DOI: 10.1148/radiol.2452061613
(Radiology 2007;245:523-531.)
© RSNA, 2007
Quantitative Real-time Catheter-based Fluorescence Molecular Imaging in Mice1
Rabi Upadhyay, BS,
Rahul A. Sheth, AB,
Ralph Weissleder, MD, PhD, and
Umar Mahmood, MD, PhD
1 From the Center for Molecular Imaging Research, Massachusetts General Hospital, Harvard Medical School, 185 Cambridge St, Simches 8226, Boston, MA 02114. Received September 18, 2006; revision requested November 16; revision received December 8; accepted December 21; final version accepted March 1, 2007. Supported in part by National Institutes of Health grants RO1-EB001872 and R24-CA92782 and by a grant from the Dana-Farber/Harvard Cancer Center Technology Innovation Fund.
Address correspondence to U.M. (e-mail: mahmood{at}helix.mgh.harvard.edu).
 |
ABSTRACT
|
|---|
Purpose: To prospectively evaluate an optical imaging system designed to perform quantitative, intravital catheter-based imaging of fluorescent molecular probes.
Materials and Methods: This study was performed according to a protocol approved by the institutional animal care committee. A fiberoptic catheter imaging system was developed to implement a normalization algorithm for real-time quantitative near-infrared (NIR) imaging. The system was validated with in vitro imaging of fluorochrome phantoms and in vivo fluorescence measurements obtained in tumors implanted in murine abdomens (n = 7) after administration of an enzyme-activatable NIR probe. Standard analysis of variance tests were used to determine significant dissimilarities in signal from distinct fluorochrome concentrations. The clinical utility of the system was further evaluated by imaging orthotopically implanted murine colonic adenocarcinomas (n = 4).
Results: Raw NIR fluorescence intensities, which were measured with a fiberoptic catheter placed above wells of varying NIR fluorochrome concentration, varied markedly (>100%) with catheter position, while the corrected NIR signal was confined to a range of values within 10% of their mean for each individual fluorochrome concentration and were significantly distinct (P < .001) between relevant concentration ranges. Similar results were observed for the in vivo measurements from the abdominally implanted tumors, with raw NIR signal varying 20% from the mean and corrected NIR signal varying only 1% from the mean. The colonic studies revealed that the correction method was robust enough for use during minimally invasive imaging procedures.
Conclusion: The authors have developed and implemented a method for quantitative real-time catheter-based fluorescence imaging that resolves NIR signal dependence on changes in catheter position.
© RSNA, 2007
|
INTRODUCTION
|
|---|
Optical imaging with fluorescent probes is a rapidly expanding modality that has the potential to substantially affect the diagnostic and therapeutic realms, as well as the realm of basic science (1). Advances in probe technology, including the development of protease-activatable near-infrared (NIR) probes, have strengthened the prospect of using molecularly selective beacons to focally highlight disease (2,3). Imaging systems that rely on direct surface illumination to measure fluorescence signal have been developed to aid in the detection of such activatable or targeted probes (4,5) and have been successful in imaging animal models of some of the most widespread human diseases. However, these epifluorescence systems often require the tissue being investigated to be within 1 cm of the surface or to be explanted prior to imaging; therefore, they have somewhat less direct clinical utility.
Minimally invasive NIR measurement devices with greater clinical utility exploit fiberoptic catheters to intravitally perform surface-weighted imaging that overcomes the depth limitations of epifluorescence systems by directly approaching the surface of interest (6). Such catheter-based systems allow simultaneous video capture of both white light (WL) and NIR signal, enabling the user to identify anatomic landmarks in the WL channel and disease processes on a molecular scale in the NIR channel. These systems have successfully extended the capabilities of epifluorescence imaging to include, among other disease applications, the intravital detection of intraperitoneal tumors, ovarian carcinomas (6), colonic adenocarcinomas (7), and peripheral lung tumors (8). In the future, interventional radiologists may directly exploit such improved fluorescence-guided minimally invasive sampling, which is analogous to the computed tomography (CT)– or fluoroscopy-guided sampling performed today, to evaluate similar lesions. Compared with anatomic imaging alone, focal fluorescence highlights such areas of disease and may improve sampling rates and decrease procedure time.
One problem with catheter-based systems is that they do not have the advantage of a static controlled imaging environment, in which distances from the illumination source to the target tissue and from the target tissue to the charge-coupled device (CCD) are fixed, that ex vivo fluorescence systems have. In such systems, illumination intensity across the target remains uniform over time, and the fluorescence emission is essentially constant to a first approximation. In contradistinction, for catheter-based systems, these distances become dynamic essentially uncontrollable variables. Photon fluence decreases as the square of the distance between the target tissue and the catheter tip increases, resulting in a marked change in NIR photon counts as one approaches or retreats from the disease being investigated. Moreover, sharp angles of incidence between the catheter and the NIR signal source cause objects closer to the catheter to appear brighter than more distant objects within the same video frame. Hence, there is a large unmet need for quantifying NIR fluorescence intensity, which would markedly increase the clinical utility of catheter systems for use in disease detection with molecularly targeted probes. An additional important benefit would be the ability to further characterize disease on the basis of quantitative assessment of probe activation or target binding.
