**Intraoperative Human Functional Brain Mapping Using Optical Intrinsic Signal Imaging**

Sameer A. Sheth, Vijay Yanamadala and Emad N. Eskandar

*Department of Neurosurgery, Massachusetts General Hospital, Harvard Medical School, Boston, USA* 

#### **1. Introduction**

26 Will-be-set-by-IN-TECH

76 Advances in Brain Imaging

URL: *http://www.opticsinfobase.org/abstract.cfm?URI=BIOMED-2008-BMD9* Wolf, M., Ferrari, M. & Quaresima, V. (2007). Progress of near-infrared spectroscopy and

Zege, E. P., Ivanov, A. I. & Katsev, I. L. (1991). *Image Transfer through a Scattering Medium*,

12(6): 062104.

Springer-Verlag.

URL: *http://link.aip.org/link/?JBO/12/062104/1*

time-of-flight distributions, *Biomedical Optics*, Optical Society of America, p. BMD9.

topography for brain and muscle clinical applications, *Journal of Biomedical Optics*

Functional brain mapping strives to describe the brain's organization as a mosaic of distinct regions, each of which subserves a particular function. Advances in our understanding of functional brain organization over the past decades have been propelled by the availability of increasingly sophisticated methods for assessing various aspects of neuronal activity *in vivo*. These methods can be broadly categorized as "direct" or "indirect" measures of neuronal activity (Figure 1). Direct techniques measure changes in electromagnetic fields resulting from neuronal action potentials and synaptic activity. Indirect techniques measure changes in other tissue properties that are related to neural activity. This distinction does not imply the superiority of direct over indirect techniques. Certain disadvantages of direct measures were the very motivation for the development of indirect measures. Indeed, the most widely used functional brain imaging modality currently is functional magnetic resonance imaging (fMRI), an indirect technique. A subset of indirect techniques are based on changes in blood flow subsequent to and produced by neural activity. These perfusiondependent functional brain imaging techniques include fMRI, positron emission tomography (PET), and others. Although they are among the most commonly used methods for investigating brain function, they rely on vascular responses that are not completely understood. In this chapter, we will focus on indirect measures of brain activity, emphasizing the technique of optical intrinsic signal imaging (OISI). We discuss the physical basis of perfusion imaging and OISI, animal and human studies of OISI to date, and its potential as a powerful intraoperative functional brain mapping tool.

#### **2. Perfusion-based functional brain imaging**

In framing OISI, we first discuss the broad category of perfusion-based imaging techniques to which it belongs. Perfusion-based brain imaging techniques measure physiological events linked to neuronal activity, such as changes in metabolism or blood flow, and include positron emission tomography (PET), functional magnetic resonance imaging (fMRI), and OISI. These techniques do not measure neuronal activity *per se*; rather, they measure surrogate metabolic and vascular markers of activity. In essence, hemodynamic responses provide a map of neuronal activity spatially and temporally broadened by passage through a vascular filter. Despite their indirect nature, however, perfusion-based brain imaging

Intraoperative Human Functional Brain Mapping Using Optical Intrinsic Signal Imaging 79

neuronal activity trigger local increases in blood flow. Whether the perfusion response is necessary to supply an increased metabolic demand is under debate. This physiological blood flow response has several measurable properties that form the basis for the techniques

The idea of neurovascular coupling dates to the late 19th century. In his tome *Principles of* 

*The fluctuations of the blood-supply to the brain … followed the quickening of mental activity almost immediately. We must suppose a very delicate adjustment whereby the circulation follows the needs of the cerebral activity. Blood very likely may rush to each region of the cortex according as it is most active, but of this we know nothing. I need hardly say that the activity of the nervous matter is the* 

James draws these conclusions from the observations of the Italian scientist Mosso, who found that cerebral blood flow was redistributed based on emotional or intellectual activity. In the same year, Roy and Sherrington published their landmark study, "On the regulation of the blood supply of the brain", in which they hypothesized a connection between neuronal activity and blood flow. Decades later in 1928, Fulton (then a neurosurgical resident under Harvey Cushing, the "father" of neurosurgery) described a clinical case involving a patient with an occipital arteriovenous malformation that produced an audible bruit when the patient engaged in visual activity [1]. This finding further strengthened the case for a causal connection between brain activity and vascular responses. Important developments occurred in the 1950s, when Kety and Sokoloff pioneered methods for measuring metabolism and blood flow changes in the whole brain using radioactive tracers [2]. Ingvar and Lassen furthered these techniques by demonstrating regional blood flow changes in response to neuronal activity in humans [3]. These methods were combined with reconstruction algorithms developed for X-ray computed tomography to give rise to PET,

Although it took over half a century to appreciate the significance of the coupling between brain activity and cerebral blood flow, the last few decades have seen a considerable advancement in our understanding of its spatiotemporal dynamics. The "hemodynamic response" refers to changes in blood flow, volume, and oxygenation generated by neuronal activation. Because perfusion-based imaging modalities are based on various aspects of the hemodynamic response, their proper interpretation requires an understanding of the

Because neurons are thought to rely on oxidative metabolism for energy production, an increase in neuronal activity leads to increased oxygen consumption, which results in extraction of oxygen from the surrounding tissue and capillaries. These events take place within the first 100-300 milliseconds following activation onset. Local oxygen extraction produces a transient increase in the relative concentration of deoxyhemoglobin (Hbr) that peaks at 0.75 – 2 seconds, depending on cortex, species, and physiological condition. Neuronal activity also triggers an increase in local cerebral blood flow (CBF), via a host of

William James (1890)

*primary phenomenon, and the afflux of blood its secondary consequence.* 

the first of the modern perfusion-based brain imaging techniques.

characteristics of the response and its relationship to neuronal activity.

**4. The hemodynamic response** 

described below.

*Psychology*, William James states,

techniques are among the most commonly used, and have provided numerous important clinical and basic research insights.

Fig. 1. Categorical division of *in vivo* functional brain mapping techniques. Direct methods measure electrical and magnetic field changes produced by neuronal action potentials and synaptic potentials. Electrical measures include both non-invasive techniques such as electroencephalography (EEG) and invasive techniques such as single unit activity (SUA) and field potential (FP) recording. The most common magnetic measure is magnetoencephalography (MEG). Indirect methods, on the other hand, are sensitive to other tissue changes that accompany neural activation. Structural changes produce variations in optical properties that follow the millisecond timecourse of neural events. Techniques taking advantage of these processes include voltage sensitive dye (VSD) imaging, which measures transmembrane voltage changes, and event-related optical signal (EROS) imaging, which measures optical scattering changes putatively produced by ionic movement. Indirect assessment is also possible using changes in blood flow elicited by neural activity. Hemodynamic events follow a much slower time course (several seconds), but are the most commonly used signals for functional imaging. A full appreciation of their importance for basic neurophysiology and functional imaging requires a detailed understanding of neurovascular coupling, or the relationship between neuronal activity and perfusion. PET, positron emission tomography; fMRI, functional magnetic resonance imaging; OISI, optical intrinsic signal imaging; NIRS, near-infrared spectroscopy.

#### **3. Neurovascular coupling**

Understanding the capacities and limitations of perfusion-based techniques requires an appreciation of the relationship between changes in neuronal activity and blood flow. In brief, the concept of "neurovascular coupling" describes the observation that increases in neuronal activity trigger local increases in blood flow. Whether the perfusion response is necessary to supply an increased metabolic demand is under debate. This physiological blood flow response has several measurable properties that form the basis for the techniques described below.

The idea of neurovascular coupling dates to the late 19th century. In his tome *Principles of Psychology*, William James states,

*The fluctuations of the blood-supply to the brain … followed the quickening of mental activity almost immediately. We must suppose a very delicate adjustment whereby the circulation follows the needs of the cerebral activity. Blood very likely may rush to each region of the cortex according as it is most active, but of this we know nothing. I need hardly say that the activity of the nervous matter is the primary phenomenon, and the afflux of blood its secondary consequence.* 

William James (1890)

James draws these conclusions from the observations of the Italian scientist Mosso, who found that cerebral blood flow was redistributed based on emotional or intellectual activity. In the same year, Roy and Sherrington published their landmark study, "On the regulation of the blood supply of the brain", in which they hypothesized a connection between neuronal activity and blood flow. Decades later in 1928, Fulton (then a neurosurgical resident under Harvey Cushing, the "father" of neurosurgery) described a clinical case involving a patient with an occipital arteriovenous malformation that produced an audible bruit when the patient engaged in visual activity [1]. This finding further strengthened the case for a causal connection between brain activity and vascular responses. Important developments occurred in the 1950s, when Kety and Sokoloff pioneered methods for measuring metabolism and blood flow changes in the whole brain using radioactive tracers [2]. Ingvar and Lassen furthered these techniques by demonstrating regional blood flow changes in response to neuronal activity in humans [3]. These methods were combined with reconstruction algorithms developed for X-ray computed tomography to give rise to PET, the first of the modern perfusion-based brain imaging techniques.

