ARTICLE pubs.acs.org/ac
Active DLP Hyperspectral Illumination: A Noninvasive, in Vivo, System Characterization Visualizing Tissue Oxygenation at Near Video Rates Karel J. Zuzak,*,†,‡,§ Robert P. Francis,‡,^ Eleanor F. Wehner,‡ Maritoni Litorja,z Jeffrey A. Cadeddu,† and Edward H. Livingston†,‡ †
University of Texas Southwestern Medical Center, 5323 Harry Hines Blvd, Dallas, Texas 75390, United States University of Texas at Arlington, 501 West First Street, Arlington, Texas 76019, United States § Digital Light Innovations, DLi, 4501 Spicewood Springs, Rd., Suite 1000, Austin, Texas 78759, United States ^ Raytheon Elcan Optical Technologies, 1601 N. Plano Road, Richardson, Texas 75081, United States z National Institute of Standards and Technology, 100 Bureau Drive, Gaithersburg, Maryland 20899, United States ‡
bS Supporting Information ABSTRACT: We report use of a novel hyperspectral imaging system utilizing digital light processing (DLP) technology to noninvasively visualize in vivo tissue oxygenation during surgical procedures. The system’s novelty resides in its method of illuminating tissue with precisely predetermined continuous complex spectra. The Texas Instruments digital micromirror device, DMD, chip consisting of 768 by 1024 mirrors, each 16 μm square, can be switched between two positions at 12.5 kHz. Switching the appropriate mirrors controls the intensity of light illuminating the tissue as a function of wavelength, active spectral illumination. Meaning, the tissue can be illuminated with a different spectrum of light within 80 μs. Precisely, predetermined spectral illumination penetrates into patient tissue, its chemical composition augments the spectral properties of the light, and its reflected spectra are detected and digitized at each pixel detector of a silicon charge-coupled device, CCD. Using complex spectral illumination, digital signal processing and chemometric methods produce chemically relevant images at near video rates. Specific to this work, tissue is illuminated spectrally with light spanning the visible electromagnetic spectrum (380 to 780 nm). Spectrophotometric images are detected and processed visualizing the percentage of oxyhemoglobin at each pixel detector and presented continuously, in real time, at 3 images per second. As a proof of principle application, kidneys of four live anesthetized pigs were imaged before, during, and after renal vascular occlusion. DLP Hyperspectral Imaging with active spectral illumination detected a 64.73 ( 1.5% drop in the oxygenation of hemoglobin within 30 s of renal arterial occlusion. Producing chemically encoded images at near video rate, time-resolved hyperspectral imaging facilitates monitoring renal blood flow during animal surgery and holds considerable promise for doing the same during human surgical interventions.
O
ptical methods have been aiding clinicians for centuries in detecting, diagnosing, and monitoring disease. For example, the field of microscopy is commonly used in pathology, while specialized surgical microscopes help surgeons perform delicate neurosurgery. Slit lamps and fundus cameras image the eye’s retina, and endoscopes can be inserted through the umbilicus, the navel, or through natural orifices such as the esophagus or rectum reducing the need for large surgical incisions. More recently, the fields of chemical physics, specifically spectroscopy, are being combined with remote satellite imaging technology for delivering an image in which each pixel measures chemical information from the tissue.1,2 Initial systems integrated a spectrometer into the beam path of a microscope for collecting optical spectra and fluorescence information used for assessing tissue, for example, identifying cancerous cells based on their r 2011 American Chemical Society
spectroscopy.35 Interest for imaging vasculature changes perfusing the skin of patients undergoing clinical and surgical procedures led to building a portable and robust system based on solid state electro-optics. The resulting Hyperspectral Imaging System, using visible light, correctly imaged a decrease in the percentage of oxyhemoglobin, HbO2, as a biomarker for visualizing ischemic tissue, modeled by briefly restricting blood flow into a finger.6 Subsequent clinical studies successfully imaged microvascular changes in the oxygenation of hemoglobin in response to nitric oxide inhibition, inhalation, and stimulation.7,8 In addition to absorbing visible light, the metalloprotein hemoglobin also Received: June 13, 2011 Accepted: August 15, 2011 Published: August 15, 2011 7424
dx.doi.org/10.1021/ac201467v | Anal. Chem. 2011, 83, 7424–7430
Analytical Chemistry
ARTICLE
Figure 1. The DLP Hyperspectral Imaging System visualizes, remotely, inherent tissue biochemistry during surgery, in vivo, with chemically encoded images at near video rates. The conceptual system rendering (A) and the physical embodiment (B) illustrates the light path originating from a pressurized 500 W Xe source (1). Broadband white light passes through a slit (2) that is collimated and focused (3) onto a dispersing grating (4) simultaneously illuminating various bandpases of the spectrum (5) onto successive columns of mirrors on the DMD array (6). To create the desired spectral illumination, the intensity of each spectral bandpass is controlled by triggering the appropriate number of micromirrors for each wavelengthdependent column on the DMD reflecting either to a heat sink (7) or to the beam shaping optics (8) illuminating the target tissue (9). Diffuse spectral reflectance images (10) are focused onto a CCD (11), which are digitized and streamed to a laptop computer (12) where they are archived and processed into chemically encoded images.
