History of Functional Magnetic Resonance Imaging

While conventional MRI (magnetic resonance imaging) and CAT (computerized axial tomogram) scan image the structural anatomy of the brain (and pathology) to a high degree of spatial resolution, functional imaging has carried this imaging capability one step further by imaging the functional activity in the brain, the ability to dynamically image functional activity in the brain at a significantly higher level of spatial and temporal resolution. Historically, prior to the imaging era, neurophysiology was studied by conducting autopsies on patients and correlating the autopsy findings with their neurological function in their life time. Well-described functional areas of the brain like the Wernicke area, Broca area, and the Herschel cortex (auditory cortex) were identified and studied by such techniques in the past. The availability of structural and functional imaging of the brain has totally altered the landscape and has given us a unique opportunity to study and understand brain function in health and disease in vivo using noninvasive imaging techniques like PET (positron emission tomogram) scan and (functional MRI) fMRI.

The scientific basis of functional imaging is the change in cerebral metabolism and cerebral blood flow (CBF) associated with neurological function in the brain. Functional imaging is based on imaging the spatial and temporal sequence of the change in cerebral metabolism and blood flow linked to any functionally related neuronal activity in the brain. By such techniques, simple functions like visual, auditory, and somatosensory activations were imaged initially. Subsequently, highly complex functions like memory, recall, emotion, anger, and speech also have been imaged. Lassen, the internationally well-known neurophysiologist, is credited with the very first publication in functional imaging. In 1978, he published his landmark study on language perception imaged with the Xenon ¹³³ technique of measuring regional cerebral blood flow (rCBF) during a language processing paradigm ( Fig. 11.1 ). Subsequently, Peterson, in 1988, imaged word processing with ¹⁵ O-labeled water with an early version of PET scan. To this day, the ¹⁵ O-labeled water technique of measuring rCBF is the gold standard for measuring CBF. And because of the short half-life of ¹⁵ O (2 min), repeated measurements of rCBF can be carried out every 15 min, thus improving the temporal sequences of measuring a functional activity. As explained later in the section, fMRI improved the temporal sequence to a much higher level. Dr. John Belliveau from Massachusetts General Hospital, Boston, published in Science images of visual activation of the brain using a gadolinium injection–enhanced MRI technique at a time when even MR imaging was in its infancy (the first MRI scanners were installed in 1984). This publication by Dr. Belliveau revealed for the first time enhanced blood flow and blood volume in the visual cortex in response to visual activation. The increase in blood flow in the brain in response to a neuronal activity was first described by Roy and Sherrington (neurophysiologists) over a 100 years ago, but the scientific proof of this response came much later. Belliveau is given the due credit by the MRI community for initiating fMRI studies, and 1991 is considered the year when fMRI truly got started, and the year 1978 is when functional imaging was invented. To commemorate this, in the year 2012 the journal Neuro Image published one entire issue of the journal on fMRI to mark the 20th anniversary of fMRI. With rapid progress in the field, Belliveau’s technique of fMRI itself became obsolete when BOLD (blood oxygen level–dependent contrast) signal was identified, captured, and imaged. Since then, BOLD (an endogenous contrast) has become the standard contrast measured in fMRI studies. Subsequently, with improved hardware (high-resolution MRI like 3 and 7 T, superior coils for signal capture) the spatial and temporal resolution of fMRI has improved significantly.

Xe ¹³³ CBF study: Visual activation (left) with rise in CBF in the visual association area and SMS (supplementary motor area). Auditory activation (right) with rise in CBF in the auditory cortex.

While PET scan is the gold standard for CBF measurement the spatial and temporal resolution of PET scan is lower compared to fMRI. CBF measurement can be repeated only after 15 min because of the 2-min half-life of the ¹⁵ O isotope. fMRI has a spatial resolution of 0.5 mm and temporal resolution of 2–5 s (because of the delay in blood flow response, as will be described later in the section). Notwithstanding these limitations, PET is very accurate for what it measures. Electroencephalogram (EEG) is the only technique with a very high temporal resolution comparable to neuronal activity. However, the spatial resolution of EEG is poor. Hence, in functional imaging, techniques of measuring fMRI and EEG simultaneously have been developed to avail of the higher spatial resolution of fMRI and the higher temporal resolution of EEG.

What Is Functional Magnetic Resonance Imaging?