The fluorescence microscopy community has resolved similar concerns regarding the quantitative ability of their instruments by imaging a uniformly fluorescent reference sample to establish a baseline sample image and then dividing all subsequent images by the reference sample image (9,10). Rather than divide by a constant reference image, however, we hypothesized that pixel-wise division of each NIR image by a simultaneously acquired WL image would allow for a dynamic frame-by-frame normalization that would account for variations in signal intensity between and within individual frames due to changes in catheter position. To implement such a correction algorithm, we developed an in-house software platform that simultaneously acquired, processed, and depicted 12-bit image data from two CCDs. The software also allowed seamless registration, segmentation, gradation (window/leveling), exposure control, and image processing of the data in real time. Thus, the purpose of our study was to prospectively evaluate an optical imaging system that we designed for quantitative intravital catheter-based imaging of fluorescent molecular probes.
|
MATERIALS AND METHODS
|
|---|
This study was performed according to a protocol approved by our institutional animal care committee and conformed to guidelines issued by the National Institutes of Health for care of laboratory animals.
Optics
The optical acquisition system was designed to accept standard fiberoptic catheters and was conceptually similar to previous designs (Fig 1a). The catheter used for all experiments was a 1.6-mm-diameter fiberoptic bundle with extrinsic excitation fibers (Edwards LifeSciences, Irvine, Calif) that included a 0.9-mm instrument channel and had a 100-cm working length. The optical train, including the cameras, was constructed from standard parts, and the entire system occupied less than a 1-cu-ft space (smaller than 0.3 m3). Modular mechanical parts were obtained from Edmund Industrial Optics (Barrington, NJ) and Thorlabs (Newton, NJ), and all optical pieces were obtained from Omega Optical (Brattleboro, Vt).
View larger version (25K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 1: Schematic of optics and software pipeline. A, Emission light from a fiberoptic catheter is optically separated by a dichroic mirror into WL and NIR light beams. These are focused onto CCD cameras with additional NIR filtering through a bandpass filter. The 12-bit data streams reach the computer via 100-Mbit connections. B, Incoming data are managed completely in random access memory to achieve real-time latency. Fully processed 8-bit data are depicted in real time, while all raw data and image processing parameters are also recorded on disks.
|
|
Fluorescence excitation and WL illumination originated from a 300-W xenon lamp (SolarMaxx 300; Sunoptics, Jacksonville, Fla) that was filtered through a 680-nm short-pass filter (decreasing heat output and false-positive NIR signal detection) before it entered the endoscope. Light collected by the endoscope was relayed to a 45° dichroic mirror that divided the signal into perpendicular beams of visible and NIR light. Achromatic lenses with a 50-mm focal length focused both beams of light onto the cameras, and an additional bandpass filter in front of the NIR camera further narrowed the NIR signal to the correct fluorescence emission band. The NIR optics of the system were optimized for excitation and emission of cyanine 5.5 (Cy5.5 NHS Ester; GE Healthcare, Chalfont St Giles, England).
Hardware
The imaging hardware included CCD cameras with high sensitivity, high dynamic range, high spatial resolution, and digitally tunable exposure times. A camera (Pixelfly VGA; PCO, Kelheim, Germany) with a color CCD was chosen for WL image capture, and another camera (Pixelfly QE; PCO) with a gray-scale CCD was chosen for NIR signal detection; both cameras have 12-bit acquisition with 16-bit data transfer (Table). An open-source driver and programming interface for the cameras enabled in-house development of custom software.
The software was designed for use with a 3-GHz Intel Pentium 4 computer (Intel, Santa Clara, Calif) with 3 GB of random access memory running the Linux operating system, kernel version 2.4. The code was written in C and C++ language and compiled with the openly available compiler (gcc, version 2.95).
Software
Raw image data were transported to the computer in 16 bits per pixel format and stored in random access memory (Fig 1b). The images were then segmented to exclude pixels outside the fiberoptic image by using the Hough transform technique to identify the circular catheter image within the CCD field of view. Given a registration matrix computed a priori for each catheter, all NIR image frames were rotated, translated, and scaled to match WL image frames with a six-parameter affine transform model. Before each camera was used to acquire a new frame, a new exposure time was calculated on the basis of a histogram of pixel intensities seen on the current image. This autoexposure procedure was used to adjust the exposure time such that the image intensity averaged across the catheter image approached a previously defined set point (usually set at the midpoint of the dynamic range of the CCD).