absorbs near-infrared radiation that varies as it binds with oxygen and is used for imaging tissue oxygenation from deeper within the tissue.9 Monitoring wound healing in below the knee amputations using spectral ranges within visible (520 to 645 nm) and near-infrared (650 to 1050 nm) visualized changes in the percentage of HbO2 from the surface microvasculature and the deeper blood vessels, respectively.10 Diffusely reflected hyperspectral imaging has also been used to quantify the concentration of biologically important chromophores such as water and fat,11 which has been used for imaging tissue hydration changes while monitoring wound healing postamputation.12 Other studies using similar hyperspectral imaging methods have found it helpful in monitoring diabetic foot ulcers,13,14 shock,15 burn depth evaluation,16 renal vasculature,17,18 and endoscopic bile duct visualization.19,20 In the future, imaging water could also enhance the accuracy of diagnosis of early and effective fasciotomy in treatment of extremity compartment syndrome helping reduce mortality and amputation rates from civilian accidents and combat casualties.21 Imaging tissue blood flow during surgery has many important clinical applications. For example, in the operating room, surgeons cannot sacrifice viable tissues when removing tumors; however, they have little guidance in determining what tissue is viable. Imaging tissue blood flow or oxygenation as a positive indicator of viable tissue would aid in their decision-making process. Currently, methodologies are limited by the need to insert probes into tissue, injection of contrast agents such as fluorescein, or point spectroscopy measures transmitting light through tissue.2224 All of these methods require a probe being invasively inserted or contacting the tissue, which are both timeconsuming and difficult and have serious limitations, making them impractical intraoperatively and thus used infrequently. As such, there is a pressing need to develop real time imaging technologies visualizing tissue oxygenation. Clinical hyperspectral imaging systems typically collect hundreds of contiguous narrow bandpasses of light in sequence
that are formatted into a 3-dimensional hyperspectral image cube consisting of two spatial dimensions and one spectral dimension.25 The data can be viewed either as a series of individual wavelength-dependent images or by plotting a spectrum from each image pixel. The measured hyperspectral image cube is then analyzed using chemometric methods, which involve multivariate mathematics specific for determining chemical composition and quantity within the tissue being studied.7,21,25 Alternatively, the hyperspectral image cube can be projected as a 2-dimensional image from which spectral information can be measured, thus creating an optical tissue phantom.26 Clinicians have expressed the need for seeing better;21,27 however, existing hyperspectral imaging systems are seriously limited for clinical utility because they are incapable of producing chemically encoded images at or near video rates. Capitalizing on the unique ability of digital light processing (DLP) technology to produce complex spectral illumination at the extremely fast switching rates of 12.5 kHz has enabled us to build a robust hyperspectral imaging system, overcoming obstacles limiting previous systems, making it practical for clinical utility, allowing surgeons and clinicians to improve care, and bringing better health care to everyone. Using the novel approach of actively illuminating with light composed of complex spectra for reflecting the chemical composition from within the tissue produces 3 chemically encoded images per second, near video rates, making hyperspectral imaging practical for improving surgical safety, reducing the risk for clinical complications and associated financial liability.
’ EXPERIMENTAL SECTION Instrument Design and Construction. Our novel hyperspectral imaging system design consists of a source projecting active spectral illumination, an imaging detector, and software synchronizing the source, detector, and spectroscopic deconvolution algorithms which produce chemically encoded images at 7425
dx.doi.org/10.1021/ac201467v |Anal. Chem. 2011, 83, 7424–7430
Analytical Chemistry
Figure 2. A scaled down example of DMD micromirror array (A) and measured spectral illumination (B). Various bandpasses of diffracted light illuminate associated DMD micromirror columns. Triggering a predetermined number of micromirrors within the columns controls the wavelength intensity illuminating the target. White squares represent “on” mirrors, reflecting light to the illumination optics from the DMD, while gray squares are “off” mirrors, reflecting light to a heat sink. Combining the individual bandpasses at their programmed intensity generates light with a measured spectrum illuminating the tissue.
