Imaging cortical absorption, scattering, and hemodynamic response during ischemic stroke using spatially modulated near-infrared illumination.

Animal Model

We examined the effects of occlusion of MCA in seven rats. Animal housing, care, and experimental protocols were carried out in conformity with the guidelines of the U.S. National Institutes of Health. The laboratory animal protocol for this work was approved by the Institutional Animal Care and Use Committee (IACUC) at the University of California, Irvine (protocol no. 2006-2671) and by the U.S. Department of Defense (DoD). Seven adult male Sprague-Dawley rats (Charles River Laboratories, Wilmington, Massachusetts) weighing between 350 and 400 g were deeply anesthetized by using barbiturate anesthesia [sodium pentobarbital 55 mg/kg initial dose intraperitoneal (ip), 2 mg/kg ip supplements as needed to maintain anesthesia] and their heads were fixed on a stereotactic apparatus. To decrease respiratory secretions and possibility of syncopal attacks, atropine (0.1 mg/kg) was injected imtra-muscularly, immediately prior to making a surgical incision and injected every 6 h. To prevent dehydration, 5 cc of normal saline was injected subcutaneously on the back of the animal prior to surgical incision. To prevent hypothermia temperature was kept constant by a using a rectal thermometer that is connected to an electric heating pad (model 50-7053-F, Harvard Appartures, Holliston, Massachusetts) by a feedback mechanism, maintaining body temperature at 37±0.5° during the entire experiment. Throughout the experiment, a surgical plane of anesthesia was maintained based on the lack of the toe pinch reflex. A midline skin incision was made and a 10×10-mm square area overlying the left somatosensory cortex was outlined. The skull over this area was thinned with a dental drill handpiece (model 3610N1) to about 150 μ m until the MCA and superior cerebral veins were visible. A vaseline wall was built around the thinned area of the skull region and filled with saline. The saline film was covered with a glass coverslip and the entire setup served as an imaging window. The saline is filled in the cranial window for four main reasons: (1) preventing the skull from drying, (2) reducing surface reflection, (3) increasing the transparency of the light into the brain, and (4) permitting a clear visualization of the cortex through the entire experiment. A craniotomy was performed on the left side of the brain outside the imaging window to expose the MCA. Two well-established animal models for cerebral ischemia have been used during our study to produce permanent MCA occlusion: cauterization and ligation. In the cauterization, the proximal MCA on the left cortex was surgically coagulated using monopolar electric cautery. In the ligation, the MCA was ligated using 4-0 silk sutures, which passed through the pial layer below the MCA and above the surface of the cortex. The rat was placed under the CCD camera such that most of the thinned skull region was contained within the camera’s field of view and that the optical axis of the camera lens was perpendicular to the center of the imaged region. All surgical procedures were performed under a surgical microscope (Carl Zeiss, Primo 1880). At the end of each experiment, the rat was euthanatized with pentobarbital sodium, Eutha-6, intravenous injection.