Each frame had a unique exposure time; therefore, pixel intensities (counts) could not be directly compared between two frames or across two cameras. However, by dividing all the pixels in each frame by the exposure time of that frame, we calculated a counts-per-second value for each pixel that could be readily compared between frames. The linearity of this new value was evaluated empirically. The NIR correction method (detailed later in this article) was then applied by using these new units. These operations were performed in a 32-bit format to maintain numerical precision.
All the raw image data and processing parameters from the registration, segmentation, automatic exposure, and image division methods were saved on a hard drive for archival purposes. All subsequent image quantitations (average signal intensity within a region of interest) for the in vitro and in vivo experiments were performed with these archived data.
In conjunction with data archiving, the custom software processed all three image streams (WL, NIR, and corrected NIR) for real-time visualization by the operator. Since each image stack was in either a 16 bits per pixel (WL and NIR) or a 32 bits per pixel (corrected NIR) format, a window/leveling algorithm was used to map the images to 8-bit images so they could be displayed on a computer monitor. The NIR and corrected NIR images were colorized to pseudocolor images with a color look-up table, and WL images were colorized with a standard Bayer color definition algorithm (tuned to the Red-Green-Blue filter on the WL camera).
NIR Correction Method
A corrected NIR image was created by dividing the raw NIR image by a simultaneously acquired WL image on a pixel-by-pixel basis. Before this occurred, all pixels outside the catheter image (as determined with the segmentation routine) were set to zero on the corrected image. Next, pixels on the WL or raw NIR images with values in the bottom 5% of either image (on the basis of a histogram of the pixels within the catheter image) were set to zero on the division image and excluded. This was done to prevent spuriously high values from arising by dividing low NIR values by low WL values. The background dark current was then subtracted from the pixels of both cameras. After this, pixel counts were converted into counts per second for each frame. Finally, the two resulting counts-per-second values were divided and saved in the corrected NIR image file. This correction method occurred in real time (less than 100 µsec, which was the shortest integration time used for either camera and more than 300 times faster than the video rate). The corrected image was displayed as an additional quadrant along with WL and raw NIR images in the real-time video display.
In Vitro Imaging
Distance and angle-dependence phantom studies were conducted for cyanine 5.5 fluorochrome, which was diluted in whole milk to increase the optical density of the solution and thereby produce meaningful WL signal intensity. As a colloidal dispersion, whole milk mimics the tissue scattering and absorption properties more accurately than do clear solutions. The dilutions were varied from 1 to 20 µM (1–20 µmol/L) cyanine 5.5 to span a realistic concentration range in a 20-g mouse that received 2 nmol of the fluorescent probe. A total of 16 phantom samples were studied. Distances from the surface of the solution were determined by immersing the catheter tip in liquid and then pulling the catheter back in 1-mm increments. A mechanical actuator with a built-in caliper was used to stabilize the catheter above the dilutions and ensure an accuracy of better than 0.1 mm. The tip was cleaned prior to measurement. For angle-based measurements, the catheter was held at a constant height above the liquid surface, and the incident angle of the catheter relative to the liquid surface was varied from 45° to 90°. For both experiments, five repeated measurements were used to generate each data point and error bar.
In Vivo Imaging
An author (R.U.) implanted a murine colon cancer cell line (CT26; American Tissue Culture Collection, Manassas, Va) in the peritoneal surface immediately deep to the anterior abdominal wall of nude mice (n = 7, Taconic, Germantown, NY). In all mice, focal carcinomas formed within 7 days after the procedure. One day prior to imaging, 150 µL (2 nmol/L) of the protease-activatable probe (Prosense; VisEn Medical, Woburn, Mass), a molecule that has fluorescence excitation and emission properties similar to those of cyanine 5.5, was administered intravenously to mice. Cleavage of this probe by proteases such as cathepsins results in a six- to 20-fold increase in fluorescence. As proteases have been shown to be up-regulated in a number of disease processes, this probe serves as a beacon for many disease states (11). The mice were administered a gas anesthetic agent (2% isofluorane, 2 L/min of oxygen), and two authors (R.U., R.A.S.) made a midline incision to expose the abdominal wall with the tumor. The abdominal skin flap on which the tumor had grown was then reflected and secured. The catheter was positioned perpendicular to the posterior surface of the skin flap with a caliper above the tissue as in the in vitro experiments, and the tumor was imaged from distances of 1–5 mm. Images were displayed in real time in WL, NIR, and corrected NIR channels and were saved as 12-bit raw data.
