**6. OIS spectroscopy**

90 Advances in Brain Imaging

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

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

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

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

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

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,

interpretation of results in these patients should still be approached cautiously.

tissue as it relies solely on measuring reflected light from the brain.

**5.6 Limitations of intraoperative OISI mapping** 

optical activity can then be verified by ESM.

of 1 mm. This represents a further limitation.

OISI.

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 response, as in other techniques.

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 absorbers: *i i i Abs c l* , where *εi* is the extinction coefficient of the *i*th absorber, *c* is the

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 term: ( ) *i i i Abs c l S* , which would also capture residual errors.

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 comparison between OIS data and other modalities such as fMRI.

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

Intraoperative Human Functional Brain Mapping Using Optical Intrinsic Signal Imaging 93

The principle advantage of NIRS over other optical techniques is its noninvasive nature. NIRS provides information about functional oxygenation and volume changes that are directly comparable to fMRI, but the apparatus is much less costly and confining. NIRS can be performed in pediatric populations much more easily than fMRI [124], and it can be transported to the bedside for clinical evaluations [125-128]. Because NIRS signals are detected several centimeters from the cortex, however, the spatial resolution of the technique is low (~1-2 cm). Spatial coverage can be increased by using arrays of emitters and detectors, but the emitter-detector spacing must be at least 2-3 cm to allow the light to

Modifications to the acquisition and analysis methods allow images to be created from NIRS data [129]. In this variant, a grid of emitters and detectors is placed on the head, providing several emitter-detectors pairs. NIRS imaging, or diffuse optical tomography (DOT), is the optical analog of PET, EEG, or MEG, in that it requires measurement of surface signals and calculation of the source distribution that could have produced them. It therefore also involves solving an inverse problem, which is again poorly constrained. Recent work has shown, however, that analyzing multi-channel NIRS data with this imaging approach

The brain mapping techniques described above, both direct and indirect, can be combined, such that the information gained by their combination surpasses their advantages individually. The most beneficial combinations usually involve compensating for limitations of one technique with another, or concurrently measuring different aspects of the same response to better understand its physiological basis. For example, combined OISI-fMRI studies have also contributed to our knowledge of the etiology of these signals [101, 130- 131]. The recent development of a system that allows simultaneous fMRI and OIS spectroscopy [132] promises to further this endeavor. This combination has also been used to test the clinical utility of intraoperative human OISI for neurosurgical guidance by

A central tenet in neurosurgery is avoidance of new postoperative neurological deficit. This goal is especially challenging when operating in or near "eloquent cortex", or regions subserving known specific functions, such as sensation, motor control, or language. Given individual neuroanatomical variations, eloquent regions must be delineated at the time of surgery, within the individual patient. The conventional method for identifying eloquent cortex intraoperatively is electrical stimulation mapping (ESM), during which regions of cortex are directly stimulated with a small electrical current using a hand-held probe. However, ESM has several limitations including limited spatial resolution, lengthy protocol time which places the patient at increased anesthesia and infection risk and increases costs, and a higher risk of seizure due to direct electrical stimulation of the cortex. Intraoperative OISI can provide maps of cortical function rapidly and without contacting the brain, therefore reducing operative time and seizure likelihood. These maps will be complementary to ESM for the localization of eloquent cortex. Furthermore, these maps can

provides better estimates of functional changes than single-channel NIRS [129].

comparing intraoperative OISI maps with pre-surgical fMRI maps [101, 130-131].

sample cortex.

**8. Conclusions** 

**7. Multi-modality approaches** 

saying that they experience more scattering. To properly account for the optical pathlength and scattering, therefore, this dependency must be taken into account.

Mayhew and colleagues performed Monte Carlo simulations to calculate the distribution of differential pathlength factors in the visible spectrum [71]. Since then, several studies have incorporated wavelength dependency into the Beer-Lambert model to more accurately simulate the behavior of light transport through highly scattering biological tissue [28, 33, 114-115]. Their results have shown that accounting for wavelength dependency is critical, especially when assessing small transients in the response such as the initial dip.

#### **6.1 Near-infrared spectroscopy (NIRS)**

In 1977 Jobsis showed that the intact human skull was not necessarily a barrier for light [116]. He found that wavelengths of light beyond the visible spectrum in the near-infrared range (~670-900 nm) can penetrate through several centimeters of skin and skull. This range is ideally situated between the strong absorption spectra of hemoglobin (<~630 nm) and water (>~950 nm), and has therefore been dubbed the "biological window" for noninvasive optical imaging [117-118].

NIRS is based on the same principles as visible range spectroscopy described above. Changes in light attenuation are fit to a modified Beer-Lambert law incorporating scattering and absorption by hemoglobin. The wavelength dependency of the pathlength must be taken into account. Differential pathlength factors can also be calculated using a Monte Carlo simulation, but another method exists in the case of NIRS. Pathlengths can be directly measured using time-resolved spectroscopy systems [113, 119]. These instruments have very fast (picosecond) detectors that are capable of measuring the time of flight of photons traveling through the head, which is directly related to the distance traveled. Alternatively, frequency domain systems can measure the phase difference between incident and remitted light [120-121].

The main difference between NIRS and visible spectroscopy is the way in which light is emitted and collected. Because the cortex is not exposed, light cannot illuminate the entire area. Instead, light is directed into the head through fiber optic guides and diffuses through the skin, skull, and cortex. A detector fiber guide is positioned a few centimeters from the emitter, and captures photons that have scattered through the head in an arc-shaped path from emitter to detector. The greater the distance between emitter and detector, the more likely it is that photons will travel through a deeper arc. On the other hand, a greater separation reduces the number of photons detected. Studies have theoretically and experimentally determined the optimal spacing (2.5-4 cm) to allow the photons to "sample" the top layers of cortex [122-123]. Because functional activation is not expected to produce changes in the skull or scalp, any differences in measured light intensity are attributed to cortical hemodynamic processes, i.e., changes in oxygenation or volume.

NIRS can be performed using broadband or laser illumination. The former situation is directly analogous to visible spectroscopy: remitted light is spectrally decomposed and captured by a camera. This approach affords excellent spectral resolution, but the emitted power per wavelength band is low. In contrast, laser diodes produce more power in a narrow wavelength band, but the number of wavelengths is limited to a few (2-4 in conventional systems), decreasing spectral resolution. Detectors for laser illumination are usually photodiodes, which are much more sensitive than CCDs.

The principle advantage of NIRS over other optical techniques is its noninvasive nature. NIRS provides information about functional oxygenation and volume changes that are directly comparable to fMRI, but the apparatus is much less costly and confining. NIRS can be performed in pediatric populations much more easily than fMRI [124], and it can be transported to the bedside for clinical evaluations [125-128]. Because NIRS signals are detected several centimeters from the cortex, however, the spatial resolution of the technique is low (~1-2 cm). Spatial coverage can be increased by using arrays of emitters and detectors, but the emitter-detector spacing must be at least 2-3 cm to allow the light to sample cortex.

Modifications to the acquisition and analysis methods allow images to be created from NIRS data [129]. In this variant, a grid of emitters and detectors is placed on the head, providing several emitter-detectors pairs. NIRS imaging, or diffuse optical tomography (DOT), is the optical analog of PET, EEG, or MEG, in that it requires measurement of surface signals and calculation of the source distribution that could have produced them. It therefore also involves solving an inverse problem, which is again poorly constrained. Recent work has shown, however, that analyzing multi-channel NIRS data with this imaging approach provides better estimates of functional changes than single-channel NIRS [129].