Functional MRI is based on the detection and imaging of BOLD (blood oxygen level–dependent contrast) signal, which is linked to any functionally related neuronal activity in the brain. BOLD signal was described by Ogawa in 1991. The origin of BOLD signal and its relation to neuronal activity is explained as follows: any functionally related neuronal activity in the brain results in an increase in metabolism in the related region of the brain. The brain, as is well known, is metabolically a very active organ (among all other organs of the body); it is 2% of the body weight, gets 15% of the cardiac output, and consumes 20% of the total body oxygen consumption. Thus, weight for weight, the brain consumes 10 times more oxygen compared to the total body average consumption. To look at this in perspective, there are 80 billion neurons in the brain and billion billion (10 ¹⁸ ) synaptic connections. Each cerebral cortex when unfolded is the size of a 12-in. Pizza and is 2 mm thick. Each 1 mm ² of cerebral cortex has approximately 90,000 neurons. And 2 mm is by far the smallest dimension (voxel) where signal is measured in fMRI. This implies that the ability to store and analyze data is enormous in the human brain, and this capability is comparable to any advanced computer in the world.

It is this change in metabolism, blood flow change in this highly complex and yet intricately connected organ we are studying in fMRI. Whenever there is an increase in metabolism in a region of the brain, there is a parallel increase in CBF to match the increase in energy requirement. The single most important factor that is responsible for the BOLD signal (the primary signal measured in fMRI) is, in addition to the rise in cerebral metabolism and a parallel increase in CBF, for a brief period the increase in blood flow is out of proportion to the increase in metabolism. It is this disproportionate increase in blood flow for 2–5 s that is responsible for the emission of the BOLD signal. This increase in metabolism and blood flow has been confirmed by other studies.

The scientific basis of fMRI is based on the seminal work of Ogawa on the variation in the magnetic properties of oxyhemoglobin (oxy Hb), which is diamagnetic, and deoxyhemoglobin (deoxy Hb), which is paramagnetic. In 1936, Linus Pauling, in his Nobel Prize–winning research, discovered that deoxy Hb has 20% more magnetic susceptibility (because of two unpaired electrons) compared to oxy Hb (which has no unpaired electrons). Ogawa realized that the standard MRI, since it is based on a proton signal (which is ubiquitous in the human body: 70% of human body is H ₂ O) is not suitable for studying human physiology. He was looking for an MRI signal that would be sensitive to physiological changes like brain metabolism. In a study in rodents, he observed that there was a big change in MRI signal (T2 weighted signal) when O ₂ concentration changes from 100% to 21% to 0%. This led to his speculation that the blood oxygen level–dependent signal can identify neuronal metabolic activity in the brain. When halothane anesthesia was administered to rats, at deeper levels of anesthesia since brain metabolism is lower and oxy Hb concentration is higher (and so BOLD signal is poor), while at lighter a level of anesthesia, brain metabolism is higher, and deoxy Hb level is also high (and so the BOLD signal is stronger). He summarized this by stating that BOLD MRI contrast signal depends on the ratio of oxy Hb to deoxy Hb, which in turn depends on the oxygen demand and supply in the brain. Sokoloff, in his autoradiographic study, demonstrated the coupling between cerebral metabolism and CBF in the brain. Autoradiography is the equivalent of present-day PET scan. Subsequent work by Fox and Raichle added further proof and confirmed the scientific basis of BOLD proposed by Ogawa. In a very elegant PET scan study (confirming the basis of fMRI) in 1988, during a visual activation study, they measured the change in cerebral metabolic rate of oxygen consumption (CMRO ₂ ), CBF, and cerebral metabolic rate of glucose consumption (CMRg). With visual activation, as expected, there was a rise in CBF and CMRg (50% and 51%, respectively). However, the rise in CMRO ₂ was only 5%. They concluded that during a neuronal activation, initially, energy is generated predominantly by anaerobic metabolism. This has been confirmed by another study where prolonged visual activation resulted in a rise in lactate concentration in the concerned region of the brain. This selective uncoupling of CMRO ₂ and CMRg (5% and 51% rise) causes an increase in oxy Hb and decrease in deoxy Hb. This change in ratio oxy Hb to deoxy Hb is responsible for the BOLD signal. Maloneck and Grinwald measured the concentration of oxy Hb and Deoxy Hb in the visual cortex of cats during visual activation using a very sensitive optical imaging technique ( Fig. 11.2 ). They reported that in the first 2 s, deoxy Hb concentration rises, and after a delay of 5 s, while deoxy Hb concentration drops, oxy Hb concentration rises in the visual cortex. This time course of rise in oxy Hb level corresponds well with the time course of the BOLD signal. Thus, while agreeing with the uncoupling hypothesis of Fox and Raichle, they explained this phenomenon on the basis of rise in CBF (rather than an increase in anaerobic metabolism).

Visual activation (left) change in reduced Hb; slight rise followed by a drop. (right) Change in oxy Hb; delayed rise with a peak in 6 s.

It has been emphasized all along that BOLD signal, while it is linked to neuronal activity, is an indirect response related to the hemodynamic change associated with neuronal activation. This is important to understand while interpreting these signal changes, especially in anesthesia studies because some anesthetic agents can influence CBF independent of the change in metabolism. In a forepaw stimulation study in rats in 1999, Mandeville demonstrated that with activation of forepaw, as BOLD signal increases, CBF and cerebral blood volume also increases. This confirms the rise in CBF with activation (forepaw in this case), which in turn causes a rise is BOLD signal. BOLD signal change is a qualitative change and not a quantitative responsive. Hence, it is the relative change in BOLD that is significant and not the absolute change in BOLD.

Practical Aspects of Functional Magnetic Resonance Imaging: How Is the Data Collected

How Is an Functional Magnetic Resonance Imaging Carried Out?

In a fMRI the entire brain has to be imaged within a short period of time to capture the BOLD signal generated during the functional activation globally in the brain. To understand and explain in neurophysiological terms how a functional activation (auditory, visual activation, memory activation, etc.) is modulated in the entire brain, imaging of the entire brain is necessary. This type of imaging has revealed that even a simple task-induces activation (and some deactivation also) simultaneously, in several interconnected but anatomically widely separated regions of the brain. The classical example would be memory activation induces BOLD response in the frontal cortex, medial temporal cortex, and even in some regions of the sensory cortex in the parietal lobe. Since this response to activation is quite rapid (as is expected) the entire brain has to be imaged in a short period of time. In practice the entire brain, which measures ≅ 24 cm from the brain stem to the vertex (in “z” axis), is imaged in less than 2 s. In a three-dimensional imaging like MRI, multiple slices of images are acquired, covering the whole brain. If the slice thickness is 4 mm (as is the common practice) with a 1 mm gap between the slices (this gap is necessary to avoid any spillage of signals from adjacent voxels), approximately 48 slices are required to cover the whole brain. These slices (one set) are acquired in 2 s. The imaging time for each slice is 40 ms. As explained earlier the BOLD signal strength is not as strong as the proton signal acquired in a standard structural MRI. As a result, multiple images have to be acquired in the same condition (paradigm). Typically, 200 to 300 sets of images are acquired, and the data is averaged. By acquiring multiple sets of images and averaging the data the signal strength can be improved significantly while decreasing the noise (higher signal-to-noise ratio). This one group of 200–300 sets is referred to as one block. Sometimes, researchers acquire two such blocks of images. A voxel (like the pixel of a camera) is the smallest volume of the brain from where data is acquired. A voxel size of 4 mm × 4 mm × 4 mm is preferred in most studies. If the voxel size is too small the signal acquired may be too weak, but the spatial resolution will be high. And if voxel size is too big, while the signal acquired will be strong, the spatial resolution will be poor.

Other Signals Studied in Functional Magnetic Resonance Imaging

As explained, the BOLD signal measured in standard fMRI is a qualitative measure of neuronal activity in the brain. This is the endogenous contrast generated during a neuronal activity (related to the hemodynamic changes induced by neuronal activity) and is commonly measured in fMRI. By referring to BOLD as an endogenous contrast, what we mean is that this contrast is generated intrinsically (and not an extraneous injection like gadolinium) at or close to the site of neuronal activity.

Another signal/technique of measurement in fMRI is the pulsed arterial spin labeling (PASL) technique. This is a noninvasive technique of measuring rCBF quantitatively in the brain. In the PASL technique, resonance is generated in the cerebral vasculature at the level of the neck (carotid and vertebral arteries) ( Fig. 11.3 ). In a high-power magnetic field, H ⁺ ions align themselves parallel and antiparallel in the magnetic field. In the PASL technique, protons in the water molecule of plasma act as the endogenous contrast. These protons in the plasma are referred to as spins. When these spins are intermittently exposed to a radiofrequency pulse, it alters the polarity of these spins transiently. With cerebral perfusion, as blood flows through the arterial tree and capillaries, the spins regain their polarity, and this alters the T1-weighted signal in MRI, which can be captured and measured. The signal change is proportional to the rCBF, and with some assumptions, absolute rCBF can be calculated. Since the change in rCBF is measured at the level of the capillaries, this reflects neuronal activity more precisely (compared to BOLD where the venous blood volume reflects the signal) since the capillaries are closer to the site of neuronal activity in the brain. However, the limitation in PASL technique is imposed by the rapid CBF, which requires that imaging be done more rapidly (to capture the change in polarity of the spins).

Pulsed arterial spin labeling: description of the figure is in the text.

Functional Magnetic Resonance Imaging Studies

fMRI started as a technique of imaging various subjective functions objectively in the brain. Consequently, all the early studies done with fMRI were functional activity–related studies. The basic procedure followed was that specific functional areas of the brain were activated, and the brain was imaged to visualize this functional activity, to locate the region of the brain where such functions were carried out. Before the fMRI era, visual activation and word processing were the two functions studies with PET scan. This was called functional imaging. With fMRI, auditory stimulation and somatosensory activation were the two early studies done at the Welcome Institute of Cognitive Neuroscience, University of London. These two studies have been described extensively in literature, though they were never published (SPM literature). For the first time ever, auditory and somatosensory activation (finger tapping) could be actually visualized in the brain, something known for a long time but never actually demonstrated in vivo in the healthy human brain. The basic technique of conducting these studies involves imaging BOLD signals first in the no-task state (resting state, no activation of the cerebral cortex) of the brain. Next, BOLD imaging is done during a state of functional activation of the brain (choice of activation will be related to the function to be studied: visual, auditory, somatosensory, memory activation, etc.). As described in the earlier section, multiple sets of images are captured, and the data is averaged. During the data analysis, the BOLD image during resting state is subtracted from the activation image. The resulting subtracted image reflects the functional activation of the brain. As a part of any fMRI study a high-resolution anatomical image is also acquired in the same imaging session. This is required because the area/region activated (or deactivated) during the fMRI has to be identified by transposing the BOLD images on the respective subject’s high-resolution anatomical image. Using the standard anatomical atlas (Talairache or Montreal Neurological Institute (MNI) atlas) ( www.talairach.org /daemon) and applying the x, y, z coordinates the exact standard anatomical location can be identified. It is important to ensure that a subject’s head position does not change during the entire imaging session. To compare the data between subjects, and across various labs, it is essential that the results of fMRI (activation/deactivation) are described with reference to the x, y, z coordinates and the specific standard atlas to which is being referred.

As mentioned, auditory and somatosensory activation were the early studies done with fMRI. Subsequently, more complex functions like memory, recall, emotion, anger, etc., were also studied and imaged based on the same principle. In anesthesia-related fMRI studies, memory, consciousness, and analgesia have been studied. The specific function being activated is referred to as a paradigm. To synchronize the fMRI imaging with the paradigm, a paradigm sequence is built in the imaging sequence. The imaging sequence includes the periods of no activation, followed by activation (repetition if require), and the high-resolution anatomical image called the T1 image. When the imaging commences the paradigm also gets activated.

With the advancement and better understanding of fMRI, our knowledge of cerebral physiology and energy consumption in a normally functioning brain has also improved over the years. As is well known, the human brain, which is 3 lbs (1.5 kg) and 2% of the body’s weight, gets 15% of the cardiac output as blood flow (50–55 mL/100 g/min) and consumes 20% of the total body oxygen (3–3.5 mL of O ₂ /100 g/min, compared to 3–3.5 mL of O ₂ /kg/min for the whole body). But when energy consumption of the brain is studied during a functional activation the incremental energy consumed is no more than 1%–3% over and above the baseline energy consumption. While this was certainly a surprising finding, this led the neuroscientists to conclude that the brain has no so-called “resting state” because the human brain is metabolically active all the time and consumes a lot of energy. Hence, the term “restless brain” would be more appropriate for the brain. It would also mean that any functional activity is predetermined in the brain. The neuronal circuitry for a function seems to preexist, and at the appropriate time, it gets activated. Because the human brain is very active, even in the resting state, this would be an argument to study the brain function in this resting no-task state (the restless brain). This study of the brain in the resting state is also called connectivity (discussed later in this chapter). Our human brain, unlike other organs/systems of the body, does not have specific functions isolated to specific regions of the brain, though some regions of the brain have been linked to specific functions (like motor area, speech area, auditory cortex, etc.). The brain is a highly interconnected organ. The various regions of the brain are connected to each other in a very complex manner and are constantly receiving and transmitting information to other regions. To look at it in perspective, there are 80 billion neurons in the brain and a billion billion (10 ¹⁸ ) synaptic connections. Any functional activity in the brain involves the coordinated activity of several functionally related, connected regions of the brain. Hence the term “connectivity” was coined to refer to this resting state activity of the brain.