For the orthotopic experiments, a urologist with 6 years of surgical experience used a previously described method (12) to implant the same cell line in the colons of four mice. Similarly, focal carcinomas formed within 7 days, and the protease-activatable probe was administered 1 day prior to imaging. Mice were administered a gas anesthetic agent, the colon was cleansed with saline, and the catheter was inserted through the rectum while air was introduced from the catheter tip through the insufflation channel. Two authors (R.U., R.A.S.) inspected the colon up to the splenic flexure. Images were displayed in real time in WL, NIR, and corrected NIR channels and were saved as 12-bit raw data.
After imaging, the mice were sacrificed with the catheter still inserted; this allowed the appropriate portion of the colon to be explanted for histologic analysis by the urologist. Gross examination of the tissue was performed after explantation to verify the presence of lesions by noting the presence of increased wall thickness at the site of the tumors. The tissue was then embedded in freezing medium, frozen, and cut into 8-µm-thick slices with a cryostat (Leica, Bannockburn, Ill) by two authors (R.U., R.A.S.). Samples were stained with hematoxylin-and eosin and viewed with a standard light microscope to determine if neoplastic tissue was present.
Statistical Analysis
Linear regression was performed with the nonlinear least-squares Marquardt-Levenberg algorithm, and coefficients of determination were computed to indicate the fraction of variance shared by exposure time and signal. All in vitro distance dependence results are presented as means ± standard errors (or as percentage changes ± percentage standard errors) calculated over a constant region of interest. This region of interest (roughly 200 000 pixels) was determined with the segmentation routine of the in-house–designed software package that was used to demarcate the catheter image within the CCD field of view. The same region of interest was used for all images acquired in this experiment. The in vitro angle dependence data represent plot profiles obtained over the same coordinates on all images.
All corrected NIR signals measured during in vitro distance dependence experiments were compared with statistical software (SAS/STAT, version 9.1.3; SAS Institute, Cary, NC) by using one-way analysis of covariance with the distance parameter set as a continuous explanatory variable. Analysis of covariance was used to specifically test whether concentration had an effect on corrected NIR signal; this test accounts for the variance in corrected signal from 1 to 5 mm and effectively adjusts all the data to the middle 3-mm distance. We performed the six pairwise comparisons only if the effect of the concentration was significant (as determined with analysis of covariance). Analysis of covariance with interaction was used to test the interaction between concentration and distance. Finally, analysis of variance was used to compare the corrected NIR signal at each individual distance measurement in all concentration groups.
The in vivo data collected from the abdominally implanted tumors represent average percentage changes from the mean for seven mice. At each distance increment for each mouse, mean values for raw and corrected NIR signal intensity were calculated over a constant region of interest. The same region of interest was used for both NIR images. Average signal intensity was then calculated for raw and corrected NIR images for each mouse across all five distances, and a percentage change from this mean was computed for each distance. These percentage changes were then averaged over all seven mice for each distance and plotted. The error bars represent percentage standard errors. For the in vivo colon imaging results, the color look-up tables for the raw and corrected NIR overlays are based on the percentage change between each pixel value and a baseline value. The same look-up table was used for both raw and corrected NIR images.
|
RESULTS
|
|---|
Exposure times for both cameras varied from frame to frame to maintain a constant average signal intensity; thus, we could not directly compare the raw pixel values of different images. Instead, we normalized the pixels of each image by using the exposure time. We justified this use of counts per second by calculating the mean pixel values within the catheter region of interest over different exposure times while maintaining a constant height from the phantom surface. For both NIR and WL cameras (Fig 2), there was a linear relationship between pixel counts and exposure times. This linearity allowed us to use the ratio of the pixel value to the exposure time (ie, the slopes of the plots [Fig 2]) as an exposure time–independent data value.
View larger version (12K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 2: Graph shows signal intensity data values versus increasing exposure times. Best-fitting lines, as determined with linear regression, are shown. Also shown are coefficients of determination (R2 values) that indicate the fraction of variance shared by exposure time and signal intensity. Linear response allows the use of counts per second as a quantitative parameter. AU = arbitrary units.
|
|
In Vitro Imaging
As the distance between the catheter tip and the surface of fluorescent dye increased, raw NIR pixel counts per second monotonically rapidly decreased. Moreover, the curves for the different concentrations of cyanine 5.5 fluorochrome did not remain within a particular counts-per-second range; rather, they exhibited a large degree of overlap. Dividing by the WL counts-per-second values in a pixel-wise manner (Fig 3) resolved these issues. Not only were the curves for the four fluorochrome concentrations more level (ie, less dependent on the distance from the catheter tip), but they were also confined to a particular range of values. There was a significant difference in corrected NIR signal among the concentrations when we accounted for distance variations with analysis of covariance (P < .001).
