G. While there was very little difference in magnetization of oxyhemoglobin and deoxyhemoglobin at 0.5 T, at 1.5 T there was a significant difference in the magnetization property. Only 0.5T and 1.5T MRI scanners were commercially available in the 1980s.
1982—The first 1.5T MRI scanner was installed at Duke University by GE (the General Electric Company). Subsequently, GE was able to foresee the clinical value of MRI and started installing 1.5T scanners in hospitals. In 1985, the FDA approved the clinical use of MRI in diagnostic imaging and insurance companies allowed hospitals to bill for MRI scanning. To GE’s credit, the 1.5T MRI scanner soon became the standard and remained the workhorse for medical imaging in neurological diagnosis for the next 20 years (when it was gradually replaced by 3T MRI scanners).
The widespread availability of MRI scanners for clinical imaging was the necessary driving force for the development of fMRI. Research scientists were able to use the scanners for fMRI during late hours and weekends. The additional requirement was better hardware, like higher-resolution gradient coils, various types of activation paradigms linked to MRI, and data storage and analysis capacity for vast amounts of data.
1991—The first fMRI study was reported by Jack Bellievu from Massachusetts General Hospital, Boston, at the Society of Magnetic Resonance Medicine meeting in 1991.6 Belliveau measured cerebral blood volume (CBV) as a function of neuronal activity. He used the dynamic susceptibility contrast technique for measuring CBV. He administered two injections of gadolinium (before and after visual stimulation) and measured CBV changes with visual activation. Belliveau’s landmark study of functional mapping of the human visual cortex was published in Science. However, the technique of gadolinium injection never became popular. Subsequently, endogenous contrast techniques—i.e., BOLD and pulsed arterial spin labeling (PASL)—were identified, and they were adopted by neuroscientists as the standard imaging contrast/signal for fMRI.
1991—Seiji Ogawa carried the work of Thulborn further. The focus of his research was imaging the oxygenation status of blood. Using a 7T magnet, Ogawa was able to show that hemoglobin can be imaged in rats in vivo. He postulated that variation in blood oxygenation could be visualized in the blood vessels of the brain. Thus, the concept of blood oxygenation as an endogenous contrast was recognized and the term BOLD (blood oxygen level dependent contrast) effect was coined.7,8 Ogawa went one step further and proposed that variation in cerebral metabolism could be imaged with BOLD contrast. As a proof, he compared the BOLD contrast of rats anesthetized with 3% halothane and that of rats anesthetized with 0.75% halothane. Ogawa observed an increase in BOLD signal at 0.75% halothane but very weak BOLD signal (implying low metabolism) at 3% halothane. The difference in the BOLD signals at the two halothane concentrations was related to the variation in the oxyhemoglobin:deoxyhemoglobin ratio (which in turn was related to the change in metabolic rate). The possible explanation for the variation in BOLD signals at the two levels of halothane anesthesia is that, with a higher metabolic rate, there is either a decrease in oxyhemoglobin (more oxygen consumption) or an increase in CBF in response to the increase in metabolism and decreased deoxyhemoglobin concentration. In another set of studies, Ogawa was able to prove that the BOLD effect was a CBF-related phenomenon. He imaged one group of rats with 100% oxygen and another group with 90% oxygen and 10% carbon dioxide. Carbon dioxide inhalation is well known to increase CBF (10% CO2 increases CBF by 300%). Inhalation of 10% CO2 totally eliminated the BOLD signal in the brain because of the increase in oxyhemoglobin concentration (Figure 1.2). This very important concept in the understanding of the origin of BOLD signal (which is the scientific basis for fMRI studies) and its relationship to metabolism was confirmed by two other studies described below.
1988—Fox and Raichle demonstrated that activation of a region of the brain decreases the oxygen extraction from blood in that region (because of an increase in blood flow out of proportion to the oxygen demand).9 This results in a rise in oxyhemoglobin concentration in response to the increase in neuronal activity and metabolism. Using PET imaging, Fox and Raichle measured the change in CBF, cerebral metabolic rate of glucose (CMRg), and CMRO2 in human volunteers when they were given a visual activation task. While there was a 50% increase in CBF and CMRg, the rise in CMRO2 was only 5% (Figure 1.3). This landmark study disproved the conventional belief, held until then, that the brain is entirely dependent on aerobic metabolism for its energy. The study confirmed that there is anaerobic metabolism initially during a neuronal activation. The increase in CBF without a proportional rise in CMRO2 (causing an alteration in the oxyhemoglobin:deoxyhemoglobin ratio) is the origin of the BOLD signal related to an increase in neuronal activity.