#### **4. The hemodynamic response**

78 Advances in Brain Imaging

techniques are among the most commonly used, and have provided numerous important

Fig. 1. Categorical division of *in vivo* functional brain mapping techniques. Direct methods measure electrical and magnetic field changes produced by neuronal action potentials and synaptic potentials. Electrical measures include both non-invasive techniques such as electroencephalography (EEG) and invasive techniques such as single unit activity (SUA)

magnetoencephalography (MEG). Indirect methods, on the other hand, are sensitive to other tissue changes that accompany neural activation. Structural changes produce variations in optical properties that follow the millisecond timecourse of neural events. Techniques taking advantage of these processes include voltage sensitive dye (VSD) imaging, which measures transmembrane voltage changes, and event-related optical signal (EROS) imaging, which measures optical scattering changes putatively produced by ionic movement. Indirect

Hemodynamic events follow a much slower time course (several seconds), but are the most commonly used signals for functional imaging. A full appreciation of their importance for basic neurophysiology and functional imaging requires a detailed understanding of neurovascular coupling, or the relationship between neuronal activity and perfusion. PET, positron emission tomography; fMRI, functional magnetic resonance imaging; OISI, optical

Understanding the capacities and limitations of perfusion-based techniques requires an appreciation of the relationship between changes in neuronal activity and blood flow. In brief, the concept of "neurovascular coupling" describes the observation that increases in

and field potential (FP) recording. The most common magnetic measure is

assessment is also possible using changes in blood flow elicited by neural activity.

intrinsic signal imaging; NIRS, near-infrared spectroscopy.

**3. Neurovascular coupling** 

clinical and basic research insights.

Although it took over half a century to appreciate the significance of the coupling between brain activity and cerebral blood flow, the last few decades have seen a considerable advancement in our understanding of its spatiotemporal dynamics. The "hemodynamic response" refers to changes in blood flow, volume, and oxygenation generated by neuronal activation. Because perfusion-based imaging modalities are based on various aspects of the hemodynamic response, their proper interpretation requires an understanding of the characteristics of the response and its relationship to neuronal activity.

Because neurons are thought to rely on oxidative metabolism for energy production, an increase in neuronal activity leads to increased oxygen consumption, which results in extraction of oxygen from the surrounding tissue and capillaries. These events take place within the first 100-300 milliseconds following activation onset. Local oxygen extraction produces a transient increase in the relative concentration of deoxyhemoglobin (Hbr) that peaks at 0.75 – 2 seconds, depending on cortex, species, and physiological condition. Neuronal activity also triggers an increase in local cerebral blood flow (CBF), via a host of

Intraoperative Human Functional Brain Mapping Using Optical Intrinsic Signal Imaging 81

Although general aspects of the hemodynamic response have been fairly well characterized, relatively little is known about the mediators of the response [39]. Hypotheses explaining the molecular communication between neural activity and blood flow fall roughly into four groups [40]. According to one, byproducts of neuronal activity such as adenosine and K+ released extracellularly cause vasodilatation in nearby blood vessels [41]. A second hypothesis promotes the role of nitric oxide (NO). In this scenario, glutamate, the predominant excitatory neurotransmitter, also acts as a mediator of neurovascular coupling. Glutamate released into the synapse binds to postsynaptic NMDA receptors and activates guanylyl cyclase to produce NO, which diffuses to neighboring blood vessels and causes vasodilatation [42]. According to a third line of thought, neuronal processes directly innervate local blood vessels and transmit vasodilatory signals through acetylcholine [43], dopamine [44], or serotonin [45]. The final and most recent argument also implicates glutamate, but highlights the role of astrocytes. Astrocytic processes ensheathing the synapse sense glutamate levels through metabotropic receptors that produce graded intracellular Ca2+ increases. The Ca2+ wave propagates to the astrocyte endfeet, which are in intimate contact with blood vessels, and triggers vasodilatation through release of various

Historically, the most commonly used perfusion-based functional brain mapping techniques have been positron emission tomography (PET) and functional magnetic resonance imaging (fMRI). The latter has experienced tremendous growth over the last ten years, eclipsing all other brain imaging techniques [47]. The demand for high-resolution imaging has also brought OISI to the forefront, along with its lower resolution but noninvasive cousin, near-

Each of these imaging modalities is based on one or more aspects of the hemodynamic response. The following discussion introduces the development, basic methodology, and relative advantages and disadvantages of these techniques, with particular emphasis given

Throughout the 1950s and 1960s, radionuclide scans were a popular tool for neurodiagnostics. The application of tomographic reconstruction techniques developed for X-ray computerized tomography (CT) to nuclear medicine heralded the rise of positron emission tomography (PET) and single photon emission computed tomography (SPECT). Introduced in 1975 by Michael Phelps [48], PET imaging was the first non-invasive (or at least minimally invasive) perfusion-based functional brain imaging modality. This technological breakthrough allowed autoradiographic measurement of metabolism [49] and blood flow [2], previously restricted to animals, to be performed in humans in the form of *in vivo* autoradiograms. The development of several biologically useful positron-emitting

PET imaging relies on the use of radioactive atoms decaying by emitting positrons, which have the same mass as electrons but a positive charge. These positron-emitting atoms are generated either through the decay of another generating element (e.g., 68Ga through the decay of 68Ge), or by direct production in a cyclotron (e.g., 15O). Molecules containing these radioactive elements are chemically synthesized, and trace amounts injected into the subject.

molecules rapidly increased the utility of PET through the late 1970s and 1980s.

molecules including prostanoids [46].

to OISI and related optical techniques.

**4.1 Positron emission tomography (PET)** 

infrared spectroscopy (NIRS).

possible mediators and incompletely understood mechanisms. The CBF response begins near the site of neural activity in small arterioles, the primary resistance vessels, and propagates retrograde to larger vessels, peaking at 2-4 seconds. The influx of oxygenated arterial blood rapidly reverses local tissue oxygenation, decreasing Hbr and increasing oxyhemoglobin (HbO2). This oxygenation change greatly overshoots baseline, resulting in relative hyperoxygenation, peaking at 3-6 seconds. As the inflowing blood drains into the venous system, these later oxygenation changes occur in medium to large veins. If neuronal activation lasts fewer than ~4 seconds, the oxygenation changes gradually return to baseline over 10-20 seconds. If activation lasts longer, blood volume and oxygenation remain elevated for the duration of stimulation at a lower "plateau" level, and then return to baseline over several seconds after stimulus offset.

The reliance of neurons on oxidative metabolism mentioned above is a recently developed hypothesis. In the late 1980s Fox and Raichle found that relative increases in CBF were six times greater than increases in oxygen utilization [5-6]. The authors suggested that this mismatch represented uncoupling between CBF and oxygen metabolism, which they interpreted as suggesting neurons use means other than oxygen metabolism to support their energy demand. For several years, a glycoltyic mechanism for ATP production was thought to support neuronal activity. In the late 1990s, however, Buxton and Frank proposed a biophysical model of the hemodynamic response that accounted for these measurements [7- 9]. They suggested that the capacity for oxygen diffusion from capillary red blood cells to neurons was reduced during the CBF response due to a decrease in capillary transit time. Blood flowing more quickly had less time to exchange oxygen with the tissue. According to this "oxygen limitation" model, a relatively larger increase in CBF was necessary to compensate for the reduced oxygen extraction, and that Fox and Raichle's data actually supported tight coupling between CBF and oxygen metabolism. This model supported the notion that oxidative metabolism accounted for the bulk of neuronal energy production. Studies using MR spectroscopy [10] and fMRI [11] provided experimental validation of this theory. In the last few years, a number of additional studies have demonstrated focal decreases in tissue oxygenation and increases in oxygen metabolism rate [12-15].