There was also a significant difference in corrected NIR signal between all four concentrations when they were compared at each individual distance measurement with analysis of variance (P < .001 for all five distances). The corrected NIR curves stayed within approximately 10% of their mean values, while the raw NIR curves ranged to up to approximately 50% of the mean values (Fig 3). However, each corrected NIR curve retained a significant nonzero slope (P < .001), and the slopes had a significant dependence on concentration (P < .001).
View larger version (34K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 3a: Distance dependence of raw NIR pixel values versus WL-corrected NIR values. (a) Schematic of the experimental design shows the various concentrations of fluorescent dye used in the phantoms. ROI = region of interest. (b) Distance dependence curves for raw NIR counts per millisecond for the concentrations given in a show a large percentage change over the distance of a few millimeters. Error bars denote the standard of error. (c) Distance dependence curves for the corrected NIR pixels for the same concentrations show an approximately flat relationship between corrected signal intensity and distance. The difference in corrected NIR signal intensity between all four concentrations is significant (P < .001), as determined with analysis of covariance. (d) Percentage change from the mean values for raw and corrected NIR. The reduced absolute slope of the corrected data (as compared with the slope of the raw data) demonstrates the reduced dependence of the signal intensity on distance.
|
|
View larger version (14K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 3b: Distance dependence of raw NIR pixel values versus WL-corrected NIR values. (a) Schematic of the experimental design shows the various concentrations of fluorescent dye used in the phantoms. ROI = region of interest. (b) Distance dependence curves for raw NIR counts per millisecond for the concentrations given in a show a large percentage change over the distance of a few millimeters. Error bars denote the standard of error. (c) Distance dependence curves for the corrected NIR pixels for the same concentrations show an approximately flat relationship between corrected signal intensity and distance. The difference in corrected NIR signal intensity between all four concentrations is significant (P < .001), as determined with analysis of covariance. (d) Percentage change from the mean values for raw and corrected NIR. The reduced absolute slope of the corrected data (as compared with the slope of the raw data) demonstrates the reduced dependence of the signal intensity on distance.
|
|
View larger version (13K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 3c: Distance dependence of raw NIR pixel values versus WL-corrected NIR values. (a) Schematic of the experimental design shows the various concentrations of fluorescent dye used in the phantoms. ROI = region of interest. (b) Distance dependence curves for raw NIR counts per millisecond for the concentrations given in a show a large percentage change over the distance of a few millimeters. Error bars denote the standard of error. (c) Distance dependence curves for the corrected NIR pixels for the same concentrations show an approximately flat relationship between corrected signal intensity and distance. The difference in corrected NIR signal intensity between all four concentrations is significant (P < .001), as determined with analysis of covariance. (d) Percentage change from the mean values for raw and corrected NIR. The reduced absolute slope of the corrected data (as compared with the slope of the raw data) demonstrates the reduced dependence of the signal intensity on distance.
|
|
View larger version (16K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 3d: Distance dependence of raw NIR pixel values versus WL-corrected NIR values. (a) Schematic of the experimental design shows the various concentrations of fluorescent dye used in the phantoms. ROI = region of interest. (b) Distance dependence curves for raw NIR counts per millisecond for the concentrations given in a show a large percentage change over the distance of a few millimeters. Error bars denote the standard of error. (c) Distance dependence curves for the corrected NIR pixels for the same concentrations show an approximately flat relationship between corrected signal intensity and distance. The difference in corrected NIR signal intensity between all four concentrations is significant (P < .001), as determined with analysis of covariance. (d) Percentage change from the mean values for raw and corrected NIR. The reduced absolute slope of the corrected data (as compared with the slope of the raw data) demonstrates the reduced dependence of the signal intensity on distance.
|
|
The correction algorithm was used to correct for changes in measured NIR fluorescence intensity due to variations in the angle of incidence between the catheter tip and the surface of the dye. The maxima for the different plot profiles generated (Fig 4) were increasingly skewed away from the center line and toward the light source (ie, the catheter) at sharper angles. The hill-shaped nature of these profiles also demonstrated the undesired effects of shading on signal across the field of view of the catheter. However, division of the two images resolved these issues: The plateau-shaped curves (Fig 4d) indicated a uniform-corrected NIR image, even at sharp angles and at either extreme of the imaging range.