near video rates. Figure 1A illustrates the light path and components for the Visible DLP Hyperspectral Imaging System. The integrated device (Figure 1B) consists of a DLP-based spectral illuminator, a focal plane array detector, and a graphical user interface, GUI. The spectral illuminator, an OL-490 Agile Light Source (Optronic Laboratories, Orlando, FL), utilizes a 768 by 1024 digital micromirror device, DMD, the basis for digital light processing, DLP, technology (Texas Instruments, Dallas, TX). The OL-490 can be programmed to emit light at a single bandpass whose center wavelength and full width half-maximum can be varied into simple bell shaped pass bands or shaped into a complex spectrum that is steady or varying over time at rates of up to 12 500 different spectra per second. The visible OL-490 used for this work has a spectral range of 380 to 780 nm, a spectral resolution of 5 nm bandpass measuring at the full width at halfmaximum, fwhm, and a spectral accuracy of 1 nm when a 150 μm slit is used. The OL-490 uses a 500 W Xenon arc lamp emitting broadband light that is passed through a micrometer slit. Various slit widths ranging from 150 to 750 μm can be used, depending on the application requirements. Smaller slit widths were found to result in narrower bandpasses but reduced illumination intensity. After passing through the slit, light is collimated and dispersed off a diffraction grating, separating the light into a plurality of bandpasses. The diffracted light strikes a DMD in such a way that the discrete wavelengths of light fall on associated columns of micromirrors. Programming the appropriate number of micromirrors in select columns to reflect toward the output illumination optics convolutes the individual bandpasses of light to illuminate the tissue with a precisely predetermined known spectrum. For example, as shown in Figure 2, a scaled down example of a DMD array, columns of micromirrors correspond to wavelengths ranging from 380 to 780 nm. Varying the number of micromirrors within a column changes the illumination intensity of that column’s associated wavelength. A series of optics focuses the light onto a flexible, 3 m long, 5 mm core diameter liquid light guide and a beam shaping optic that illuminates the subject with precisely tuned spectroscopic light. Diffuse spectral reflectance from the tissue is focused by a standard 50 mm lens (Nikon, Melville, NY) onto a charged coupled device (CCD) Focal Plane Array detector. Where wavelength-dependent image data is detected, digitized, and streamed to a computer for storage and further processing. The imaging detector, a CoolSNAP HQ2 (Princeton Instruments, Trenton, NJ) using a Sony ICX285
ARTICLE
CCD with a 1392 1040 array and 6.45 μm square pixel detectors, provides high spatial resolution and binning capability for optimizing sensitivity. A 20 MHz, 14 bit analog to digital converter, positioned next to the CCD on the board of the CoolSNAP, minimizes the time for digitizing the spectroscopic images. Electronic shuttering minimizes camera vibration, providing high shutter speeds while minimizing movement artifacts in the image. A Dell Latitude D630 laptop computer (Round Rock, TX) with an Intel Core 2 Duo 2.6 GHz processor and 4 GB ram is used for managing the electro-optics and digital signal processing algorithms. A custom computer program written in C sharp (AAVA Technologies, Plano, TX) synchronizes the spectral illumination of the DMD with the CCD detector, archives the raw hyperspectral image data, and applies novel digital signal processing and chemometric methods for imaging the tissue’s inherent biochemical composition at near video rate. System Characterization. The operating characteristics of the DLP Hyperspectral Imaging system were evaluated for parameters typically used during human surgery. Meaning, all unsterilized equipment is required to be at least 3 feet from the patient according to operating room guidelines to which all instrument parameters were determined accordingly. At this distance, a 350 μm slit provides the best combination of illumination intensity and spectral resolution. The system was characterized by programming the OL-490 (serial number 08000003) to illuminate with all column mirrors turned on and sequentially incrementing the center wavelength every 10 nm from 380 to 780 nm. OL-490 irradiance was measured at the LLG using a 50 mm integrating sphere and an Ocean Optics USB2000+ visible (350 to 1000 nm) spectrometer with a 50 μm UV/vis Premium Fiber and SpectraSuite capture and display software (Ocean Optics, Dunedin, FL). Figure 3A depicts the measured spectral profile of the OL-490 having a spectral range of 380 to 780 nm, as specified by the manufacturer, with an irradiance intensity of 71% or greater between 510 and 625 nm and at least 10% between 420 and 750 nm. Wavelength Calibration. A calibration equation maximizing tuning accuracy was determined (eq 1). The expected center wavelengths, λp, were sent electronically to the OL-490 and plotted against the irradiating center wavelengths measured experimentally, λm. The measured center wavelength was determined as the wavelength bisecting spectral separation between two wavelengths that are 50% of the maximal bandpass value. Linearly regressing the experimentally measured versus the expected, determined a calibration equation, with a correlation of R2 = 0.99. λM ¼ 1:00ðλp Þ þ 0:52