The debate over the physiology of energy metabolism has occurred almost in parallel with the debate over its implication for perfusion-based functional imaging. Some have suggested that a local increase in Hbr due to decreased tissue oxygenation would generate a transient drop in blood oxygen-level dependent (BOLD) signal ("initial dip" in fMRI parlance) [16-17]. Its close relationship to neuronal metabolism would putatively allow the initial dip to serve as a spatially accurate mapping signal for functional imaging. Although several studies have observed the dip with BOLD fMRI [18-27] and OIS spectroscopy [13, 28-31], others have not [32-36], and its existence remains somewhat controversial. Some have suggested that slightly different spatiotemporal dynamics between cerebral blood flow (CBF) and volume (CBV) could create an initial dip in BOLD signal without an increase in oxygen metabolism [9, 37]. The observation that CBV increases at least 0.5 seconds after tissue pO2 begins to drop [15], however, makes that possibility unlikely. There now seems to be general agreement that the dip exists and is related to increased oxygen consumption. Its small size [38] and susceptibility to baseline physiological conditions [7, 32] however, jeopardize its usefulness as a mapping signal.

possible mediators and incompletely understood mechanisms. The CBF response begins near the site of neural activity in small arterioles, the primary resistance vessels, and propagates retrograde to larger vessels, peaking at 2-4 seconds. The influx of oxygenated arterial blood rapidly reverses local tissue oxygenation, decreasing Hbr and increasing oxyhemoglobin (HbO2). This oxygenation change greatly overshoots baseline, resulting in relative hyperoxygenation, peaking at 3-6 seconds. As the inflowing blood drains into the venous system, these later oxygenation changes occur in medium to large veins. If neuronal activation lasts fewer than ~4 seconds, the oxygenation changes gradually return to baseline over 10-20 seconds. If activation lasts longer, blood volume and oxygenation remain elevated for the duration of stimulation at a lower "plateau" level, and then return to

The reliance of neurons on oxidative metabolism mentioned above is a recently developed hypothesis. In the late 1980s Fox and Raichle found that relative increases in CBF were six times greater than increases in oxygen utilization [5-6]. The authors suggested that this mismatch represented uncoupling between CBF and oxygen metabolism, which they interpreted as suggesting neurons use means other than oxygen metabolism to support their energy demand. For several years, a glycoltyic mechanism for ATP production was thought to support neuronal activity. In the late 1990s, however, Buxton and Frank proposed a biophysical model of the hemodynamic response that accounted for these measurements [7- 9]. They suggested that the capacity for oxygen diffusion from capillary red blood cells to neurons was reduced during the CBF response due to a decrease in capillary transit time. Blood flowing more quickly had less time to exchange oxygen with the tissue. According to this "oxygen limitation" model, a relatively larger increase in CBF was necessary to compensate for the reduced oxygen extraction, and that Fox and Raichle's data actually supported tight coupling between CBF and oxygen metabolism. This model supported the notion that oxidative metabolism accounted for the bulk of neuronal energy production. Studies using MR spectroscopy [10] and fMRI [11] provided experimental validation of this theory. In the last few years, a number of additional studies have demonstrated focal

decreases in tissue oxygenation and increases in oxygen metabolism rate [12-15].

The debate over the physiology of energy metabolism has occurred almost in parallel with the debate over its implication for perfusion-based functional imaging. Some have suggested that a local increase in Hbr due to decreased tissue oxygenation would generate a transient drop in blood oxygen-level dependent (BOLD) signal ("initial dip" in fMRI parlance) [16-17]. Its close relationship to neuronal metabolism would putatively allow the initial dip to serve as a spatially accurate mapping signal for functional imaging. Although several studies have observed the dip with BOLD fMRI [18-27] and OIS spectroscopy [13, 28-31], others have not [32-36], and its existence remains somewhat controversial. Some have suggested that slightly different spatiotemporal dynamics between cerebral blood flow (CBF) and volume (CBV) could create an initial dip in BOLD signal without an increase in oxygen metabolism [9, 37]. The observation that CBV increases at least 0.5 seconds after tissue pO2 begins to drop [15], however, makes that possibility unlikely. There now seems to be general agreement that the dip exists and is related to increased oxygen consumption. Its small size [38] and susceptibility to baseline physiological conditions [7, 32] however,

baseline over several seconds after stimulus offset.

jeopardize its usefulness as a mapping signal.

Although general aspects of the hemodynamic response have been fairly well characterized, relatively little is known about the mediators of the response [39]. Hypotheses explaining the molecular communication between neural activity and blood flow fall roughly into four groups [40]. According to one, byproducts of neuronal activity such as adenosine and K+ released extracellularly cause vasodilatation in nearby blood vessels [41]. A second hypothesis promotes the role of nitric oxide (NO). In this scenario, glutamate, the predominant excitatory neurotransmitter, also acts as a mediator of neurovascular coupling. Glutamate released into the synapse binds to postsynaptic NMDA receptors and activates guanylyl cyclase to produce NO, which diffuses to neighboring blood vessels and causes vasodilatation [42]. According to a third line of thought, neuronal processes directly innervate local blood vessels and transmit vasodilatory signals through acetylcholine [43], dopamine [44], or serotonin [45]. The final and most recent argument also implicates glutamate, but highlights the role of astrocytes. Astrocytic processes ensheathing the synapse sense glutamate levels through metabotropic receptors that produce graded intracellular Ca2+ increases. The Ca2+ wave propagates to the astrocyte endfeet, which are in intimate contact with blood vessels, and triggers vasodilatation through release of various molecules including prostanoids [46].

Historically, the most commonly used perfusion-based functional brain mapping techniques have been positron emission tomography (PET) and functional magnetic resonance imaging (fMRI). The latter has experienced tremendous growth over the last ten years, eclipsing all other brain imaging techniques [47]. The demand for high-resolution imaging has also brought OISI to the forefront, along with its lower resolution but noninvasive cousin, nearinfrared spectroscopy (NIRS).

Each of these imaging modalities is based on one or more aspects of the hemodynamic response. The following discussion introduces the development, basic methodology, and relative advantages and disadvantages of these techniques, with particular emphasis given to OISI and related optical techniques.

#### **4.1 Positron emission tomography (PET)**

Throughout the 1950s and 1960s, radionuclide scans were a popular tool for neurodiagnostics. The application of tomographic reconstruction techniques developed for X-ray computerized tomography (CT) to nuclear medicine heralded the rise of positron emission tomography (PET) and single photon emission computed tomography (SPECT). Introduced in 1975 by Michael Phelps [48], PET imaging was the first non-invasive (or at least minimally invasive) perfusion-based functional brain imaging modality. This technological breakthrough allowed autoradiographic measurement of metabolism [49] and blood flow [2], previously restricted to animals, to be performed in humans in the form of *in vivo* autoradiograms. The development of several biologically useful positron-emitting molecules rapidly increased the utility of PET through the late 1970s and 1980s.

PET imaging relies on the use of radioactive atoms decaying by emitting positrons, which have the same mass as electrons but a positive charge. These positron-emitting atoms are generated either through the decay of another generating element (e.g., 68Ga through the decay of 68Ge), or by direct production in a cyclotron (e.g., 15O). Molecules containing these radioactive elements are chemically synthesized, and trace amounts injected into the subject.

Intraoperative Human Functional Brain Mapping Using Optical Intrinsic Signal Imaging 83

required only 11. This surge in popularity is a consequence of the technique's noninvasiveness and balance between sensitivity and resolution. In addition, many types of studies can be performed in clinical MRI magnets, which are increasingly available in modern health care facilities. Over the past decade, great progress has been made in improving the versatility, spatial and temporal resolution, and clinical utility of fMRI.

The phenomenon of nuclear magnetic resonance (NMR) was observed independently by Bloch and Purcell in 1945, a finding for which they were jointly awarded the Nobel Prize in Physics in 1952. NMR describes the behavior of atomic nuclei in magnetic fields. Because nuclei possess both charge and spin, they also possess an intrinsic magnetic moment. When

anti-parallel to the field and precess about the field like a gyroscope precessing in the Earth's gravitational field. If molar quantities of spinning atoms are considered, a small

 . The resonance phenomenon dictates that externally applied energy at this frequency will be transferred to the spinning atoms. Because the Larmor frequency is in the radio wave spectrum, the applied energy is called the radiofrequency (RF) pulse. The RF

still precessing, the time varying magnetic field they produce in the transverse plane induces current in a coil of wire oriented perpendicular to it. Thus transverse magnetization

This signal decays by two processes: longitudinal (spin-lattice) and transverse (spin-spin)

spins due to precession at different rates, and occurs with a time constant T2. This difference arises from spins experiencing slightly different magnetic field strengths, caused either by interactions with other spins (pure T2 effects) or by small field inhomogeneities (T2\* effects).

Imaging based on NMR involves spatially encoding the position of different spins using magnetic field gradients. Whereas NMR can be performed with any atom possessing non zero spin, imaging applications prefer spin ½ nuclei because they have only two possible energy levels. Several spin ½ atoms have been used for imaging and spectroscopy applications, including 3He [50], 13C [51], 31P [52], and 129Xe [53], but by far the most

Biological imaging using NMR was developed in the late 1970s, with the first human images appearing in 1977 [54]. The word "nuclear" was dropped in the mid 1980s to avoid the associated negative connotation, in favor of the term magnetic resonance imaging (MRI). In the early 1990s, Belliveau, Rosen, and colleagues performed the first functional studies using

Variations in the latter provide contrast for the most commonly used form of fMRI.

common is 1H because of its abundance in the form of H2O in biological material.

, and occurs with a characteristic time constant T1. The second describes dephasing of

and the atom's inherent properties (called its gryomagnetic constant,

Precession occurs at a characteristic frequency (the Larmor frequency,

), nuclei align their magnetic moments parallel or

parallel

):

), which is a

back to its equilibrium position, parallel

, resulting in a net magnetization vector *<sup>M</sup>*

, rotating it towards the transverse plane. Because the spins are

placed in an external magnetic field ( *B*

fraction more tend to align parallel to *B*

to *B* .