View larger version (31K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 4a: Angle dependence for WL, raw NIR, and corrected NIR. (a) Schematic of the experimental design shows the angles at which the images were obtained. (b) Plot profile of WL images acquired at the angles given in a. A dashed line was drawn through the maxima of each hill-shaped curve to highlight the increasing skew as the catheter was held at more acute angles. The dashed line is in contrast with the solid black line, which is a theoretical line without skew. (c) Plot profile of raw NIR images obtained simultaneously with the WL images in b. A similar skew is shown as the angle of the catheter decreases. (d) Plot profile of corrected NIR images calculated from the WL and raw NIR images shown in b and c. Each plateau-shaped curve correctly represents the uniform fluorescent signal across the well of dye. In addition, each curve is symmetric and shows no skew. AU = arbitrary units.
|
|
View larger version (21K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 4b: Angle dependence for WL, raw NIR, and corrected NIR. (a) Schematic of the experimental design shows the angles at which the images were obtained. (b) Plot profile of WL images acquired at the angles given in a. A dashed line was drawn through the maxima of each hill-shaped curve to highlight the increasing skew as the catheter was held at more acute angles. The dashed line is in contrast with the solid black line, which is a theoretical line without skew. (c) Plot profile of raw NIR images obtained simultaneously with the WL images in b. A similar skew is shown as the angle of the catheter decreases. (d) Plot profile of corrected NIR images calculated from the WL and raw NIR images shown in b and c. Each plateau-shaped curve correctly represents the uniform fluorescent signal across the well of dye. In addition, each curve is symmetric and shows no skew. AU = arbitrary units.
|
|
View larger version (20K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 4c: Angle dependence for WL, raw NIR, and corrected NIR. (a) Schematic of the experimental design shows the angles at which the images were obtained. (b) Plot profile of WL images acquired at the angles given in a. A dashed line was drawn through the maxima of each hill-shaped curve to highlight the increasing skew as the catheter was held at more acute angles. The dashed line is in contrast with the solid black line, which is a theoretical line without skew. (c) Plot profile of raw NIR images obtained simultaneously with the WL images in b. A similar skew is shown as the angle of the catheter decreases. (d) Plot profile of corrected NIR images calculated from the WL and raw NIR images shown in b and c. Each plateau-shaped curve correctly represents the uniform fluorescent signal across the well of dye. In addition, each curve is symmetric and shows no skew. AU = arbitrary units.
|
|
View larger version (22K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 4d: Angle dependence for WL, raw NIR, and corrected NIR. (a) Schematic of the experimental design shows the angles at which the images were obtained. (b) Plot profile of WL images acquired at the angles given in a. A dashed line was drawn through the maxima of each hill-shaped curve to highlight the increasing skew as the catheter was held at more acute angles. The dashed line is in contrast with the solid black line, which is a theoretical line without skew. (c) Plot profile of raw NIR images obtained simultaneously with the WL images in b. A similar skew is shown as the angle of the catheter decreases. (d) Plot profile of corrected NIR images calculated from the WL and raw NIR images shown in b and c. Each plateau-shaped curve correctly represents the uniform fluorescent signal across the well of dye. In addition, each curve is symmetric and shows no skew. AU = arbitrary units.
|
|
In Vivo Imaging
As the catheter was withdrawn 1–5 mm from the surface of the tumor, the raw NIR counts decreased dramatically, exhibiting a ±20% variation from the mean signal across the five distance measurements (Fig 5). On the other hand, the corrected NIR signal demonstrated little variation as the catheter was pulled away from the tumor (±1% change from the mean value).
View larger version (10K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 5: Quantitative analysis of in vivo imaging of implanted abdominal tumors after administration of an activatable NIR probe. Raw and corrected NIR values of seven tumors implanted in the abdominal wall were acquired intraoperatively with a catheter. Curves are shown as percentage change in signal intensity from the mean signal intensity over a 5-mm distance. Error bars denote standard of error. The dashed line remains relatively flat, showing the lack of change in corrected NIR signal intensity as the catheter is moved farther away. This is in contrast to the solid line, which varies substantially in raw NIR signal intensity even though the same tumor is being imaged.
|
|
Colonoscopy was feasible in all mice, and there were no deaths related to this procedure. The two colonic tumor foci were not visible with WL imaging at far or medium distances (Fig 6a). At close range, one of the lesions was barely discernible because of a mild contour irregularity, while the second lesion remained unseen given the small size of the tumor foci. In the raw NIR channel, the lesions were visible because of protease activation of the NIR probe at the site of the tumors. However, the two lesions appeared as one because of the dominance of the closer lesion in the area of the total NIR signal. In this case, the dominance was due to the smaller distance of the proximal lesion compared with the more distal distance of the other lesion from the catheter tip. Moreover, in the raw NIR channel, the lesions appeared dim when they were far away and much brighter when they were close. In contradistinction, (a) both lesions were visible in the corrected NIR channel, (b) they appeared to be of equal signal intensity despite the greater distance to the more distal lesion, and (c) their corrected signal intensity remained relatively constant despite a wide variation in distance. As expected, the lesions appeared larger at smaller distances from the catheter tip.