ð1Þ
On average, this OL-490 was found to tune 0.52 nm below the expected wavelength, which is well within the manufacturer’s specification of (1 nm. The bandwidth characteristics were measured at the fwhm which was found to be a constant 8.55 ( 0.23 nm (mean ( standard deviation) when averaged over the spectral range, which is 1.45 nm narrower than the manufacturer specification of 10 nm, Figure 3B. A wavelength resolution of 0.39 nm was determined by the 400 nm spectral range that is dispersed over 1024 columns of micromirrors. Irradiance Stability. System irradiance was monitored over a two hour period using a photodiode (Hamamatsu, Bridgewater, NJ) calibrated at the National Institute of Standards and Technology, NIST, Figure 3C. The OL-490 was fitted with a 350 μm slit and programmed to output a 10 nm fwhm bandpass centered at 7426
dx.doi.org/10.1021/ac201467v |Anal. Chem. 2011, 83, 7424–7430
Analytical Chemistry
ARTICLE
Figure 4. Color tile characterization, a color tile standard placed in the field of view of the DLP Hyperspectral imager, which measured its normalized reflectance spectrum, O, and compared it to the normalized spectrum measured by NIST, b. Averaging over the entire spectral range, the spectra deviated by 3.2% with a correlation coefficient of R2 = 0.99.
Figure 3. Characterization of the OL-490 agile light source (SN: 08000003) utilizing a 350 μm slit. A set of wavelength-dependent column micromirrors are positioned to reflect a bandpass from the grating and slit. The bandpasses emitting from sequentially activated columns of wavelength-dependent micromirrors from the 768 by 1024 DMD micromirror array are measured and plotted every 10 nm, determining the spectral range, 380 to 780 nm, with a spectral accuracy of 0.52 nm (A). The bandwidth characteristics measured at the fwhm of the bandpass are shown, dots, at every 10 nm as the spectral illuminator is tuned through its spectral range (B) yielding an overall average bandwidth of 8.55 ( 0.23 nm, which is well below the manufacturer specification of 10 nm, solid line. A photodiode was used for testing the stability of the source over time (C). Steady state was reached within 2535 min after which the irradiance fluctuated (0.12%.
580 nm, the median wavelength spanning the spectral range of the photodiode. The LLG of the OL-490 is interfaced 9.1 cm from an aspherical lens assembly having a 15 light shaping diffuser (Thor Laboratories, Newton, NJ) that was 3.5 cm from the NIST calibrated photodiode. The analogue signal is amplified with a DLPCA-200 amplifier (FEMTO Messtechnik GmbH, Berlin) and digitized with an NI USB-6216 DAQ and LabView (National Instruments, Austin, TX), sampling once every second for two hours. Integrating the signal accumulated by the photodiode every minute indicated the source reaches steady state
2535 min after igniting the Xe plasma. Steady state was determined as 99.53 ( 0.12% of the maximal plateau value found after 35 min. To achieve the most stable results, this measure recommends collecting experimental data 35 min after igniting the Xe plasma, and turning off surgical and room lights. Color Tile Standardization. Figure 4 depicts normalized reflectance spectra of a standard ceramic color tile (Avian Technologies, Wilmington, OH). The standard tile was placed in the field of view of the DLP hyperspectral imager and actively illuminated with 46 sequential bandpass illuminations spanning 460 to 640 nm at 4 nm increments. Reflected wavelengthdependent images of the tile are collected, and the resulting spectra of 225 pixels are sampled, averaged, and plotted. A comparison of a reflectance spectrum of one of the tiles measured by the DLP Hyperspectral Imager, the circles, with one provided by NIST, the solid dots, found them to be positively correlated, R2= 0.99, and deviating 3.2% when averaging over the spectral range. Field of View and Spatial Resolution. A calibrated USAF 1951 resolution target (Newport, Irvine, CA) is placed 0.91 m from the illuminating beam shaping optic and 50 mm Nikon lens of the detector, with the aperture opened to an f-stop of 1.4. Spatial resolution is limited by diffraction occurring at various locations of the system, for example, the tissue, electro optics, and interference fringes28 of the complete system, and can be computed.29 Experimentally, spatial resolution is defined here as the minimum distance between two lines that can be resolved. This was determined by percent contrast until reaching Rayleigh’s minimum distance criterion, as described in previous work.25 Binning the detector, the practice of electronically merging charge from adjacent pixel detectors on the focal plane array