1996—Maloneck and Grinvald, using a sensitive optical electrode, were able to trace the change in oxyhemoglobin and deoxyhemoglobin concentration during neuronal activation.10 They demonstrated that, initially for 2 seconds after neuronal activation, the oxyhemoglobin level drops, and over the next 5 to 6 seconds there is a rise in oxyhemoglobin concentration (Figure 1.4). The change in BOLD signal response follows the change in oxyhemoglobin concentration. This hemodynamic change with neuronal activation, which is the genesis of the BOLD signal, is now common knowledge among fMRI scientists.
The year 1991–1992 is considered the beginning of fMRI, coinciding with the studies by Ogawa, who coined the term BOLD and identified it as a qualitative measure of neuronal metabolism. To commemorate this discovery, in 2012 the journal NeuroImage published an entire issue devoted to the 20th anniversary of fMRI. Peter Bandettini, Director of the NIH section on neuroimaging, served as editor-in-chief for the issue.11 Over 120 articles related to the development of fMRI were published in the issue.
Early fMRI Studies
After Ogawa’s landmark BOLD studies, several other fMRI studies followed. Kwong et al. did a visual activation study, while Ogawa’s group repeated the visual activation study with varying periods of stimulation and signal acquisition.8,12 Bandettini’s group did a motor activation study (what is now commonly known as a finger-tapping study).13 Blamire did a study with a 2-second visual stimulation and demonstrated that there is a delay in BOLD response (because of the hemodynamic response to neuronal activation).14 The delay in hemodynamic response was demonstrated earlier by Grinwald’s study. Now it is an accepted practice to acknowledge that, while BOLD response reflects neuronal activity, it is an indirect response related to hemodynamic change. In all of the studies with sensory (visual, auditory)/motor activation, the region of the brain being activated could be visualized. fMRI was an objective method for visualizing subjective responses in the brain. It has become a standard technique in cognitive neuroscience. Correlation of phenotype with behavioral functions has been facilitated by fMRI. In psychiatric diseases, correlation between genetic alterations and the resultant phenotypic changes in the brain is also feasible.
As of today, three types of fMRI studies have been carried out:
• Task-activation studies—the conventional fMRI studies
• Regional CBF studies
• Connectivity studies
Task-activation studies are similar to the studies carried out by Kwong, Bandettini, Blamaire, and many others. Since BOLD is a qualitative measure of neuronal activity, it is a dimensionless number. No inference can be reached based on the absolute BOLD value. It is the relative change in BOLD that is significant. The basic design in task-activation studies is that a BOLD image is captured during the resting stage and again during task activation. The difference in signal between activation and resting state is considered to be related to the task. It is necessary to image the entire brain in order to identify the various regions where BOLD (and hence neuronal activity) is altered because of the task. Furthermore, imaging speed has to be rapid to capture the BOLD changes in the entire brain. The relative signal change induced by task activation is ~ 2% to 3% above the baseline.15,16 And there are several sources of noise (extraneous signals) that can interfere with or corrupt the signal being measured. The common sources of noise are intrinsic thermal noise, imperfections in scanner hardware, patient’s head motion, physiological factors (respiratory and cardiac oscillations), variations in neuronal activity unrelated to the task, and alterations in behavioral performance in the subject over time. These are explained more in detail in the following chapters. In order to optimize the signal-to-noise ratio (SNR), multiple images are required and the acquired data sets are averaged. This is the classical task-activation study. Basic sensory modalities, like visual, auditory, and somatosensory activation, as well as motor activation, were used in the early fMRI studies. Subsequently, higher-order functions, like memory retention, recall, emotion, and decision-making, also have been studied. The basic aim in all the studies is to understand how and where in the brain the functions are modulated. Subsequently the effect of medication on these functions was also studied.