Instrument

A schematic of the experimental arrangement of the spatial light illumination is illustrated in Fig. 1 . Periodic illumination patterns of various spatial frequencies with a 120-deg phase-shifting step between three adjacent patterns were projected onto the brain from a commercial digital light projector (BENQ, PB8260) controlled by a computer. The light intensity distribution of this pattern has the form of an amplitude-modulated sinusoidal wave. These patterns were created on a personal computer using PowerPoint software and projected sequentially using Microsoft Office’s ActiveX controls via an external LabView program (Ver. 7). The patterns were converted from white light projection to NIR using a filter wheel (Spectral Product, AB302) placed immediately at the output of the projector. We used five narrow NIR bandpass filters between 680 and 980 nm (Andover Corp.) placed on a five-position wheel. The diffusely reflected light, which contains information about the brain tissue optical properties, was recorded by a 16-bit CCD camera (Cascade, 512F) mounted normal to the brain surface. The optical axis of the camera lens was perpendicular to the center of the imaging region. The camera is equipped with an imaging objective (Nikon f/1.8 lens, Japan) lens combined with an extender and is capable of imaging up to 30 frames/s at full 512×512 resolution. The distance between the CCD lens and the brain was about 100 mm. A 6×6-mm area was imaged onto the CCD with resolution of 0.012 mm/pixel for both the x and y axes. As a part of the calibration procedure, a lab jack was used to bring the rat’s head to focus within the CCD view. To compensate for the wavelength-dependent illumination and detection throughput, image exposure times were chosen at each wavelength, prior to each experiment as a part of system calibration, to fill the dynamic range of the camera, ranging from 10 ms (QE~ 50% at 680 nm) to 1 s (QE ~ 10% at 980 nm). The mirror appearing in the setup was aligned at a small angle of incidence, ~15 deg to the vertical axis, to obtain projection illumination at a small angle, and therefore avoid the detection of specular reflected light. Polarized filters (Meadowlark Optics, VLM-200-IR-R) in both the illumination and detector paths were aligned with the polarization planes orthogonal to each other for additional specular elimination. The entire system is controlled by a personal computer through a LabView platform. As already mentioned, the principle component of our working system is a commercial digital light projector that is used to produce the patterned illumination. However, to fit the projector to our demands such as illumination stability, high intensity, and scalability to a small region of interest, the following modification steps were required. The white light source (Mercury lamp, OSRAM, 300 W) and the power supply of the projector were replaced with a 300-W quartz-tungsten-halogen (QTH) (EIKO, EXR, 82v) bulb and with a corresponding power supply. We decide to use this bulb since it gives a smooth broadband spectral profile along the NIR region and stable intensity over time (light output variation less then 0.6%). As shown in Sec. 3, these fluctuations are negligible compared to the signals arriving from the brain during acute ischemic injury. To boost intensity from the QTH bulb and to use the entire spectral band, the color wheel inside the projector was carefully removed, enabling the projector to work as usual. To prevent warming of the projector, a hybrid hot mirror (Reynard Corp., C-R00670-00) was introduced between the QTH bulb and the optics inside the projector. Both the bulb and the hybrid mirror were maintained inside the projector through careful mechanical design and holders. To collimate the patterns on the surface of brain, to obtain a small (~10×10-mm) and adjustable field of view, and to compensate for aberrations derived from using white source, a new lens system was design based on a set of achromatic lenses (LA-series, ThorLabs), replacing the original lens system of the projector.

Optical Property Calculation and Calibration

A detailed description of spatially modulated imaging theory and optical property measurements in turbid media have already been treated in detail, 11 but the basic features are reproduced here in the interest of clarity and completeness. Briefly, three sinusoidal patterns with a 120-deg phase shift between each are projected at the same spatial frequency denoted by I 0 ( ω x ), I 120 ( ω x ), I 240 ( ω x ). The diffusely reflected light is composed of two components: dc ( f x =0) and ac ( f x ≠ 0) reflectance. The amplitude of the ac and the dc diffuse reflectance, respectively, can be expressed as 12 (1) A ac ( ω x ) = 2 3 { [ I 0 ( ω x ) − I 120 ( ω x ) ] 2 + [ I 120 ( ω x ) − I 240 ( ω x ) ] 2 + [ I 240 ( ω x ) − I 0 ( ω x ) ] 2 } 1 / 2 , A dc ( ω x ) = 1 3 [ I 0 ( ω x ) + I 120 ( ω x ) + I 240 ( ω x ) ] . In the frequency domain A ac and A dc are given by (2) A ac ( ω x ) = ( I s ) [ MTF sys ( ω x ) ] [ R d ( ω x ) ] , A dc ( ω x = 0 ) = ( I s ) [ MTF sys ( ω x = 0 ) ] [ R d ( ω x = 0 ) ] , where ω x is the radial spatial frequency given by ω x =2 πf x , f x is the spatial frequency, I s is the source intensity, and MTF sys is the modulation transfer function of the entire system (projector, tissue, and CCD). To extract precisely the diffuse reflectance measurement, random system artifacts such as source strength variability, nonuniform response of the CCD camera, electronic noise caused by the projector-DMD element and the CCD, and optics distortion must be removed. Therefore, we performed calibration based measurements from a tissue-simulating phantom (siloxane with TiO 2 particles) with known optical values a priori at each wavelength to obtain A ac ( ω x ) ref and A dc ( ω x ) ref , respectively. Equation (2) is then divided by these measurements and multiplied by a model prediction for the same turbid phantom, R d ( ω x ) ref_pred . This process can be described through the following: (3) [ A ac ( ω x ) A ac ( ω x ) ref ] [ R d ( ω x ) ref_pred ] ⇒ ( I s ) [ MTF sys ( ω x ) ] [ R d ( ω x ) ] ( I s ) [ MTF sys ( ω x ) ] [ R d ( ω x ) ref ] R d ( ω x ) ref_pred . Note that by using the reference’s known optical properties we can calculate the diffuse reflectance of the model-based prediction, R d ( ω x ) ref_pred using Monte Carlo simulation. As R d ( ω x ) ref_pred approximates R d ( ω x ) ref , Eq. (3) will provide accurate diffuse reflectance measurements for each spatial frequency and each spatial location. By measuring reflectance values at multiple spatial frequencies, R d ( ω x ) can be solved to recover the optical properties of the tissue. We used both a least-squares multifrequency fitting algorithm based on Levenberg-Marquart method in MATLAB , and rapid two-frequency look-up table (LUT) approaches to extract separately the best-fit estimates of absorption and reduced scattering coefficients from R d ( ω x ). Equation (3) is also applied for the dc case.

Imaging and Data Processing

Imaging was performed through the surgically created imaging window on anesthetized rats held in a stereotactic frame. The reflected image of the region of interest (ROI) of the cortex was acquired on a CCD camera while projecting spatially modulated NIR light onto the region of interest. Image sets of each of the five wavelengths were acquired over a 6×6-mm 2 brain surface, with six spatial frequencies ranging from 0 to ~0.26 mm −1 . Images at each frequency were obtained as shown in Eq.(1) . The resulting 90 images (6 spatial frequencies, 3 phase shifts for each frequency, 5 wavelengths) provide a quantitative frequency response (or modulation transfer function, MTF) of the diffuse reflectance of the brain [see Fig. 2(c) ]. Diffuse reflectance images were obtained from the ROI selected by the investigators, and processed by fitting frequency-dependent data to a Monte Carlo model. To separate light absorption from light scattering and subsequently generate chromophore maps faster, we used only two spatial frequencies (from the six frequencies): 0 mm −1 (dc frequency) and 0.16 mm −1 (ac frequency) in our experiments. As a result, multispectral maps were generated approximately every ~1.5 min to study the changes in cortical perfusion with time following experimental manipulation. Imaging was started before induction of ischemia to establish baseline chromophore concentrations and repeated during and after experimental intervention. Throughout the entire study, baseline images were obtained between 15 and 25 min before induction of experimental injury to establish baseline chromophore concentrations. Thus, each rat served as its own control, decreasing the number of animals required for the study. Postinjury imaging was started approximately 15 min after MCA occlusion and continued for 1 h. The image analysis was performed off-line on a PC computer with custom software developed in MATLAB (version 7.04, The Math-Works, Inc). All images before analysis were digitally filtered by a 2-D Gaussian low-pass filter with size of 9×9 with standard deviation (FWHM) of σ =1.5 using the fspecial function in MATLAB and a 3×3 binning operation was performed. Monte Carlo Simulation A Monte Carlo (MC) method was used to define the relationship between the measured reflectance and the optical properties of the rat cortex. We adopted the use of a White Monte Carlo (WMC) technique 13 to accelerate processing time. In WMC, libraries for a specified scattering coefficient and null absorption are built and the reflectance function R’ 0 ( μ’ s 0 , μ a =0) is calculated. The diffuse reflectance function for other scattering values and any set of absorption, R ( μ’ s , μ a ≠0) , is then obtained by multiplying R’ 0 by scaling factors (Ref. 13 , Eq. (2) , p. 2775). Once the diffuse reflectance, R ( μ’ s , μ a ), is obtained, a Fourier-Bessel transform is used to predict the diffuse reflectance in the spatial frequency domain. 14 In this study, a WMC simulation was performed for a collimated point source launching 10 million photons into the medium with a detector numerical aperture of 0.22. All simulations assumed a semiinfinite homogenous medium with index of refraction n =1.43, anisotropy factor g =0.9, and the Henyey-Greenstein phase-function. The grid resolution (d r ) was 0.09 mm, making the maximum spatial frequency greater than 10 mm −1 . For five wavelengths and six spatial frequencies, typical simulation times were less then 6 h on a current desktop system (AMD Turion 1.6-GHz CPU).

Statistical Considerations

Comparison of optical and chromophore properties pre- and post-MCA occlusion were performed by using a paired t test. Differences in properties were considered to be significant at a probability level of less than 0.05 ( p