prior to digitization using the on-chip circuitry, improved detector sensitivity measured by shorter exposure times but decreased image resolution. For typical surgical parameters, the field of view was found to be 12.7 cm in diameter, and for binning 1 1, 2 2, 3 3, and 4 4 detector pixels, the exposure times were 11.32, 1.08, 0.60, 0.26 ms while the spatial resolution was determined to be 0.13, 0.18, 0.28, and 0.36 mm per line, respectively. Data Acquisition and Processing Time. The system can be programmed to illuminate the tissue with a series of separate bandpass spectra, which for this experiment consisted of 126 different center wavelengths. The measured reflectance spectrum at each image pixel can be plotted or deconvoluted using supervised multivariate chemometrics into the percentage of oxyhemoglobin, 7427
dx.doi.org/10.1021/ac201467v |Anal. Chem. 2011, 83, 7424–7430
Analytical Chemistry the sequential bandpass illumination method. Alternatively, the micromirrors of the digital micromirror array can also be programmed to illuminate the tissue with a precisely predetermined continuous complex spectrum. For this experiment, the complex spectral illuminations were derived from commonly available spectra for oxyhemoglobin, HbO2, deoxyhemoglobin, Hb, and the broad band spectrum of the source.30 Their respective reflectance images are stored and processed, and a resulting chemically encoded image for the percentage of HbO2 at each image pixel is presented to the computer screen, the complex spectral illumination method. It was found that in both methods (sequential bandpass illumination and complex spectral illumination), when acquiring a hyperspectral data cube, the greatest amount of time is spent acquiring data (9072 ms vs 216 ms) followed by processing the data (8432 ms vs 216 ms) and finally exposing the focal plane array (5040 ms vs 6 ms), respectively. For a more detailed description please see the Supporting Information section. Visualizing Tissue Oxygenation in Vivo. As part of an ongoing animal study approved by the Institutional Animal Care and Use Committee at UT Southwestern Medical Center (Protocol#2008-0251) for exploring different renal vascular clamping methods, the DLP Hyperspectral Imager was used to image spatial profiles of oxygenation before, during, and after occluding the renal artery of a live anesthetized porcine kidney, in order to determine the best surgical outcome. Yorkshire swine (∼75 lbs) were sedated using an intramuscular injection of tiletamine Hcl and zolazepam (6 mg/kg) and then placed under general inhalational anesthesia using 0.53% isoflurane. During anesthesia, the fraction of inspired oxygen was maintained at 100%. Once anesthetized, a surgeon surgically exposes the kidney, and using visible hyperspectral illumination, the kidney was imaged. The corresponding wavelength-dependent reflectance images are formatted into a hyperspectral image cube containing a visible spectrum at each image pixel, which can be archived into a medical tissue database. A multivariate least-squares method determines the percent HbO2 from the spectrum measured at each detector image pixel.25 Sampling a 9 9 square pixel area from the digital images of the kidney, chemically encoded for the percentage of HbO2, provides a quick, noninvasive optical biopsy method. These 81 sampled pixels are averaged to determine a mean ( standard deviation for the oxygenation of blood perfusing the kidney tissue before and after the renal artery and venous occlusion. Using the complex spectral illumination method results in visualizing percentages of HbO2 perfusing the tissue at near video rate. Sampling image pixels, an optical biopsy explained previously, can be evaluated statistically at near video rates during surgery, aiding in evaluating tissue health and oxygenation, painlessly and without the need for removing tissue.
’ RESULTS AND DISCUSSION For the purposes of this work, hyperspectral imaging is defined as the process of implementing multiple spectra composed of multiple wavelengths for determining the chemical nature of a scene. Specifically, the DLP Hyperspectral Imaging system described here is a platform technology applicable to a variety of clinical venues. This results in real time mapping and visualization of tissue oxygenation, assisting surgeons and physicians in making more informed decisions quickly during the high stress of surgery. DLP technology has the unique ability to project precisely
ARTICLE
Figure 5. DLP Hyperspectral Imaging using serial bandpass illumination was collected before occluding the renal artery (A), 60 min after arterial occlusion (B), and 30 min after removing the restriction and reperfusing the kidney with blood (C). Images are chemically encoded for the relative percentage of oxygenated hemoglobin as indicated by the color bar. Optically sampling the kidney from the area indicated by the circle and averaging 81 spectra within that area are depicted for the associated control, occlusion, and reperfusion conditions (A1, B1, and C1).