CBF has been measured using the arterial spin labeling (ASL) technique. ASL is recognized as another endogenous contrast in fMRI studies. In the ASL technique, quantitative measurement of CBF can also be done. In the ASL technique, protons (H+ ions, which are ubiquitous in blood) act as the endogenous contrast. These protons are referred to as spins.17 The spins in the feeding artery to the brain region (cerebral circulation) are transiently exposed to radiofrequency pulses. When exposed to a radiofrequency pulse, the polarity of the spins is altered—the spins become inversely labeled. When the inversely labeled spins flow through the vascular tree, they change the spin-lattice relaxation rate and alter the T1-weighted MRI signal. The change in T1-weighted signal is proportional to the CBF. As seen in Figure 1.5, on the arterial side, there are white circles (spins), which are deflected down (polarized protons or spins). While passing through the capillaries, some of the spins regain their polarity (gray circles with arrow pointing up) and some remain in a polarized state. The proportion of spins that change/regain their polarity is related to the regional CBF. Based on the change in the T1 signal, the regional CBF is computed.
Because of the flow–metabolism coupling in the brain, change in blood flow is proportional to change in metabolism under physiological conditions. However, the coupling relationship holds good only for global measurement of CBF and for measurement in a larger area of the brain. As a baseline measure, CBF does reflect cerebral metabolism. With age, there is neuronal degeneration in the brain, leading to a decrease in cerebral metabolism. This is more regionally distributed in certain areas of the brain (like the frontal and temporal lobes). The change in cerebral metabolism and blood flow with age has been studied with the ASL technique. Because of the high CBF (the brain is only 2% of the body’s weight, but it receives 15% of cardiac output and accounts for 20% of the body’s total oxygen consumption), the mean transit time of blood in the brain is short. Hence, imaging time in ASL has to be rapid in order to capture the change in polarity of the spins, which results in a slight decrease in spatial resolution.
Task-activation studies work well while the focus is on one or two localized functions of the brain. However, there are some limitations to task-activation studies. Brain consumes the bulk of its energy in the resting state (20% of total body oxygen consumption). As mentioned, the increase in energy consumption during a task is only 2% to 3% above the baseline. In a task- activation study, the baseline energy state of the brain is totally ignored and the primary focus is on the incremental energy consumption during the task. Many of the task-activation studies have identified multiple regions of activation (smaller scattered areas) in addition to the primary sensory/motor activation regions. Initially, these smaller areas of activation were interpreted as electrical noise, which is common in fMRI (as mentioned earlier). Subsequently, neuroscientists recognized that brain function should be studied like the electroencephalogram (EEG), since the various anatomical regions of brain are connected to each other and they constantly exchange information within and between each other all the time. This is the functional state of the brain while it is at rest as well as during the execution of a task. Hence the term resting brain is inappropriate and restless brain would be more appropriate. This has led to the concept of connectivity in brain function. Connectivity refers to temporally correlated signals between functionally related but anatomically distinct regions of the brain. In the last 10 years, the focus of fMRI studies has shifted from task activation studies to connectivity studies. There is a growing interest in studying connectivity in the resting state as well as during task activation. Biswal et al. first reported that, in the resting state, spontaneous fluctuation in low-frequency (0.01–0.1 Hz) BOLD activity has a striking spatial correspondence with sensorimotor regions of the cerebral cortex.18 Several specific connectivity networks (also referred to as resting-state networks) associated with specific functions have been identified over the years. These networks are distinct and are present during the awake state, during physiological sleep, and under anesthesia also (albeit with a decrease in activity). The spatial pattern of the networks extends beyond the anatomical landmarks in the brain. Some of the networks identified in the resting state are the same regions that get activated during a task.
fMRI Data Acquisition and Analysis—The Challenge
fMRI Data Acquisition
fMRI is technically very complex, requiring specific data acquisition parameters, and the data analysis (described in the next section) is also intense, requiring multiple steps. Because of the time involved and the technical challenges in data analysis, in spite of being an important diagnostic tool in neuroscience, fMRI imaging is still not a real-time imaging technique. This means that, in fMRI, the images do not appear as we are doing the study (although they do appear during a conventional structural MRI and CT scan)—the data has to be preprocessed and analyzed using specific software developed for fMRI data analysis. After the data analysis, fMRI images are created based on the data. Although some basic programs for real-time imaging are available for neurosurgical applications, the programs were written for specific simple applications, such as motor and speech activation.