 *B*

to *B*

function of *B*

energy exerts a torque on *M*

produces a detectable signal.

relaxation. The first describes the relaxation of *M*

The labeled molecules circulate according to their biological properties, all the while emitting positrons. The emitted positrons travel a short distance (on the order of a few mm), until they encounter their anti-particles, electrons, with which they annihilate. The annihilation produces a characteristic release of so-called synchrotron energy in the form of two 511 keV gamma rays, which travel in opposite directions, close to 180º apart.

Because of the gamma rays' high energy, PET system detectors employ elements with high atomic number and therefore high stopping power. The original detectors used thalliumdoped sodium iodide (NaI[Tl]) crystals, but modern scanners use either bismuth germinate (BGO; Bi4Ge3O12) or lutetium oxyorthosilicate (LSO). The detector electronics use a coincidence detection system that reduces background radiation and scatter by rejecting events that are not recorded almost simultaneously (within the preset coincidence time window) on both sides of the head. The source is assumed to exist on a line connecting the two detectors that recorded the events, and its position is determined based on the coincidence time difference. Events are recorded in this manner, and the source distribution within the brain is then calculated by solving the inverse problem using filtered backprojection or iterative approaches.

One of the advantages of PET imaging is the diversity of metabolic, hemodynamic, and biochemical processes that can be assessed using different tracers. Those that are most relevant for hemodynamic functional brain mapping are 15O-labelled water (H2 15O) and 18Flabelled 2-fluoro-2-deoxy-D-glucose (18FDG). H2 15O distributes within the circulation and collects proportionally to the regional blood flow. H215O PET is therefore the most common method for non-invasive cerebral blood flow (CBF) measurement. 18FDG is actively transported into cells via glucose transporters, where it is phosphorylated in the cytoplasm and thereby sequestered intracellularly. Its 2-fluoro group prevents it from undergoing further glycolysis, so the amount accumulated within the cell indicates the rate of glucose metabolism.

More than 25 years since its introduction, PET imaging is still an important and widely used technique for clinical diagnosis and basic science research. PET offers versatility for measuring a range of physiological processes and the ability to quantify them in absolute terms. Its reliance on radiopharmaceuticals, however, limits the potential subject population and number of studies that can be performed in the same subject. In addition, PET affords rather poor spatial (several millimeters) and temporal (10s of seconds to minutes) resolution.

SPECT imaging is similar to PET in its reliance on exogenous radioisotope contrast agents. SPECT radioisotopes undergo gamma decay by emitting a single high-energy photon, as its name implies, as opposed to the two gamma rays produced by positron-electron annihilation in PET. As in PET, the contrast agents can be chosen to measure physiological responses such as CBF (e.g., 99mTc-hexamethylpropyleneamineoxime [HMPAO]). Many of the radiopharmaceuticals are commonly used in nuclear medicine and therefore do not require a cyclotron for their production. Factors such as these make SPECT imaging significantly cheaper than PET, although it suffers from even lower spatial resolution.

#### **4.2 Functional magnetic resonance imaging (fMRI)**

In 2001, fMRI surpassed EEG as the most widely used brain imaging technique, in terms of the number of published papers. Whereas EEG required 72 years to reach this level, fMRI

The labeled molecules circulate according to their biological properties, all the while emitting positrons. The emitted positrons travel a short distance (on the order of a few mm), until they encounter their anti-particles, electrons, with which they annihilate. The annihilation produces a characteristic release of so-called synchrotron energy in the form of

Because of the gamma rays' high energy, PET system detectors employ elements with high atomic number and therefore high stopping power. The original detectors used thalliumdoped sodium iodide (NaI[Tl]) crystals, but modern scanners use either bismuth germinate (BGO; Bi4Ge3O12) or lutetium oxyorthosilicate (LSO). The detector electronics use a coincidence detection system that reduces background radiation and scatter by rejecting events that are not recorded almost simultaneously (within the preset coincidence time window) on both sides of the head. The source is assumed to exist on a line connecting the two detectors that recorded the events, and its position is determined based on the coincidence time difference. Events are recorded in this manner, and the source distribution within the brain is then calculated by solving the inverse problem using filtered back-

One of the advantages of PET imaging is the diversity of metabolic, hemodynamic, and biochemical processes that can be assessed using different tracers. Those that are most relevant for hemodynamic functional brain mapping are 15O-labelled water (H215O) and 18Flabelled 2-fluoro-2-deoxy-D-glucose (18FDG). H215O distributes within the circulation and collects proportionally to the regional blood flow. H215O PET is therefore the most common method for non-invasive cerebral blood flow (CBF) measurement. 18FDG is actively transported into cells via glucose transporters, where it is phosphorylated in the cytoplasm and thereby sequestered intracellularly. Its 2-fluoro group prevents it from undergoing further glycolysis, so the amount accumulated within the cell indicates the rate of glucose

More than 25 years since its introduction, PET imaging is still an important and widely used technique for clinical diagnosis and basic science research. PET offers versatility for measuring a range of physiological processes and the ability to quantify them in absolute terms. Its reliance on radiopharmaceuticals, however, limits the potential subject population and number of studies that can be performed in the same subject. In addition, PET affords rather poor spatial (several millimeters) and temporal (10s of seconds to minutes) resolution. SPECT imaging is similar to PET in its reliance on exogenous radioisotope contrast agents. SPECT radioisotopes undergo gamma decay by emitting a single high-energy photon, as its name implies, as opposed to the two gamma rays produced by positron-electron annihilation in PET. As in PET, the contrast agents can be chosen to measure physiological responses such as CBF (e.g., 99mTc-hexamethylpropyleneamineoxime [HMPAO]). Many of the radiopharmaceuticals are commonly used in nuclear medicine and therefore do not require a cyclotron for their production. Factors such as these make SPECT imaging significantly cheaper than PET, although it suffers from even lower spatial resolution.

In 2001, fMRI surpassed EEG as the most widely used brain imaging technique, in terms of the number of published papers. Whereas EEG required 72 years to reach this level, fMRI

two 511 keV gamma rays, which travel in opposite directions, close to 180º apart.

projection or iterative approaches.

**4.2 Functional magnetic resonance imaging (fMRI)** 

metabolism.

required only 11. This surge in popularity is a consequence of the technique's noninvasiveness and balance between sensitivity and resolution. In addition, many types of studies can be performed in clinical MRI magnets, which are increasingly available in modern health care facilities. Over the past decade, great progress has been made in improving the versatility, spatial and temporal resolution, and clinical utility of fMRI.

The phenomenon of nuclear magnetic resonance (NMR) was observed independently by Bloch and Purcell in 1945, a finding for which they were jointly awarded the Nobel Prize in Physics in 1952. NMR describes the behavior of atomic nuclei in magnetic fields. Because nuclei possess both charge and spin, they also possess an intrinsic magnetic moment. When placed in an external magnetic field ( *B* ), nuclei align their magnetic moments parallel or anti-parallel to the field and precess about the field like a gyroscope precessing in the Earth's gravitational field. If molar quantities of spinning atoms are considered, a small fraction more tend to align parallel to *B* , resulting in a net magnetization vector *<sup>M</sup>* parallel to *B* .

Precession occurs at a characteristic frequency (the Larmor frequency, ), which is a function of *B* and the atom's inherent properties (called its gryomagnetic constant, ): *B* . The resonance phenomenon dictates that externally applied energy at this frequency will be transferred to the spinning atoms. Because the Larmor frequency is in the radio wave spectrum, the applied energy is called the radiofrequency (RF) pulse. The RF energy exerts a torque on *M* , rotating it towards the transverse plane. Because the spins are still precessing, the time varying magnetic field they produce in the transverse plane induces current in a coil of wire oriented perpendicular to it. Thus transverse magnetization produces a detectable signal.

This signal decays by two processes: longitudinal (spin-lattice) and transverse (spin-spin) relaxation. The first describes the relaxation of *M* back to its equilibrium position, parallel to *B* , and occurs with a characteristic time constant T1. The second describes dephasing of spins due to precession at different rates, and occurs with a time constant T2. This difference arises from spins experiencing slightly different magnetic field strengths, caused either by interactions with other spins (pure T2 effects) or by small field inhomogeneities (T2\* effects). Variations in the latter provide contrast for the most commonly used form of fMRI.

Imaging based on NMR involves spatially encoding the position of different spins using magnetic field gradients. Whereas NMR can be performed with any atom possessing non zero spin, imaging applications prefer spin ½ nuclei because they have only two possible energy levels. Several spin ½ atoms have been used for imaging and spectroscopy applications, including 3He [50], 13C [51], 31P [52], and 129Xe [53], but by far the most common is 1H because of its abundance in the form of H2O in biological material.