View larger version (95K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 6a: In vivo colonoscopy video frames show real-time depiction capabilities (similar results seen in all four mice imaged). (a) WL, raw NIR, and corrected NIR images from far, medium (mid), and close distances to the tumor foci. Raw NIR signal intensity initially results in underestimation and then results in overestimation of fluorochrome concentration as distance decreases between the catheter tip and the tumor. Corrected signal intensity shows constant concentration of both adenocarcinoma foci at all distances. (b) Histologic analysis of the lesions depicted in a enabled us to confirm the two small tumor foci (marked by arrows in top section and magnified in bottom sections).
|
|
View larger version (121K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 6b: In vivo colonoscopy video frames show real-time depiction capabilities (similar results seen in all four mice imaged). (a) WL, raw NIR, and corrected NIR images from far, medium (mid), and close distances to the tumor foci. Raw NIR signal intensity initially results in underestimation and then results in overestimation of fluorochrome concentration as distance decreases between the catheter tip and the tumor. Corrected signal intensity shows constant concentration of both adenocarcinoma foci at all distances. (b) Histologic analysis of the lesions depicted in a enabled us to confirm the two small tumor foci (marked by arrows in top section and magnified in bottom sections).
|
|
|
DISCUSSION
|
|---|
We have developed a correction method that is used to resolve the issues of distance and angle dependence in catheter-based intravital optical imaging and that results in continuously acquired truly quantitative fluorescence measurements. We hypothesized that division of each NIR image by a simultaneously acquired WL image would allow for a dynamic frame-by-frame normalization that would account for variations in signal intensity between and within individual frames due to catheter position changes. To implement this normalization algorithm, we developed a software platform that was used to perform real-time necessary registration, segmentation, exposure time adjustment, and gradation (window/leveling) to acquire and process images from WL and NIR cameras.
In our study, we demonstrated both the dramatic variability of raw NIR signal caused by changes in catheter position and the effectiveness of our correction technique in generating a corrected NIR signal that accounts for these dependencies. The virtual replication of the in vitro data by the in vivo data suggests the robustness of this normalization algorithm. Furthermore, the results of the in vitro experiments show that the correction method can be used to discriminate between samples of different fluorochrome concentration. Thus, it is impossible to assign any counts-per-millisecond range to a specific fluorochrome concentration. On the other hand, signal intensity values from the corrected NIR image can be directly correlated with fluorochrome concentration. With this method, it is possible to assign a true fluorescence value to a target under investigation during minimally invasive procedures; when used in conjunction with enzyme-activatable probes, this technology may allow the in vivo assessment of gene expression profiles within abnormal lesions, an advancement that could be used, for example, in the intravital staging of cancers on the basis of protease expression (13).
We have evaluated the correction schemes described in this article over a wide range of distances, especially when these distances are compared with the diameter of the imaging catheters. Illumination sources on imaging catheters are often circumferentially located relative to the lens in front of the light-receiving fibers. However, they can be placed along one axis of the catheter face or asymmetrically placed in one point adjacent to the lens. While we have initially evaluated such catheters and found comparable results, a limitation of our study is that we have not exhaustively quantitatively tested their behavior as we did for the catheter system used throughout our study. Future studies may be performed to evaluate such variable designs for changes in ability to correct for distance and angle, especially for very close (less than 1 mm) and very far (more than 10 mm) conditions.
We believe that our technique will help pave the way for the evolution of catheter-based systems into fully quantitative and more clinically useful optical fluorescent imaging devices. In the future, interventional radiologists may be able to use fluorescence guidance in a fashion similar to that of fluoroscopic, CT, or ultrasonographic guidance. The high target-to-background ratios that are possible with targeted and especially activatable fluorescence agents (14) will allow easier detection and sampling of suspicious lesions and will allow real-time guidance during catheter advancement (eg, in the peritoneum). Vascular interventions after saline flushing will enable identification of atherosclerotic regions more likely to rupture and hence more likely to benefit from intervention compared with other segments with a similar disease burden (15). In situ characterization of lesions may be possible by means of quantitative reporting of the amount of fluorescent ligand binding. Such enhanced sampling may help to ensure imaging of the most aggressive regions within lesions. Moreover, as the techniques described herein are expandable to more than one NIR fluorescent wavelength, such characterization will become more precise and may help to target at-risk predisease tissue.
|
ADVANCES IN KNOWLEDGE
|
|---|
- It is possible to quantitate optical signal while continually varying parameters, such as catheter distance and angle.
- There is a one-to-one relation between corrected near infrared fluorescence signal and fluorochrome concentration.
|
IMPLICATION FOR PATIENT CARE
|
|---|
- We present the in vivo translation of this development by showing the ability to correct for these factors in real time during murine colonoscopy and laparoscopy. Human translation is possible, and the developed technique may be applied to interventional radiology–based procedures.
|
ACKNOWLEDGMENTS
|
|---|
The authors thank Herlen Alencar, MD, and Marco Maricevich, MD, for help with surgical procedures and histologic analysis and Elkan Halpern, PhD, for help with statistical analysis.
|
FOOTNOTES
|
|---|
Abbreviations: CCD = charge-coupled device • NIR = near infrared • WL = white light
See also Science to Practice in this issue.