For clinicians interested in fMRI, it is important to have some basic knowledge of how fMRI data are acquired and are subsequently analyzed and interpreted. Since fMRI is a measure of neuronal activity in the brain, it is essential to image the entire brain in fMRI in order to capture the neuronal activity in the whole brain (or alteration in connectivity) in response to a stimulus. Any functional activity in the brain is known to be modulated in multiple regions of the brain—it is a well-coordinated activity in several regions that results in sensory perception (or a motor activity). In studies done in the past, brainstem and cerebellum were frequently not included in imaging, but now it is standard practice to image the entire brain, up to and including the brainstem, in most fMRI studies. It is also imperative that the entire brain be imaged within a short period of time—before the BOLD signals dissipate. The time required for imaging the whole brain is referred to as the repetition time (TR). Based on data from many studies, it has been determined that the optimum TR is 2 seconds (2,000 milliseconds). A TR of less than 2 seconds has no advantage. Multiple slices of the brain are imaged to cover the entire brain in 2 seconds. Field of view (FOV) is the area covered in one slice, and matrix size refers to the number of data points in each plane (x- and y-axis). A matrix size of 64 × 64 means there are 4,096 data points in one slice. The matrix size is usually kept as a multiple of 2 in order to facilitate fast Fourier transformation (FFT), which is required for image construction later. If the FOV is 240 × 240 mm (cross-sectional area being imaged) each data point is 3.75 mm apart (240/64 = 3.75 mm). This distance between data points (which depends on the FOV and the matrix size) determines the spatial resolution of the scan, which is expressed in terms of voxels. Voxels (like the pixels in a camera) are the 3D volume in which the signal is measured. For a FOV of 240 mm and matrix size of 64 × 64, the voxel size is 3.75 × 3.75 mm. The third dimension is the thickness of the slices. Isotropic voxels are those where all three dimensions are the same (e.g., 3.75 × 3.75 × 3.75 mm). Isotropic voxels are preferred because signals are always measured at the center of the voxel. The number of slices acquired depends on the dimensions of the patient’s brain and the thickness of the slice. For an adult with an average brain dimension, 35 slices are required. Thus, if there are 4,096 data points in one slice and there are 35 slices, the total number of data points in the entire brain is 122,880 (4,096 × 35). This is referred to as one volume. In an activation study, the change in signal with activation is ~ 2% to 3% above the baseline. Since the fMRI signal is frequently corrupted by noise, in order to optimize the signal-to-noise ratio (SNR), multiple volumes of images are acquired. Typically, 150 or 200 volumes are acquired in an activation study. With a 2-second TR, the time required for acquiring one set of images consisting of 200 volumes is 400 seconds. This is one set of data. Frequently, two such data sets are acquired for each condition. Subsequently the data is averaged during data analysis.
fMRI Data Analysis
Because of the large volume of data (122,880 data points in one brain volume), programs specifically designed for fMRI data analysis are required for preprocessing and statistical analysis. Many programs, like SPM (Statistical Parametric Mapping), Free Surfer, Analysis of Functional Neuro Image (AFNI), and BrainVoyager, have been designed specifically for fMRI data analysis. (Because PET scan and EEG analysis also involve a large volume of data, SPM and certain other programs have the capability for PET scan and EEG analysis as well.) There are three essential steps in preprocessing: realignment, normalization, and smoothing.