predetermined complex spectral illumination, which is not possible using current bandpass-based optical methods. Utilizing the much faster and DLP-based system has enabled us to build a hyperspectral imaging system that operates in near real time, making the technology practical for clinical and surgical use. This novel ability of DLP technology to illuminate tissue with complex spectra reduces the need for collecting hundreds of wavelengthdependent images, thereby increasing the speed at which images are encoded for the percentage of oxyhemoglobin by a factor of 111 times faster than systems used in previous work.10,19,25 In Vivo Determination of Tissue Oxygenation Measurement During Renal Ischemia. As a proof of principle study, the utility of DLP hyperspectral imaging was established for intraoperative monitoring of kidney tissue oxygenation in response to renal vascular occlusion. Figure 5 depicts chemically encoded images determined using the sequential bandpass illumination method. In this experiment, a series of 126 bandpass illuminations were used. Relative contribution of oxyhemoglobin within the kidney was determined from the measured spectrum at each image pixel through a multivariate least-squares analysis.25 Sampling pixels of the kidney from the image color encoded for the percentage of oxyhemoglobin, the area indicated by the black circles in Figure 5AC reveals its average spectra as plotted in Figures 5A1C1. Prior to occluding the renal artery and vein, the kidney reflects a well oxygenated oxyhemoglobin spectrum, Figure 5A1, which when deconvoluted corresponds to 71.33 ( 0.78% oxyhemoglobin, (mean ( std) and is color encoded as red pixels, Figure 5A. Figure 5B1 shows an average deoxygenated hemoglobin spectrum measured sixty minutes postvascular occlusion, indicating tissue oxygenation was reduced to 43.52 ( 2.86% HbO2 (color-encoded as a light blue hue), Figure 5B. Physiologically, hemoglobin within the tissue became deoxygenated in response to renal artery and venous occlusion shunting blood flow away from the kidney while tissue metabolism and oxygen demand remained unaltered. After removing the restriction 7428
dx.doi.org/10.1021/ac201467v |Anal. Chem. 2011, 83, 7424–7430
Analytical Chemistry
ARTICLE
Figure 6. Regional ischemia was not seen using a standard clinical camera (A) as visualized by the chemically encoded hyperspectral images. After occluding the renal artery, DLP Hyperspectral Imaging using complex spectral illumination indicated the lower pole had become ischemic (B) while the upper pole was well oxygenated. Upon further inspection, an anteriorly placed secondary artery was discovered, which when clamped resulted in deoxygenating the entire kidney (C).
from the renal artery and vein, blood reperfused the kidney, and thirty minutes later, a partially oxygenated hemoglobin spectrum was measured, Figure 5C1, corresponding to 76.09 ( 2.07% HbO2 (color-encoded as a dark red hue), Figure 5C. The measured reperfusion spectrum indicated hemoglobin having increased oxygen saturation, a typical response following a period of ischemia that is associated with a transient increase in blood flow due to a buildup of vasodilator metabolites released by the vasculature in response to the ischemia, reactive hyperemia.31 Regional Ischemia Visualized. The simplicity and speed of the use of complex spectral illumination allows the clinician to monitor the tissue continuously at 3 frames per second capturing unexpected events and imaging dynamic changes in blood flow. For example, using a standard Satinsky vascular clamp, the surgeon occluded the renal artery of a pig expecting blood to be shunted away from the kidney as in the previous trials. Immediately after clamping the main renal artery, real-time hyperspectral imaging using our complex spectral illumination method visualized ischemic changes in only a portion of the kidney which was not indicated by direct visualization, Figure 6A. Specifically, the lower pole of the kidney was blue, Figure 6B, indicating approximately 43% oxygenated hemoglobin as expected when occluding the renal artery. Surprisingly, the upper pole of the kidney contained on average 60% oxyhemoglobin, visualized as a yellow area. Physically, re-examining the vascular clamp and kidney revealed the presence of an obscured anteriorly placed secondary artery perfusing the upper pole of the kidney, which when occluded resulted in the entire kidney becoming ischemic, Figure 6C. Releasing the clamp from the secondary artery while keeping the primary artery occluded indicated blood reperfused the upper lobe within 1 s. Time Resolved Optical Biopsy. Figure 7 plots a time progression monitoring the level of oxygenated hemoglobin within the kidney tissue before and during renal arterial occlusion. A group of pixels representing a small area of kidney tissue is sampled as indicated by the box on Figure 7A,B, a digital optical biopsy. The complex spectral illumination method visualizes the percentage of oxyhemoglobin at each image pixel at a rate of
Figure 7. Time resolved optical biopsy monitors changes in the percentage of oxygenated hemoglobin perfusing the kidney tissue before and during renal arterial occlusion at 3 chemically encoded images per second. Restricting the renal artery and vein, at the time indicated by the vertical line on the time plot, shunts blood away from the kidney. The DLP Hyperspectral Imaging System utilizing the complex spectral illumination method provides an image encoded for the percentage of oxyhemoglobin every 0.33 s. As an example, two of these images, A and B, are depicted, illustrating the spatial sampling location on the kidney, square box, and the value for the average of the 169 pixels sampled within the box and time sampled on the time plot, C.