Author contributions: Guarantor of integrity of entire study, U.M.; study concepts/study design or data acquisition or data analysis/interpretation, all authors; manuscript drafting or manuscript revision for important intellectual content, all authors; manuscript final version approval, all authors; literature research, R.U., R.A.S., U.M.; experimental studies, R.U., R.A.S., U.M.; statistical analysis, R.U., R.A.S., U.M.; and manuscript editing, all authors
Authors stated no financial relationship to disclose.
|
References
|
|---|
- Weissleder R, Mahmood U. Molecular imaging. Radiology 2001;219:316–333. [Abstract/Free Full Text]
- Weissleder R, Tung CH, Mahmood U, Bogdanov A Jr. In vivo imaging of tumors with protease-activated near-infrared fluorescent probes. Nat Biotechnol 1999;17:375–378. [CrossRef][Medline]
- Tung CH, Bredow S, Mahmood U, Weissleder R. Preparation of a cathepsin D sensitive near-infrared fluorescence probe for imaging. Bioconjug Chem 1999;10:892–896. [CrossRef][Medline]
- Mahmood U, Tung CH, Tang Y, Weissleder R. Feasibility of in vivo multichannel optical imaging of gene expression: experimental study in mice. Radiology 2002;224:446–451. [Abstract/Free Full Text]
- Mahmood U, Tung CH, Bogdanov A Jr, Weissleder R. Near-infrared optical imaging of protease activity for tumor detection. Radiology 1999;213:866–870. [Abstract/Free Full Text]
- Funovics MA, Weissleder R, Mahmood U. Catheter-based in vivo imaging of enzyme activity and gene expression: feasibility study in mice. Radiology 2004;231:659–666. [Abstract/Free Full Text]
- Funovics MA, Alencar H, Su HS, Khazaie K, Weissleder R, Mahmood U. Miniaturized multichannel near infrared endoscope for mouse imaging. Mol Imaging 2003;2:350–357. [CrossRef][Medline]
- Figueiredo JL, Alencar H, Weissleder R, Mahmood U. Near infrared thoracoscopy of tumoral protease activity for improved detection of peripheral lung cancer. Int J Cancer 2006;118:2672–2677. [CrossRef][Medline]
- Zwier JM, Van Rooij GJ, Hofstraat JW, Brakenhoff GJ. Image calibration in fluorescence microscopy. J Microsc 2004;216:15–24. [Medline]
- Model MA, Burkhardt JK. A standard for calibration and shading correction of a fluorescence microscope. Cytometry 2001;44:309–316. [CrossRef][Medline]
- Mahmood U, Weissleder R. Near-infrared optical imaging of proteases in cancer. Mol Cancer Ther 2003;2:489–496. [Abstract/Free Full Text]
- Alencar H, King R, Funovics M, Stout C, Weissleder R, Mahmood U. A novel mouse model for segmental orthotopic colon cancer. Int J Cancer 2005;117:335–339. [CrossRef][Medline]
- Duffy MJ. Proteases as prognostic markers in cancer. Clin Cancer Res 1996;2:613–618. [Abstract]
- Jiang T, Olson ES, Nguyen QT, Roy M, Jennings PA, Tsien RY. Tumor imaging by means of proteolytic activation of cell-penetrating peptides. Proc Natl Acad Sci U S A 2004;101:17867–17872. [Abstract/Free Full Text]
- Chen J, Tung CH, Mahmood U, et al. In vivo imaging of proteolytic activity in atherosclerosis. Circulation 2002;105:2766–2771. [Abstract/Free Full Text]
Related Article
-
Is It Possible to Quantify Fluorescence during Optical Endoscopy?
- Peter L. Choyke
Radiology 2007 245: 307-308.
[Full Text]
[PDF]
This article has been cited by other articles:
|
|
|
|
 
J. Waldeck, F. Hager, C. Holtke, C. Lanckohr, A. von Wallbrunn, G. Torsello, W. Heindel, G. Theilmeier, M. Schafers, and C. Bremer
Fluorescence Reflectance Imaging of Macrophage-Rich Atherosclerotic Plaques Using an {alpha}v{beta}3 Integrin-Targeted Fluorochrome
J. Nucl. Med.,
November 1, 2008;
49(11):
1845 - 1851.
[Abstract]
[Full Text]
[PDF]
|
|
|
|
|
|
|
P. L. Choyke
Is It Possible to Quantify Fluorescence during Optical Endoscopy?
Radiology,
November 1, 2007;
245(2):
307 - 308.
[Full Text]
[PDF]
|
|
|
TOP
ABSTRACT
INTRODUCTION