Preprocessing is specifically required in fMRI analysis for two reasons—there is frequently some inadvertent movement of the head during scanning that needs to be corrected for. When multiple volumes of data are being averaged, it is essential that all the slices be in the same anatomical alignment. Realignment of data is also referred to as rigid transformation, since the shape of the brain slices is not altered in this step. Realignment can be done manually (if the correction required is large because of significant movement of the head), and/or finer realignment in x-, y-, and z-axes can be carried out using the automated program. When fMRI data is acquired and analyzed, the anatomical localization of the functional activity cannot be identified from functional images. To facilitate anatomical localization, a high-resolution 3D anatomical image (called a T1 scan) is acquired as part of fMRI and the functional slices are overlaid on the anatomical images to identify the anatomical correlation of the functional activity. However, when several subjects are being studied, the anatomical configuration of each subject’s brain is different. While each individual subject’s data can be referenced to the T1 image for each subject, for between-subject group comparison, a standard reference anatomical atlas is necessary. The two standard brain atlases referenced internationally are the Talairach atlas (www.talairach.org) and the MNI (Montreal Neurological Institute) atlas. Both of the atlases use the x, y, and z coordinates for the appropriate anatomical region of the brain that corresponds to a voxel. This step in preprocessing is called normalization. The basic steps in normalization are—the high-resolution T1 anatomical image of the brain is normalized to the template image (Talairach or MNI atlas). When this is done, the T1 image of the subjects gets warped slightly to fit the shape of the standard brain. In the next step (since the T1 image and all functional images are of the same alignment and configuration), the normalization parameters for the T1 image are applied to all the functional images. This ensures that the functional images of all subjects are of the same configuration. This is the nonrigid transformation step (since the shape of the brain is slightly warped to fit the standard atlas image). Once this is done, all the data from several subjects can be averaged to measure the change in neuronal activity in the group as a whole and the Talairach atlas (or MNI) can be used for identifying the anatomical localization of the regions of interest (ROI). Smoothing is the final step in preprocessing. In this step, 3D Gaussian filter is used in averaging data from several adjacent voxels in order to improve the statistical (functional) validity, which at the same time reduces the spatial resolution of the data. Typically, smoothing is done to a voxel size of 6 × 6 × 6 mm. As a result, if 4 × 4 × 4 mm voxels are acquired, there will be more than three times as many 4-mm voxels if the 6-mm smoothing filter is applied. This increases the signal strength and improves the statistical validity of the data.
Programs like SPM and Free Surfer are MATLAB-based programs for preprocessing and statistical analysis. (MATLAB is a purely mathematical program for handling statistical analysis of a large volume of data.) As already mentioned, in fMRI there are 122,880 data points in one brain volume. There are 200 such volumes of data for one set, and two such sets of data are acquired in most studies. Thus, there are 400 volumes of data with 122,880 data points. MATLAB has the capability to carry out calculations on such multiple data points within a few seconds or minutes.
Statistical tests like the t test can be carried out on the various data points (corresponding to individual voxels). Based on this, a “t” map is created and plotted on the 3D image of the brain. Because of the large number of data points in each subject’s brain, standard statistical techniques with a significance level of P < .05 or < .01 may be invalid in the fMRI studies. As an example, if P < .05 is chosen, out of 122,000 voxels, in 2,440 voxels (5% of voxels) activation may be purely by chance. If a standard statistical correction like the Bonferroni is applied, it could lead to very conservative results because the fMRI data in individual voxels are not totally independent of each other. The adjacent voxels have activity related to each other. Hence the corrections recommend in fMRI statistics are false discovery rate (FDR) or family-wise error (FWE). FWE is more conservative than FDR. Sometimes P < .001 is also chosen—this is less conservative than FDR and FWE.
Future of fMRI: Has It Affected the Practice of Medical Science?
fMRI brought together physics-based technology and cognitive neuroscience. Experts from a wide range of disciplines, such as medical physics, imaging, neuroscience, psychiatry, clinical psychology, neurology, and several others, worked together to improve our understanding of brain function in health and disease. In the last 20 years, fMRI has matured from being a research tool to become a very commonly used diagnostic tool in cognitive neuroscience. In 2010, a survey was done on the various articles published in the Journal of Cognitive Neuroscience. It was observed that in 52% of the published articles, fMRI technology had been used. fMRI is now accepted as a fundamental tool for studying the link between brain and behavior. It has improved our understanding of the phenotype of many neurological and psychiatric disorders. In scientific publications, PubMed citations on fMRI compare with eight other high-impact keywords (e.g., knockout mice, HIV, MRI, telomere, IVF—fields associated with several Nobel Prizes in recent years). The number of publications on fMRI has increased exponentially in the last 20 years. However, in the field of psychiatry, as per the comments of Tom Insel (Director of the National Institute of Mental Health), in spite of over 4,000 publications, fMRI has not changed the routine practice of psychiatry. Common psychiatric disorders like schizophrenia, bipolar disorder, and depression have been studied extensively with fMRI, which has greatly improved our understanding of these conditions. Dr. Insel added that this will change in the future because the new concept of psychiatric disease as an abnormality of circuitry (as opposed to the age-old concept of psychiatric disease as being related to specific regions of the brain) is likely to change the diagnosis, prevention, and treatment of these diseases. The connectivity concept has added a new dimension to the description of both normal brain function and brain function in disease.