three images per second. Sampling pixel values from each image can be plotted to monitor, in real time, changes in renal tissue oxygenation resolved over time. Specifically, 169 pixels from each oxymetric image are sampled, averaged, and plotted over time, Figure 7C. The data is normalized by determining the ratio between the control value and experimental values over time. For 7429
dx.doi.org/10.1021/ac201467v |Anal. Chem. 2011, 83, 7424–7430
Analytical Chemistry this plot, each kidney serves as its internal control, which is the averaged sampled pixels collected prior to occluding the renal artery. Note the optical biopsy location corresponding to each time point on the plot was sampled consistently within the same tissue location over the entire monitoring period. Thirty to thirty five seconds after occluding the renal artery, the hemoglobin saturation decreased sigmoidally to 64.73 ( 1.57% of its unclamped value, Figure 7C. The nonlinear trend is most likely a result of cooperative binding effects between oxygen and hemoglobin as described by hemoglobin’s oxygen dissociation curve.32
’ CONCLUSION We have developed, characterized, and demonstrated proofof-principle results for a novel visible-DLP hyperspectral imaging system. This system is capable of active spectral illumination that facilitates rapid hyperspectral imaging in real time. This technological advance in imaging speed resulted in a system that is fast enough to be used during human surgical procedures. Hyperspectral Imaging systems obtaining images using serial bandpass illumination have been too slow to be used during human surgery because prolonged image acquisition times caused substantial motion artifacts. The system developed in the laboratory synchronizes the spectral illuminator with an imaging detector that collects and digitizes diffuse reflectance hyperspectral images that are processed, resulting in oxymetric encoded images. Using DMD technology allows the option to illuminate with either a series of simple bandpasses or complex spectra. Light reflected by the tissue is processed using supervised chemometric algorithms that quantify the level of oxygenated hemoglobin at each image pixel. Illuminating with bandpass spectra over a spectral range facilitates characterization of tissue-specific spectra that subsequently can be used for determining the chemical composition of tissue, measured noninvasively and in vivo. DLP technology makes it possible to illuminate the tissue with complex spectra producing chemically encoded images at near video rates, offering a practical solution for monitoring blood flow and visualizing tissue oxygenation in real time, as is needed during surgery. ’ ASSOCIATED CONTENT
bS
Supporting Information. Additional information as noted in text. This material is available free of charge via the Internet at http://pubs.acs.org.
’ AUTHOR INFORMATION Corresponding Author
*Address: Digital Light Innovations, DLi, 4501 Spicewood Springs, Rd., Suite 1000, Austin, Texas 78759. E-mail: kzuzak@ DLinnovations.com. Fax: 512-615-4635.
’ ACKNOWLEDGMENT Texas Instruments is acknowledged for providing major funding for developing the DLP Hyperspectral Imaging System. DLP and the DLP logo are registered trademarks of Texas Instruments. Further funding was also provided by the HudsonPenn endowment and the Smith endowments located at the University of Texas at Southwestern Medical Center. Finally, the study was supported in part by Federal funding from the Department of Energy to the University of Texas at Arlington.
ARTICLE
’ REFERENCES (1) Goetz, A. F. H.; Vane, G.; Solomon, J. D.; Rock, B. N. Science 1985, 228, 1147–1153. (2) Gat, N. Proc. SPIE 2000, 4056, 50–64. (3) Fernandez, D. C.; Bhargava, R.; Hewitt, S. M.; Levin, I. W. Nat. Biotechnol. 2005, 23, 469–474. (4) Georgakoudi, I.; Jacobson, B. C.; Muller, M. G.; Sheets, E. E.; Badizadegan, K.; Carr-Locke, D. L.; Crum, C. P.; Boone, C. W.; Dasari, R. R.; Dam, J. V.; Feld, M. S. Cancer Res. 2002, 62, 682–687. (5) Wachman, E. S.; Niu, W.; Farkas, D. L. Biophys. J. 1997, 73, 1215–1222. (6) Zuzak, K. J.; Schaeberle, M. D.; Levin, I. W.; Lewis, N. E.; Freeman, J.; McNeil, J. D.; Cancio, L. C. IEEE 1999, 2, 1118. (7) Zuzak, K. J.; Schaeberle, M. D.; Gladwin, M. T.; Cannon, R. O., III; Levin, I. W. Circulation 2001, 104, 2905–2910. (8) Zuzak, K. J.; Gladwin, M. T.; Cannon, R. O., III; Levin, I. W. Am. J. Physiol. Heart Circ. Physiol. 2003, 285, H1183–H1189. (9) Shaw, R. A.; Mansfield, J. R.; Kupriyanov, V. V.; Mantsch, H. H. J. Inorg. Biochem. 2000, 79, 285–293. (10) Zuzak, K. J.; Hariharan, S.; Naik, S. C.; Mandhale, M.; Livingston, E. H. IEEE 2007, 9–11. (11) Pham, T. H.; Bevilacqua, F.; Spott, T.; Dam, J. S.; Tromberg, B. J.; Andersson-Engels, S. Appl. Opt. 2000, 39, 6487–6497. (12) Zuzak, K. J.; Perumanoor, T. J.; Naik, S. C.; Mandhale, M.; Livingston, E. H. IEEE 2007, 16–18. (13) Greenman, R. L.; Panasyuk, S.; Wang, X.; Lyons, T. E.; Dinh, T.; Longaoria, L.; Giurini, J. M.; Freeman, J.; Khaodhiar, L.; Veves, A. Lancet 2005, 366, 1711–1717. (14) Khaodhiar, L.; Dinh, T.; Schomacker, K. T.; Panasyuk, S. V.; Freeman, J. E.; Lew, R.; Vo, T.; Panasyuk, A. A.; Lima, C.; Giurini, J. M.; Lyons, T. E.; Veves, A. Diabetes Care 2007, 30, 903–910. (15) Cancio, L. C.; Batchinsky, A. I.; Mansfield, J. R.; Panasyuk, S.; Hetz, K.; Martini, D.; Jordan, B.; Tracey, B.; Freeman, J. E. J. Trauma 2006, 60, 1087–1095. (16) Cross, K. M.; Leonardi, L.; Payette, J. R.; Gomez, M.; Levasseur, M. A.; Schattka, B. J.; Sowa, M. G.; Fish, J. S. Wound Repair Regener. 2007, 15, 332–340. (17) Crane, N. J.; Gillern, S. M.; Tajkarimi, K.; Levin, I. W.; Pinto, P. A.; Elster, E. A. J. Urol. 2010, 184, 1279–1285. (18) Crane, N. J.; McHone, B.; Hawksworth, J.; Pearl, J. P.; Denobile, J.; Tadaki, D.; Pinto, P. A.; Levin, I. W.; Elster, E. A. J. Am. Coll. Surg. 2008, 206, 1159–1166. (19) Zuzak, K. J.; Naik, S. C.; Alexandrakis, G.; Hawkins, D.; Behbehani, K.; Livingston, E. H. Anal. Chem. 2007, 79, 4709–4715. (20) Zuzak, K.; Naik, S.; Alexandrakis, G.; Hawkins, D.; Behbehani, K.; Livingston, E. Am. J. Surg. 2008, 195, 291–497. (21) Cancio, L. C. Proc. SPIE 2010, 7596, 5.1–5.8. (22) Vanderkooi, J. M.; Erecinska, M.; Silver, I. A. Am. J. Cell Physiol. Cell Physiol. 1991, 260, C1131–C1150. (23) Leung, F. W.; Morishita, T.; Livingston, E. H.; Reedy, T.; Guth, P. H. Am. J. Physiol. Gastrointest. Liver Physiol. 1987, 252, G797–G804. (24) Severinghaus, J. W.; Astrum, P. B. J. Clin. Monit. 1986, 2, 60–73. (25) Zuzak, K. J.; Schaeberle, M. D.; Lewis, E. N.; Levin, I. W. Anal. Chem. 2002, 74, 2021–2028. (26) Rice, J. P.; Allen, D. W. Proc. SPIE 2009, 7334. (27) Livingston, E. H. Proc. SPIE 2010, 7596, 2.1–2.9. (28) Bunting, C. A.; Carolan, P. G.; Forrest, M. J.; Noonan, P. G.; Sharpe, A. C. Rev. Sci. Instrum. 1988, 59, 1488–1490. (29) Shirley, E. L. Appl. Opt. 2004, 43, 2609–2620. (30) Zijlstra, W. G.; Buursma, A.; Van Assendelft, O. W. Visible and Near Infrared Absorption Spectra of Human and Animal Haemoglobin; VSP BV: The Netherlands, 2000; pp 62. (31) Coffman, J. D.; Gregg, D. E. Am. J. Physiol. 1960, 199, 1143–1149. (32) Berne, R. M.; Levy, M. N. Principles of Physiology, 3rd ed.; Mosby Year Book: St. Louis, MO, 2000; pp 335.
7430
dx.doi.org/10.1021/ac201467v |Anal. Chem. 2011, 83, 7424–7430