Biological imaging using NMR was developed in the late 1970s, with the first human images appearing in 1977 [54]. The word "nuclear" was dropped in the mid 1980s to avoid the associated negative connotation, in favor of the term magnetic resonance imaging (MRI). In the early 1990s, Belliveau, Rosen, and colleagues performed the first functional studies using

Intraoperative Human Functional Brain Mapping Using Optical Intrinsic Signal Imaging 85

64]. Spins outside the region of interest are labeled with an RF pulse. Labeled spins are allowed to enter the imaging region after a suitable time delay (0.5 – 2 s), where they exchange with tissue water. The amount of signal detected in the imaging slice is proportional to the flow rate into the slice. Advantages of CBF fMRI include better spatial specificity than BOLD, with less emphasis on large draining veins. In addition, whereas BOLD signals include a complex mixture of CBV, CBF, and oxygenation contributions, CBF fMRI can isolate and quantify the CBF component. This technique suffers, however, from relatively poor temporal resolution (several seconds) and lower sensitivity than BOLD [65]. The original reports of functional imaging using MRI used CBV contrast [55-57]. Current CBV fMRI studies use exogenous paramagnetic contrast agents with a long blood half-life, avoiding the need for kinetic tracer models and allowing repeated imaging. The large magnetic field disruptions introduced by the contrast agents lead to decreased T2\*-weighted signals in proportion to CBV. The temporal resolution of CBV fMRI is similar to that of BOLD fMRI, and its contrast-to-noise ratio (CNR) is much higher. The major drawback is the requirement for contrast injection, which also precludes human studies, although clinical

Intraoperative OISI maps the brain by measuring activity-related changes in cortical light reflectance. Activity related reflectance changes were first demonstrated in nervous tissue in vitro more than 50 years ago [67] and have since been observed in vivo in rodents, cats, nonhuman primates, and humans. It is a particularly attractive brain-mapping modality because it can rapidly assess the functional activity of large cortical areas with very high spatial resolution (50–100 µm). Because of its versatility, OISI has been used to characterize numerous physiological phenomena, including neurovascular coupling [67-69], hemodynamic refractory periods [70], vasomotion [71], the organization of the visual cortex [71-73], cortical plasticity [71-76], cortical spreading depression [77], seizure [78-79], and language organization in the human brain [80]. Haglund and colleagues [81] were the first to observe optical signals in humans during seizure and cognitive tasks. Since then, the authors of studies on intraoperative OISI have described optical signal evolution in human cortex [82], the mapping of primary sensory and motor cortices [70], and the delineation of

OISI detects perfusion-related and metabolic signals that are coupled to neuronal activity, including hemoglobin concentration and oxygenation changes, cytochrome oxidation changes, and light scattering caused by altered blood volume, blood flow, and cell swelling [15, 17, 30, 84-86], which in turn create a functional map of the brain. Each of these different phenomena is observed and best quantified at different imaging wavelengths [30, 87]. For example, imaging at 610 nm best detects deoxyhemoglobin concentration changes because the absorbance of oxyhemoglobin is much less than that of deoxyhemoglobin at 610 nm [30]. Thus, OISI at 610 nm is analagous to BOLD fMRI [20], which capitalizes on local magnetic susceptibility changes due to differences between deoxyhemoglobin and oxyhemoglobin

trials are underway [66].

**5.1 Intrinsic optical signals** 

concentration.

**5. Optical intrinsic signal imaging (OISI)** 

language cortices within [83] and across languages [80].

an injected paramagnetic contrast agent [55-57]. Contrast agents with high magnetic susceptibility produce large magnetic field gradients in the local environment, decreasing T2\* in proportion to the amount present. Using tracer kinetic models, they compared the decrease in T2\* weighted signal during visual activity and rest and generated a CBV weighted functional image of visual cortex [55].

At the same time, Ogawa and colleagues showed in animals that functional images could be generated without an exogenous contrast agent [58]. They took advantage of the natural difference in magnetic properties between oxy- (HbO2) and deoxyhemoglobin (Hbr). Whereas the former is weakly diamagnetic and has little effect on magnetic fields, the latter is paramagnetic and causes local field disturbances. Hbr therefore acts as an endogenous contrast agent. The CBF response produced by functional activity reduces local Hbr content and therefore increases signal strength.

These lines of research converged in 1992, when these two groups and a third almost simultaneously demonstrated that this intrinsic oxygenation-based contrast could be used to map brain activity [59-61]. This so-called blood oxygenation level dependent (BOLD) contrast is by far the most common in fMRI studies. BOLD fMRI has spearheaded the surge in functional brain imaging over the last decade, surpassing PET due to its superior spatial and temporal resolution and avoidance of exogenous radioactive tracers.

Because it is based on blood Hbr content, BOLD fMRI temporal characteristics closely follow those of oxygenation changes, starting within 1-2 seconds of stimulation onset and peaking at 4-6 seconds. Spatially, it emphasizes venous structures because oxygenation changes are most prominent in medium to large veins. This bias tends to decrease the spatial specificity of conventional T2\* BOLD fMRI somewhat, as veins are often millimeters away from the neuronal areas they drain. Spatial precision of a few millimeters may be sufficient for many types of cognitive studies investigating the entire brain with moderate resolution (and is certainly superior to that of PET, EEG, or MEG), but it is not sufficient for high resolution studies of columnar functional architecture.

In 1996, Malonek and Grinvald observed a small increase in Hbr before the CBF-induced hyperoxygenation [17]. They attributed this transient deoxygenation to an increase in oxidative metabolism that decreased local tissue oxygen tension before the onset of the CBF response. They suggested that imaging based on the dip could improve spatial specificity, since it was restricted to metabolically active areas.

A transient increase in Hbr would appear as a brief decrease in BOLD signal prior to the conventional positive BOLD response. Over the next several years, many investigators looked for this "initial dip" in a variety of cortices and species. It has been successfully identified in the visual cortex of cat [22-23], monkey [21], and human [18-20, 24-27], and human motor cortex [25, 27]. Although these studies identified the dip in the BOLD signal timecourse, only one study has been able to generate a map using this signal [23]. Other groups have tried, but found that the maps were not reproducible [62]. Indeed, several fMRI studies in rat somatosensory cortex have been unable to detect the dip altogether [34-36]. These discrepancies may be due to anesthesia, differences in cortical architecture, or other effects.

Alternatives to BOLD fMRI include CBF- (or perfusion-) weighted fMRI and CBV-weighted fMRI. In CBF fMRI, 1H spins in arterial blood water are used as endogenous flow tracers [6364]. Spins outside the region of interest are labeled with an RF pulse. Labeled spins are allowed to enter the imaging region after a suitable time delay (0.5 – 2 s), where they exchange with tissue water. The amount of signal detected in the imaging slice is proportional to the flow rate into the slice. Advantages of CBF fMRI include better spatial specificity than BOLD, with less emphasis on large draining veins. In addition, whereas BOLD signals include a complex mixture of CBV, CBF, and oxygenation contributions, CBF fMRI can isolate and quantify the CBF component. This technique suffers, however, from relatively poor temporal resolution (several seconds) and lower sensitivity than BOLD [65].

The original reports of functional imaging using MRI used CBV contrast [55-57]. Current CBV fMRI studies use exogenous paramagnetic contrast agents with a long blood half-life, avoiding the need for kinetic tracer models and allowing repeated imaging. The large magnetic field disruptions introduced by the contrast agents lead to decreased T2\*-weighted signals in proportion to CBV. The temporal resolution of CBV fMRI is similar to that of BOLD fMRI, and its contrast-to-noise ratio (CNR) is much higher. The major drawback is the requirement for contrast injection, which also precludes human studies, although clinical trials are underway [66].

#### **5. Optical intrinsic signal imaging (OISI)**

84 Advances in Brain Imaging

an injected paramagnetic contrast agent [55-57]. Contrast agents with high magnetic susceptibility produce large magnetic field gradients in the local environment, decreasing T2\* in proportion to the amount present. Using tracer kinetic models, they compared the decrease in T2\* weighted signal during visual activity and rest and generated a CBV

At the same time, Ogawa and colleagues showed in animals that functional images could be generated without an exogenous contrast agent [58]. They took advantage of the natural difference in magnetic properties between oxy- (HbO2) and deoxyhemoglobin (Hbr). Whereas the former is weakly diamagnetic and has little effect on magnetic fields, the latter is paramagnetic and causes local field disturbances. Hbr therefore acts as an endogenous contrast agent. The CBF response produced by functional activity reduces local Hbr content

These lines of research converged in 1992, when these two groups and a third almost simultaneously demonstrated that this intrinsic oxygenation-based contrast could be used to map brain activity [59-61]. This so-called blood oxygenation level dependent (BOLD) contrast is by far the most common in fMRI studies. BOLD fMRI has spearheaded the surge in functional brain imaging over the last decade, surpassing PET due to its superior spatial

Because it is based on blood Hbr content, BOLD fMRI temporal characteristics closely follow those of oxygenation changes, starting within 1-2 seconds of stimulation onset and peaking at 4-6 seconds. Spatially, it emphasizes venous structures because oxygenation changes are most prominent in medium to large veins. This bias tends to decrease the spatial specificity of conventional T2\* BOLD fMRI somewhat, as veins are often millimeters away from the neuronal areas they drain. Spatial precision of a few millimeters may be sufficient for many types of cognitive studies investigating the entire brain with moderate resolution (and is certainly superior to that of PET, EEG, or MEG), but it is not sufficient for high resolution

In 1996, Malonek and Grinvald observed a small increase in Hbr before the CBF-induced hyperoxygenation [17]. They attributed this transient deoxygenation to an increase in oxidative metabolism that decreased local tissue oxygen tension before the onset of the CBF response. They suggested that imaging based on the dip could improve spatial specificity,

A transient increase in Hbr would appear as a brief decrease in BOLD signal prior to the conventional positive BOLD response. Over the next several years, many investigators looked for this "initial dip" in a variety of cortices and species. It has been successfully identified in the visual cortex of cat [22-23], monkey [21], and human [18-20, 24-27], and human motor cortex [25, 27]. Although these studies identified the dip in the BOLD signal timecourse, only one study has been able to generate a map using this signal [23]. Other groups have tried, but found that the maps were not reproducible [62]. Indeed, several fMRI studies in rat somatosensory cortex have been unable to detect the dip altogether [34-36]. These discrepancies may be due to anesthesia, differences in cortical architecture, or other effects.

Alternatives to BOLD fMRI include CBF- (or perfusion-) weighted fMRI and CBV-weighted fMRI. In CBF fMRI, 1H spins in arterial blood water are used as endogenous flow tracers [63-

and temporal resolution and avoidance of exogenous radioactive tracers.

weighted functional image of visual cortex [55].

and therefore increases signal strength.

studies of columnar functional architecture.

since it was restricted to metabolically active areas.

Intraoperative OISI maps the brain by measuring activity-related changes in cortical light reflectance. Activity related reflectance changes were first demonstrated in nervous tissue in vitro more than 50 years ago [67] and have since been observed in vivo in rodents, cats, nonhuman primates, and humans. It is a particularly attractive brain-mapping modality because it can rapidly assess the functional activity of large cortical areas with very high spatial resolution (50–100 µm). Because of its versatility, OISI has been used to characterize numerous physiological phenomena, including neurovascular coupling [67-69], hemodynamic refractory periods [70], vasomotion [71], the organization of the visual cortex [71-73], cortical plasticity [71-76], cortical spreading depression [77], seizure [78-79], and language organization in the human brain [80]. Haglund and colleagues [81] were the first to observe optical signals in humans during seizure and cognitive tasks. Since then, the authors of studies on intraoperative OISI have described optical signal evolution in human cortex [82], the mapping of primary sensory and motor cortices [70], and the delineation of language cortices within [83] and across languages [80].

#### **5.1 Intrinsic optical signals**

OISI detects perfusion-related and metabolic signals that are coupled to neuronal activity, including hemoglobin concentration and oxygenation changes, cytochrome oxidation changes, and light scattering caused by altered blood volume, blood flow, and cell swelling [15, 17, 30, 84-86], which in turn create a functional map of the brain. Each of these different phenomena is observed and best quantified at different imaging wavelengths [30, 87]. For example, imaging at 610 nm best detects deoxyhemoglobin concentration changes because the absorbance of oxyhemoglobin is much less than that of deoxyhemoglobin at 610 nm [30]. Thus, OISI at 610 nm is analagous to BOLD fMRI [20], which capitalizes on local magnetic susceptibility changes due to differences between deoxyhemoglobin and oxyhemoglobin concentration.

Intraoperative Human Functional Brain Mapping Using Optical Intrinsic Signal Imaging 87

Commercial systems for clinical intraoperative OISI imaging are not yet available. Investigational systems can be developed from an existing operating microscope with the addition of a camera, camera controller, personal computer, and software to control image

Fig. 3. Intraoperative photograph showing CCD camera mounted on operating microscope. The cortex is illuminated with white light, and the reflected light is filtered at a particular wavelength of interest. Images are captured by a very sensitive CCD camera (shown with green arrow), and reflectance changes between rest and stimulation are measured. These changes correspond to evolving aspects of the hemodynamic response generated as a

OISI has been extensively used in animals to date and has paved the way for major breakthroughs in our understanding of the functional organization, physiology and pathophysiology of the brain [30, 68-69, 72, 74-78, 84-85, 92, 94-97]. OISI studies in animals have defined the functional topography of visual [72] and other cortices, helped elucidate the coupling between electrophysiology and perfusion-related signals, characterized the robustness of neurovascular response capacities, and described perfusion-related changes induced by pathophysiological processes such as cortical spreading depression and seizure. These studies have repeatedly demonstrated the versatility of this modality and its

consequence of underlying cortical activity.

**5.3 OISI in animal models** 

acquisition and analysis.

Fig. 2. Hemoglobin absorption spectrum. Optical intrinsic signal imaging (OISI) relies on changes in cortical light reflectance produced by the hemodynamic response. The most important absorber in the visible spectrum is hemoglobin (Hb). Because oxy- (HbO2) and deoxyhemoglobin (Hbr) absorb light differentially, OISI is wavelength dependent. By selecting different imaging wavelengths, different aspects of the hemodynamic response can be assessed. Both Hb species absorb equally at isosbestic points (549, 569 nm; green light), so reflectance changes at these wavelengths emphasize changes in total Hb, a measure of blood volume. The increased absorbance of Hbr in the 605-630 nm range (red light) permits estimation of oxygenation changes by imaging in this range.

#### **5.2 OISI mapping**

Optical maps are integrated comparisons between the cortex at rest and during prescribed activity. OISI can only be performed intraoperatively in humans because it must be performed following craniotomy and dural reflection. A CCD camera is used to detect small optical changes (0.5–5%) and is mounted onto an operating microscope (Figure 3) or other support structure for imaging. The cortex is then epi-illuminated with white light, and the CCD camera captures the reflected light after it passes through a band-pass filter. Maps of functional change are calculated by comparing images during activation to images at rest [70, 80, 82-83, 89-91]. Multiple trials are averaged to increase the SNR.

Respirophasic and cardiophasic movements of the brain are significant sources of noise during imaging. Using a glass plate to immobilize the cortex [81], synchronizing image acquisition with respiration and heart rate [82] and using image registration [80-81, 83] allow for a reduction in this noise. Vascular artifacts from blood vessels are another major source of noise. Focusing 1-2 mm below the cortical surface [87] and imaging only immediately after stimulus onset [92] can minimize this artifact when the area being imaged is close to large vessels [13, 71, 93].

Commercial systems for clinical intraoperative OISI imaging are not yet available. Investigational systems can be developed from an existing operating microscope with the addition of a camera, camera controller, personal computer, and software to control image acquisition and analysis.

Fig. 3. Intraoperative photograph showing CCD camera mounted on operating microscope. The cortex is illuminated with white light, and the reflected light is filtered at a particular wavelength of interest. Images are captured by a very sensitive CCD camera (shown with green arrow), and reflectance changes between rest and stimulation are measured. These changes correspond to evolving aspects of the hemodynamic response generated as a consequence of underlying cortical activity.

#### **5.3 OISI in animal models**

86 Advances in Brain Imaging

Fig. 2. Hemoglobin absorption spectrum. Optical intrinsic signal imaging (OISI) relies on changes in cortical light reflectance produced by the hemodynamic response. The most important absorber in the visible spectrum is hemoglobin (Hb). Because oxy- (HbO2) and deoxyhemoglobin (Hbr) absorb light differentially, OISI is wavelength dependent. By selecting different imaging wavelengths, different aspects of the hemodynamic response can be assessed. Both Hb species absorb equally at isosbestic points (549, 569 nm; green light), so reflectance changes at these wavelengths emphasize changes in total Hb, a measure of blood volume. The increased absorbance of Hbr in the 605-630 nm range (red light) permits

Optical maps are integrated comparisons between the cortex at rest and during prescribed activity. OISI can only be performed intraoperatively in humans because it must be performed following craniotomy and dural reflection. A CCD camera is used to detect small optical changes (0.5–5%) and is mounted onto an operating microscope (Figure 3) or other support structure for imaging. The cortex is then epi-illuminated with white light, and the CCD camera captures the reflected light after it passes through a band-pass filter. Maps of functional change are calculated by comparing images during activation to images at rest

Respirophasic and cardiophasic movements of the brain are significant sources of noise during imaging. Using a glass plate to immobilize the cortex [81], synchronizing image acquisition with respiration and heart rate [82] and using image registration [80-81, 83] allow for a reduction in this noise. Vascular artifacts from blood vessels are another major source of noise. Focusing 1-2 mm below the cortical surface [87] and imaging only immediately after stimulus onset [92] can minimize this artifact when the area being imaged

estimation of oxygenation changes by imaging in this range.

[70, 80, 82-83, 89-91]. Multiple trials are averaged to increase the SNR.

**5.2 OISI mapping** 

is close to large vessels [13, 71, 93].

OISI has been extensively used in animals to date and has paved the way for major breakthroughs in our understanding of the functional organization, physiology and pathophysiology of the brain [30, 68-69, 72, 74-78, 84-85, 92, 94-97]. OISI studies in animals have defined the functional topography of visual [72] and other cortices, helped elucidate the coupling between electrophysiology and perfusion-related signals, characterized the robustness of neurovascular response capacities, and described perfusion-related changes induced by pathophysiological processes such as cortical spreading depression and seizure. These studies have repeatedly demonstrated the versatility of this modality and its

Intraoperative Human Functional Brain Mapping Using Optical Intrinsic Signal Imaging 89

The ultimate goal of intraoperative mapping is to predict when resection of a cortical area will cause functional deficits. As such, ESM is the current gold standard for intraoperative mapping because it produces reversible lesions such as would occur permanently with resection. However, ESM interpretation is complicated by the fact that cortical stimulation may disrupt remote areas via stimulation of neuronal projections [103-104], and this may lead to imprecise maps [105]. Thus, direct activation-based techniques such as OISI that rely on the detection of local neuronal activity may create a clearer picture when used in

The spatial resolution of ESM is also one of its important drawbacks. Resection within 1 cm of essential areas identified by ESM increases the likelihood of postoperative neurological deficits [106-107]. Because of this local current spread, the resolution of this techique is relatively coarse. On the other hand, OISI offers a resolution as high as 50 to 100 µm as demonstrated by

Rapidity of assessment is another consideration. ESM requires the testing of several sites, at different current levels, during numerous tasks. The process requires several repetitions and is relatively quite time-consuming. Intraoperative OISI provides a faster assessment of the cortical surface, as the entire field of view can be imaged at once. This advantage is particularly important when mapping multiple tasks in an area that covers several possible

Another advantage of OISI is its non-tactile nature, as it relies solely on light reflection. ESM requires direct contact with the brain, and the application of current to its surface. This process can precipitate abnormal electrical propagation known as after-discharge activity, which can escalate into a clinical seizure. Intraoperative seizures are not only dangerous, but

Furthermore, essential areas identified by ESM are consistently demonstrated by intraoperative OISI maps, suggesting that they do indeed offer a relatively high sensitivity. Thus, cortical regions not demonstrating task-related OISI activity can probably be resected without functional consequence. However, vascular spread, as previously described, may produce false-positive results in OISI maps by highlighting secondary and/or non-essential cortical areas, thus hindering our ability to produce a maximal resection. Thus, OISI may not completely replace ESM but may be used in conjunction as a complementary modality to

OISI can also be used in conjunction with fluorescent dyes that allow for more precise physical localization of pathological tissues such as tumors. Dyes were first used in animal studies [15, 96, 107-108] for this purpose. For example, optical imaging of an intravenously injected dye in rodents demonstrates intracranial tumors with high accuracy [107], and this has correlated well with human studies as well. Preoperative injection of 5-aminolevulinic acid, a precursor of fluorescent porphyrin, could be used to identify malignant gliomas with 85% sensitivity and 100% specificity [108]. The 5-aminolevulinic acid accumulates within the malignant tissue, where it is converted to its fluorescent derivative, which is then imaged intraoperatively using optical imaging with special optics. This is a powerful adjunctive application to optical imaging that can provide spatial information of the relationship

also often preclude further mapping, as the brain is relatively depressed afterwards.

numerous studies, potentially allowing for a much finer delineation of eloquence.

**5.5 Advantages of intraoperative OISI mapping** 

conjunction with established methods like ESM.

functional representations.

improve the accuracy of cortical maps.

between pathological tissue and essential cortical areas.

numerous potential applications. The specificity of optical maps generated through OISI has been consistently confirmed in these various studies when compared to other established invasive methodologies, including single unit activity measurement [87, 95], maximum field potential measurement [96-97], and cytochrome oxidase-staining [68, 95]. OISI has also been used for mapping seizure propagation in the cortex [78] and for identifying epileptogenic foci [79].

#### **5.4 OISI in humans**

The first OISI maps of human function were created by asking the patient to engage in simple motor tasks such as tongue movement and simple language task such as visual object-naming exercises. Motor tasks demonstrated clear optical signals in the motor cortex, and language tasks demonstrated activity in both Broca and Wernicke areas [81]. Optical responses in humans typically appear within 1 second, peak between 3 and 4 seconds, and disappear by 9 seconds, similar to those observed in animal models [82, 89]. Indeed, this similar timing of optical signals across species suggests that we are indeed imaging similar phenomena, and because we have clearer physiological correlates in animals obtained through invasive methodologies, we can extrapolate that we are imaging the same neural activity. In line with this assertion, hemodynamic refractory periods originally observed in rodents were also observable in humans during OISI mapping [70]. In further corroboration, all human OISI studies to date indicate that the observed reflectance changes are spatially correlated with somatosensory evoked potential (SSEP) data. Median and ulnar nerve stimulation leads to a decrease in optical signal obtained at 610 nm in an area that colocalizes with the largest SSEPs in both somatosensory and motor cortices [82]. More importantly, OISI maps similarly colocalize with intraoperative electrocortical stimulation mapping (ESM), the current gold standard of cortical mapping [80-81, 83]. Areas that are identified by ESM as essential for a specific task consistently demonstrate optical activity [80-81, 83].

A notable phenomenon with optical signals is "vascular spread." OISI maps demonstrate signal in some surrounding areas (approximately 25%) that are not identified as essential on ESM. This has been demonstrated in both rodent models [98-100] and human intraoperative imaging. Spread may occur in part because intraoperative OISI is detecting both essential and secondary cortices whereas ESM is only detecting essential areas. Spread may also be related to imprecise physiological coupling of neuronal activity, metabolism, and perfusion. The sensitivity, specificity, positive predictive value, and negative predictive value of intraoperative OISI relative to ESM as a gold standard have not been fully quantified [101- 102], and such quantification will be essential to the broadening of intraoperative OISI as a fundamental clinical tool.

OISI has also begun to provide a finer resolution of cortical mapping with regards to task specific activity. For example, OISI demonstrates differentially activated gyri during tongue movement and naming tasks, activities which require similar movements but which are otherwise distinct [81]. Median and ulnar nerve stimulation produces distinct maps within the same gyrus, possibly providing a fine resolution of somatosensory and motor activity in the cortex [82]. Distinct maps of face, thumb, and index and middle fingers have also been obtained within the same gyrus [70], and although there is some overlap between these areas, the areas of maximum optical signal are distinct for each task. Such specificity has also been observed in language areas (Broca and Wernicke areas) [83].

#### **5.5 Advantages of intraoperative OISI mapping**

88 Advances in Brain Imaging

numerous potential applications. The specificity of optical maps generated through OISI has been consistently confirmed in these various studies when compared to other established invasive methodologies, including single unit activity measurement [87, 95], maximum field potential measurement [96-97], and cytochrome oxidase-staining [68, 95]. OISI has also been used for mapping seizure propagation in the cortex [78] and for identifying epileptogenic

The first OISI maps of human function were created by asking the patient to engage in simple motor tasks such as tongue movement and simple language task such as visual object-naming exercises. Motor tasks demonstrated clear optical signals in the motor cortex, and language tasks demonstrated activity in both Broca and Wernicke areas [81]. Optical responses in humans typically appear within 1 second, peak between 3 and 4 seconds, and disappear by 9 seconds, similar to those observed in animal models [82, 89]. Indeed, this similar timing of optical signals across species suggests that we are indeed imaging similar phenomena, and because we have clearer physiological correlates in animals obtained through invasive methodologies, we can extrapolate that we are imaging the same neural activity. In line with this assertion, hemodynamic refractory periods originally observed in rodents were also observable in humans during OISI mapping [70]. In further corroboration, all human OISI studies to date indicate that the observed reflectance changes are spatially correlated with somatosensory evoked potential (SSEP) data. Median and ulnar nerve stimulation leads to a decrease in optical signal obtained at 610 nm in an area that colocalizes with the largest SSEPs in both somatosensory and motor cortices [82]. More importantly, OISI maps similarly colocalize with intraoperative electrocortical stimulation mapping (ESM), the current gold standard of cortical mapping [80-81, 83]. Areas that are identified by ESM as essential for a

A notable phenomenon with optical signals is "vascular spread." OISI maps demonstrate signal in some surrounding areas (approximately 25%) that are not identified as essential on ESM. This has been demonstrated in both rodent models [98-100] and human intraoperative imaging. Spread may occur in part because intraoperative OISI is detecting both essential and secondary cortices whereas ESM is only detecting essential areas. Spread may also be related to imprecise physiological coupling of neuronal activity, metabolism, and perfusion. The sensitivity, specificity, positive predictive value, and negative predictive value of intraoperative OISI relative to ESM as a gold standard have not been fully quantified [101- 102], and such quantification will be essential to the broadening of intraoperative OISI as a

OISI has also begun to provide a finer resolution of cortical mapping with regards to task specific activity. For example, OISI demonstrates differentially activated gyri during tongue movement and naming tasks, activities which require similar movements but which are otherwise distinct [81]. Median and ulnar nerve stimulation produces distinct maps within the same gyrus, possibly providing a fine resolution of somatosensory and motor activity in the cortex [82]. Distinct maps of face, thumb, and index and middle fingers have also been obtained within the same gyrus [70], and although there is some overlap between these areas, the areas of maximum optical signal are distinct for each task. Such specificity has

specific task consistently demonstrate optical activity [80-81, 83].

also been observed in language areas (Broca and Wernicke areas) [83].

foci [79].

**5.4 OISI in humans** 

fundamental clinical tool.

The ultimate goal of intraoperative mapping is to predict when resection of a cortical area will cause functional deficits. As such, ESM is the current gold standard for intraoperative mapping because it produces reversible lesions such as would occur permanently with resection. However, ESM interpretation is complicated by the fact that cortical stimulation may disrupt remote areas via stimulation of neuronal projections [103-104], and this may lead to imprecise maps [105]. Thus, direct activation-based techniques such as OISI that rely on the detection of local neuronal activity may create a clearer picture when used in conjunction with established methods like ESM.

The spatial resolution of ESM is also one of its important drawbacks. Resection within 1 cm of essential areas identified by ESM increases the likelihood of postoperative neurological deficits [106-107]. Because of this local current spread, the resolution of this techique is relatively coarse. On the other hand, OISI offers a resolution as high as 50 to 100 µm as demonstrated by numerous studies, potentially allowing for a much finer delineation of eloquence.

Rapidity of assessment is another consideration. ESM requires the testing of several sites, at different current levels, during numerous tasks. The process requires several repetitions and is relatively quite time-consuming. Intraoperative OISI provides a faster assessment of the cortical surface, as the entire field of view can be imaged at once. This advantage is particularly important when mapping multiple tasks in an area that covers several possible functional representations.

Another advantage of OISI is its non-tactile nature, as it relies solely on light reflection. ESM requires direct contact with the brain, and the application of current to its surface. This process can precipitate abnormal electrical propagation known as after-discharge activity, which can escalate into a clinical seizure. Intraoperative seizures are not only dangerous, but also often preclude further mapping, as the brain is relatively depressed afterwards.

Furthermore, essential areas identified by ESM are consistently demonstrated by intraoperative OISI maps, suggesting that they do indeed offer a relatively high sensitivity. Thus, cortical regions not demonstrating task-related OISI activity can probably be resected without functional consequence. However, vascular spread, as previously described, may produce false-positive results in OISI maps by highlighting secondary and/or non-essential cortical areas, thus hindering our ability to produce a maximal resection. Thus, OISI may not completely replace ESM but may be used in conjunction as a complementary modality to improve the accuracy of cortical maps.

OISI can also be used in conjunction with fluorescent dyes that allow for more precise physical localization of pathological tissues such as tumors. Dyes were first used in animal studies [15, 96, 107-108] for this purpose. For example, optical imaging of an intravenously injected dye in rodents demonstrates intracranial tumors with high accuracy [107], and this has correlated well with human studies as well. Preoperative injection of 5-aminolevulinic acid, a precursor of fluorescent porphyrin, could be used to identify malignant gliomas with 85% sensitivity and 100% specificity [108]. The 5-aminolevulinic acid accumulates within the malignant tissue, where it is converted to its fluorescent derivative, which is then imaged intraoperatively using optical imaging with special optics. This is a powerful adjunctive application to optical imaging that can provide spatial information of the relationship between pathological tissue and essential cortical areas.

Intraoperative Human Functional Brain Mapping Using Optical Intrinsic Signal Imaging 91

however, OISI presents important new advances that can potentially improve clinical

One potential shortcoming of OISI, alluded to above, is its ambiguous etiology. Studies using OISI usually report "activity" as a certain fractional change in reflectance from baseline. These reflectance changes incorporate changes in absorbance and scattering related to a number of physiological processes. Specific hemodynamic processes can be isolated to some degree by choosing appropriate wavelengths, but other contributions certainly exist. This drawback was addressed in the late 1990s by Malonek and Grinvald [16-17]. They disambiguated the various contributions of absorbance and scattering by developing a variant of OISI known as OIS spectroscopy. In its most common form, broadband light reflected from the cortex is focused on a primary image plane containing a spectrographic slit instead of a detector. The one-dimensional column of light is then incident upon a diffraction grating that disperses the light into its constituent wavelengths along a second orthogonal axis. This two-dimensional "spatio-spectral" image is then refocused on a second image plane and captured by a camera. The *x*-dimension of the image represents the wavelength of light at a particular point along the slit, and the *y*-dimension represents vertical position. Spatio-spectral images are taken over time to capture the hemodynamic

This approach essentially sacrifices one dimension of spatial information for an extra dimension of spectral information. The advantage gained is the ability to fit the spectra acquired over time to a model containing physiological parameters that are known to change during the hemodynamic response and affect light reflectance. Models are usually based on the Beer-Lambert law, which describes light attenuation in the presence of

absorber concentration, and *l* is the pathlength through the tissue. In living tissue under normal physiological circumstances, Hbr and HbO2 are the most important absorbers in visible wavelengths. Cytochrome oxidase also absorbs in the visible range, but its oxidation state only changes in cases of extremely low oxygen saturation. Because its absorbance is also an order of magnitude smaller than hemoglobin, it is generally not considered an important model component [71]. Early models incorporated scattering as an additive linear

Spectral data are fit to the model to extract the timecourses of the model parameters, i.e., Hbr and HbO2. OIS spectroscopy therefore provides changes in physiological variables rather than (somewhat arbitrary) reflectance changes. This advantage allows for more direct

The results derived from OIS spectroscopy are only as valid as the model. The most important model refinements have been better consideration of wavelength dependency. The fact that different wavelengths of light penetrate biological tissue to different depths has been recognized for several years [113]. Longer wavelengths penetrate deeper into tissue and therefore travel through a longer pathlength (*l* in the above equations), another way of

, which would also capture residual errors.

comparison between OIS data and other modalities such as fMRI.

, where *εi* is the extinction coefficient of the *i*th absorber, *c* is the

outcomes when used in conjunction with other established modalities.

**6. OIS spectroscopy** 

response, as in other techniques.

*i Abs c l* 

absorbers: *i i*

term: ( ) *i i i Abs c l S* 

Importantly, like all intraoperative imaging modalities, OISI presents a distinct advantage over preoperative functional mapping (as provided by fMRI or PET) because it can correct for "brain shift" following craniotomy and dural reflection [109]. This is a non-trivial problem that confounds our ability to rely on preoperative functional mapping alone. Intraoperative OISI requires only minimal modification of the neurosurgical equipment already found in the operating room and does not impact the surgery or affect normal brain tissue as it relies solely on measuring reflected light from the brain.

#### **5.6 Limitations of intraoperative OISI mapping**

While neurovascular and neurometabolic coupling to neuronal activity appear to be consistent in numerous studies to date [15, 80, 82-83, 86-87, 95-97], the major limitation of intraoperative OISI continues to be the fact that the signal does not directly arise from neuronal activity. This becomes particularly relevant when dealing with pathological cortex, as is frequently the case during neurosurgical interventions, where the coupling may not be as tight as in normal cortex. This is a major question that remains to be elucidated and arguably can only be investigated with high-resolution intraoperative measures such as OISI.

Vascular lesions such as arteriovenous malformations (AVMs) present perhaps the most important challenge. Abnormal vascular networks may provide altered and unreliable signal in cortical areas adjacent to AVMs. While several studies have found that perfusionrelated mapping signals can be detected directly adjacent to AVMs and can therefore be used reliably to predict essential language sites identified by ESM [110-112], the interpretation of results in these patients should still be approached cautiously.

Vascular spread in OISI maps is another confounder in the use of intraoperative OISI as a single modality. Indeed, using it as a lone modality may produce significant false-positive results that would prevent maximum resection of pathological and non-eloquent tissue. Until we understand this spread phenomenon better and are able to control for it, intraoperative OISI cannot replace ESM. However, it may provide an important complementary modality – for example, intraoperative OISI can be used to rapidly map cortical areas of interest with high spatial resolution, and those areas found to demonstrate optical activity can then be verified by ESM.

Another potential drawback of OISI is its limited signal-to-noise ratio (SNR). In language mapping trials, SNR values range from 5:1 to 9:1 when averaging four trials. This limited SNR can be attributed to patient head motion as well as respirophasic and cardiophasic cortical movements. While SNR can be improved by increasing the number trials that are averaged, reducing cortical movements by using a glass plate [81], or synchronizing image acquisition with respiration and pulse [82], this is still a challenge that remains to be addressed. Furthermore, increasing the number of trials elongates the time required for the procedure, a distinct disadvantage. Furthermore, unlike fMRI and PET, which afford threedimensional maps, intraoperative OISI produces surface maps, usually to a maximum depth of 1 mm. This represents a further limitation.

As the utilization of OISI increases, we will begin to understand its strengths and weaknesses to a greater extent, potentially enabling the development of auxiliary technologies that augment these strengths or overcome these weaknesses. On the whole, however, OISI presents important new advances that can potentially improve clinical outcomes when used in conjunction with other established modalities.