Meanwhile, with regard to the question about the impact of fMRI on clinical medicine, there are four areas where fMRI is being used as a diagnostic tool:
1. In neurosurgery, fMRI is frequently being used in the preoperative planning for brain tumor surgery and for surgical resection of epileptic foci close to eloquent regions of the brain (e.g., speech and motor areas). In 2007, a CPT code for fMRI in preoperative planning was introduced for billing purposes. Thus, it is no longer considered a research application if fMRI is done for preoperative planning in neurosurgery.
2. In the study of consciousness, fMRI is very helpful in differentiating various altered states of consciousness, such as vegetative state versus minimally conscious state. Based on fMRI, it is possible to determine a prognosis for individual cases (since minimally conscious state has a better prognosis than vegetative state).
3. In stroke and TBI (traumatic brain injury), fMRI can be used for follow-up of patients during recovery.
4. In psychiatric diseases like schizophrenia, depression, bipolar disorder, and attention deficit disorder, fMRI is used for imaging the functional disorder in the brain and for assessing the efficacy of treatment (although this is not yet considered a standard of care).
Meanwhile, many clinicians have looked at why this potentially valuable mode of functional imaging is not being translated into a clinically useful diagnostic tool. In order for fMRI to be used as a diagnostic tool in clinical neuroscience, a large volume of normative data from healthy subjects is required for comparison to data from patients in disease states. At this time, most fMRI studies are done in small groups of volunteers or patients. In the meantime, meta-analytic techniques combining the activation maps from several studies are being used to create an activation-likelihood map for a task. In schizophrenia, the maps have confirmed that, neurologically, schizophrenia is a disease affecting multiple regions of the brain, because of the change in connectivity. Nevertheless, for it to be used as a routine diagnostic measure in clinical practice, fMRI has to be a real-time imaging technique. Furthermore, data analysis techniques in fMRI still vary between research centers, there is no standard technique of data acquisition and analysis, and no attempt been made to cross validate the results of the various techniques.
Progress in fMRI
The slow progress in the incorporation of fMRI as a diagnostic tool in clinical medicine does not seem to be impeding progress in fMRI research. Several major research projects funded mostly by the National Institutes of Health (NIH) and other funding agencies are in progress in institutions all over the world. The Human Connectome Project (HCP) is a $30 million NIH-funded project to chart the human connectome and its variability in 1,200 healthy adults using fMRI, diffusion MRI, task-activated fMRI, EEG, and MEG (magnetoencephalography). Many studies of brain function across the lifespan include fMRI as the primary tool. The NKI (Nathan Kline Institute) Rockland Sample is a study in which 1,000 subjects 5 to 85 years old are being followed with fMRI (and other behavioral functions) for 5 years in order to create a broad and deep phenotype characterization of brain function across the lifespan. The ADNI (Alzheimer’s Disease Neuroimaging Initiative) and MRN (Mind Research Network) are two other such studies. Meanwhile, many studies on alteration in functional connectivity and regional cerebral blood flow with age and their correlation with behavioral changes have also been carried out.
On the technology side, fMRI, which began with 1.5T magnets, has progressed to 3T magnets in most institutions. Some institutions have acquired 7T magnets, and even an 11T magnet is being studied at the University of Minnesota. Higher-tesla magnets improve the spatial resolution of the functional images. MR spectroscopy (MRS) is another technique being developed. MRS has the capability to measure the concentration of a neurochemical compounds (including neurotransmitters like glutamate and GABA) and metabolites in the brain in vivo and noninvasively. MRS requires a high-field magnet, and this has been the driving force for the development of 4T, 7T, and now the 11T magnets.
In conclusion, in the 20th century, structural imaging techniques improved significantly, resulting in the availability of high-resolution MRI in many centers. The impetus for fMRI came from PET scan based functional imaging. fMRI was the adaptation of functional imaging techniques into MRI using endogenous contrast (BOLD). Over the last 20+ years with the progress in MRI technology and computing techniques, fMRI has developed into an important diagnostic tool in neuroscience. In the following chapters, the basics of fMRI technology, data acquisition, data analysis, and interpretation, as well as emerging clinical applications, are discussed in detail.
1. Lassen NA, Ingvar DH, Skinhoj E. Brain function and blood flow. Sci Am. 1978;239:62–71.Find this resource:
2. Sokoloff L, Reivich M, Kennedy C, et al. The [14C]deoxyglucose method for the measurement of local cerebral glucose utilization: theory, procedure, and normal values in the conscious and anesthetized albino rat. J Neurochem. 1977;28:897–916.Find this resource:
3. Petersen SE, Fox PT, Posner MI, Mintun M, Raichle ME. Positron emission tomographic studies of the cortical anatomy of single-word processing. Nature. 1988;331:585–589.Find this resource:
4. Pauling L, Coryell CD. The magnetic properties and structure of hemoglobin, oxyhemoglobin and carbonmonoxyhemoglobin. Proc Natl Acad Sci U S A. 1936;22:210–216.Find this resource:
5. Thulborn KR, Waterton JC, Matthews PM, Radda GK. Oxygenation dependence of the transverse relaxation time of water protons in whole blood at high field. Biochim Biophys Acta. 1982;714:265–270.Find this resource:
6. Belliveau JW, Kennedy DN Jr, McKinstry RC, et al. Functional mapping of the human visual cortex by magnetic resonance imaging. Science. 1991;254:716–719.Find this resource:
7. Ogawa S, Lee TM, Kay AR, Tank DW. Brain magnetic resonance imaging with contrast dependent on blood oxygenation. Proc Natl Acad Sci U S A. 1990;87:9868–9872.Find this resource:
8. Ogawa S, Tank DW, Menon R, et al. Intrinsic signal changes accompanying sensory stimulation: functional brain mapping with magnetic resonance imaging. Proc Natl Acad Sci U S A. 1992;89:5951–5955.Find this resource:
9. Fox PT, Raichle ME, Mintun MA, Dence C. Nonoxidative glucose consumption during focal physiologic neural activity. Science. 1988;241:462–464.Find this resource:
10. Malonek D, Grinvald A. Interactions between electrical activity and cortical microcirculation revealed by imaging spectroscopy: implications for functional brain mapping. Science. 1996;272:551–554Find this resource:
11. Bandettini PA. Twenty years of functional MRI: the science and the stories. NeuroImage. 2012;62:575–588.Find this resource:
12. Kwong KK, Belliveau JW, Chesler DA, et al. Dynamic magnetic resonance imaging of human brain activity during primary sensory stimulation. Proc Natl Acad Sci U S A. 1992;89:5675–5679.Find this resource:
13. Bandettini PA, Wong EC, Hinks RS, Tikofsky RS, Hyde JS. Time course EPI of human brain function during task activation. Magn Reson Med. 1992;25:390–397.Find this resource:
14. Blamire AM, Ogawa S, Ugurbil K, et al. Dynamic mapping of the human visual cortex by high-speed magnetic resonance imaging. Proc Natl Acad Sci U S A. 1992;89:11069–11073.Find this resource:
15. Raichle ME. Two views of brain function. Trends Cogn Sci. 2010;14:180–190.Find this resource:
16. Raichle ME, Mintun MA. Brain work and brain imaging. Annu Rev Neurosci. 2006;29:449–476.Find this resource:
17. Golay X, Hendrikse J, Lim TC. Perfusion imaging using arterial spin labeling. Top Magn Reson Imaging. 2004;15:10–27.Find this resource:
18. Biswal B, Yetkin FZ, Haughton VM, Hyde JS. Functional connectivity in the motor cortex of resting human brain using echo-planar MRI. Magn Reson Med. 1995;34:537–541.Find this resource: