1.

Introduction

Cancer detection and treatment in hollow organs such as the GI tract, airways, urinary tract, and uterus depends heavily on endoscopic techniques that conventionally utilize white light to illuminate, visually inspect the space of interest, and guide biopsy and excision of suspicious lesions. A major theme driving current endoscopic research is that malignant and premalignant lesions that are flat or small may appear macroscopically similar to inflamed, or even normal tissues. This limitation may hinder early detection of aggressive cancers, often leaving collection of random and repeat biopsies as the only viable strategy available for surveillance in high risk patients. In an effort to address these problems, several optical methods have been investigated as a tool to complement white light endoscopy for more accurate and complete treatment. These include microscopic imaging techniques such as optical coherence tomography (OCT) endo-cytoscopy, and confocal fluorescence microscopy, light scattering techniques such as Raman spectroscopy, and fluorescence imaging techniques which involve the excitation and detection of either native tissue fluorophores or the use of exogenous fluorescent materials and precursors.^1,2

From a clinician’s point of view, the ideal optical adjunct modality for endoscopic procedures would increase the sensitivity and specificity of detecting malignant and premalignant lesions, reliably identify the lateral and deep margins required for complete resection, detect residual tumor within a previous resection bed, adapt easily to work with existing endoscopic technology, not require the use of toxic or inconvenient chemicals or pharmaceutics, and allow simultaneous display of a conventional white light image such that endoscopic surgery could be guided by such modality in real time. Fluorescence based imaging techniques, in particular, have shown promise in being able to deliver on many of these endoscopic requirements and are already available in a number of clinical applications including visible autofluorescence imaging in gastrointestinal (GI) endoscopy³ and bronchoscopy.^4,5

Autofluorescence (AF) signal in the visible part of the spectrum tends to be higher in normal tissue than in cancer. Therefore, lesion identification relies on detection of a decrease in fluorescence compared to normal tissue.² The most important endogenous fluorophores in the visible wavelengths are the reduced form of nicotinamide adenine dinucleotide, NADH, and flavins which are involved in cellular metabolism and the structural proteins collagen and elastin.⁶ In the near-infrared (NIR) part of the spectrum, the most important endogenous fluorophores are thought to be porphyrins⁷ which in most cases are more highly concentrated in malignant than benign tissue⁸ and are able to be excited by visible light with relatively long wavelengths compared to molecules that emit visible fluorescence. Our group has previously demonstrated that several types of malignant tumors can be differentiated in vitro from contiguous normal tissue by inducing excitation under long-wavelength laser illumination and imaging the resulting autofluorescence in the near infrared.^9,10 In particular, an ex vivo study of normal and tumor bladder tissue has indicated that the tumor exhibits a lower NIR AF intensity than normal tissue under identical excitation conditions, while areas of necrotic tissue exhibit a much stronger signal.⁹ The potential advantages of using longer excitation wavelengths are greater tissue penetration depth of the probe light and reduced artifacts arising from non-uniform concentration of blood which is a very strong absorber of the visible light. In addition, by appropriate separation of the visible white light and NIR autofluorescence image components, simultaneous collection of both images is possible using conventional endoscope designs with each image acquired at a different frame rate as needed without imposing any limitation on the acquisition of the complementary image component.

In this paper, we describe a prototype instrument that we developed to simultaneously conduct ordinary visual endoscopy together with NIR autofluorescence imaging via parallel image acquisition. The two images are recorded concurrently, without the need to switch back and forth between imaging modes and the instrument interfaces with an ordinary endoscope. For our initial investigation using this instrument, we conducted an in vivo pilot study of bladder tumors to build on our previous work with NIR autofluorescence of bladder cancer using ex vivo specimens and define the current limitation for further development of the system.

2.

Materials and Methods

Figure 1 shows a schematic representation of the prototype NIR autofluorescence endoscopy instrument. The system utilizes a standard cystoscope lens, and the accompanying illumination fiber, but the illumination source and imaging systems are modified. Specifically, the input side of cystoscope’s light guide, which transports the external white light to the light pillar of the cystoscope, is attached to a specially made illumination assembly designed to provide monochromatic laser illumination at 650 nm, and visible white light with all spectral components above 650 nm removed. The white light is obtained from a xenon endoscopy light source and is transported to the illumination assembly with a 4 mm diameter liquid light guide. The output of the light guide is collected and refocused in to the input of the cystoscope’s light guide using a system of lenses after passing through three 650 nm short-pass interference filters to allow only transmission of photons in the visible light spectrum. This light was combined with the output of a CW laser operating at 650 nm,PLT Technology, Model 650-4-F, which was transported into the illumination assembly with a fiber after passing through a 650 nm narrow-band filter to ensure monochromatic illumination. This provides a combined visible white light and 650 nm laser light in to the tissue target location. The output power at the tip of the cystoscope is about 220 mW of visible light and about 150 mW of 650 nm laser light.

Fig. 1

Schematic diagram depicting the main optical components of the prototype imaging system providing simultaneous acquisition of visible light scattering and NIR autofluorescence images.

The resultant images at the output of the cystoscope lens are coupled in to an image-preserving fiber optic bundle using a standard $f=25\text{-}\mathrm{mm}$ C-mount endoscopy video coupler. This image preserving fiber bundle has a $4\mathrm{mm}\times 4\mathrm{mm}$ square field of view and a length of 268 cm. The overall transmission of the image preserving fiber bundle in the visible and NIR spectral region is on the order of 40 percent, while the individual fibers are 10 μm in diameter with a core diameter of 8 μm. The image preserving fiber provides a reliable way to avoid achromatic aberrations of the two transmitted image components, the visible light scattering image and the NIR AF image. In this arrangement, the two image components are transported into the imaging assembly where they are split using a system of lenses and a cold mirror which reflect the visible image component and transmit the NIR image component. The NIR image is passed through a 690 nm long-pass filter before being projected onto a high sensitivity $512\times 512$ pixel, air cooled Electron Multiplying charge couple device (EMCCD) camera. The visible image reflected by the cold mirror is passed through a 650 nm notch filter to remove any scattered light from the laser. It is then collected by a digital red-green-blue (RGB) CCD. The signals from both CCDs are separately processed by a personal computer and displayed simultaneously on two separate monitors. AVT Smartview software was used to process and display of RGB images,15 frames per second, and an image was saved every 15th frame,1 image per second. $\mathrm{Winview}/32$ was used to process and display the NIR AF images with exposure time of 500 ms and saved on the computer. The instrumentation discussed above was assembled in to a portable system that was easily transportable into the operating room. This 4-level portable system contained the light sources and power supplies at the lower level, the illumination and imaging assemblies in the middle sections, while the computer and monitors were positioned at the top for easy access and visibility.

We conducted preliminary testing of the system’s ability to acquire NIR AF images while simultaneously displaying the conventional color, RGB, image in 21 patients undergoing transurethral resection of bladder tumors at UC Davis Medical Center. These experiments were approved by the UC Davis institutional review board. After the tumor and the region of interest were initially defined via standard video cystoscopy, the standard coupler and video camera enabling the conventional RGB imaging were detached from the cystoscope and were replaced by the coupler and image preserving fiber bundle of the prototype system. The input of the cystoscope’s light guide was detached from the conventional light source and coupled to the illumination assembly of the prototype system. The system was then used to capture RGB and NIR AF images of the region of interest while these images were displayed in separate adjacent monitors which guided the operator. The entire process of connecting the prototype system to the cystoscope, acquiring the in vivo data, and reverting to the regular RGB camera was accomplished within a time window of about 2 min. These measurements were performed both before and after tumor resection. Data was also recorded for each patient’s tumor size, grade, and depth of invasion based on subsequent pathology report.

For each patient, the RGB images were used to manually map the location of apparently normal tissue and tumor tissue on the corresponding NIR images. Upon selection of the area of interest in the RGB image, the corresponding NIR image was identified by a time tag. Because of the difficulty in normalizing excitation intensity with the current setup, we did not attempt to quantify the signal in the current set of experiments.

3.

Results

The experiments performed were aiming to satisfy four main technical objectives:

Demonstration of the type of images obtained using this system is shown in Fig. 2. This image was obtained, in vivo, from a normal bladder tissue area exhibiting normal vascularity. The image on the left side is the white light RGB image and the image on the right side is the NIR AF image that was simultaneously acquired. The latter image is smooth, showing no features that could correlate to the location of blood concentration shown in the corresponding RGB image. This also provides an example of our findings that moderate amounts of blood do not give rise to corresponding image artifacts in the NIR AF images.

Fig. 2

RGB and corresponding NIR images of a normal tissue area with well defined vasculature.

Figure 3 shows the RGB and corresponding NIR AF image from the location of a tumor. The polypoid tumor is clearly visible in the RGB image, but it only appears as a dark object in the NIR AF image. This is in agreement with the previous ex vivo study where the tumors were detected with lower intensity compared to normal bladder tissue.⁹ Pathology classified this tumor as High Grade Ta, indicating the superficial and non-invasiveness of the lesion, but with cellular features that are consistent with a relatively aggressive cancer.

Fig. 3

RGB and NIR images of a 1 cm papillary bladder tumor; ${\mathrm{T}}_a$ high grade urothelial carcinoma.

Images of more advanced tumors are shown in Fig. 4. The RGB and NIR image pair shown on the top represent a sessile, high grade, muscle invasive, T2 tumor. A characteristic observation, in this case, is that the NIR autofluorescence intensity arising from different part of the tumor significantly varies. Specifically, the upper part of the tumor provides a stronger autofluorescence signal than the lower part of the tumor. This part also appears as a less colored, whiter, feature in the RGB image. A similar effect is observed in the bottom image pair which shows a papillary, high grade T1 tumor at a closer range. A well-defined structure clearly identifiable in the RGB image provides a much stronger signal in the NIF autofluorescence image.

Fig. 4

RGB and NIR images of tumors exhibiting a different amount of NIR autofluorescence at different locations. Top images: sessile high grade, muscle invasive bladder tumor. Bottom images: papillary high grade, ${\mathrm{T}}_1$ (invasive into lamina propria but not into underlying muscle) bladder tumor.

This was a relatively common observation during our preliminary in vivo experiment, where certain parts of more advanced tumors exhibit a stronger signal than the rest of the tumor or that of the normal tissue. We hypothesize that this may arise from the presence of necrotic tissue in more advanced tumors and the presence of calcifications, as we have observed in our ex vivo studies. However, in many cases, the signal was moderately higher than normal tissue or the rest of the tumor such as in the case shown in top image pair in Fig. 4. It is unclear whether this observation is clinically valuable. However, we do not exclude the possibility that some more advanced tumors have increased concentration of porphyrins, or other type of NIR fluorescence producing biomolecule, which give rise to NIR autofluorescence as is discussed in more detail in the next section.

Figure 5 shows the RGB and corresponding NIR AF images after a previous resection months earlier of a high grade T1 tumor. The perimeter of the lesion appeared as a very bright object, similar to tumor with necrotic tissue. The pathology from this lesion was reactive urothelium and necrosis. Visually, it was the necrotic component, the lighter colored areas, that gave rise to the visually different increased strength of the NIR autofluorescence signal. The rest of the lesion of reactive urothelium was not differentiated from normal tissue in the NIR AF image.

Fig. 5

RGB and NIR images of reactive urothelium with focal areas of necrosis.

Increased florescence was also observed after cauterization of tissue, which occurs typically during endoscopic surgery for removal of bladder tumors. A high electrical current is used to cut through the tissue, and then at a lower current to cauterize areas of bleeding and to ablate cancer cells. This effect is exemplified in Fig. 6. The cauterized area is visible in the RGB image from the brownish appearance of the tissue. The corresponding NIR AF image exhibits increased emission compared to the normal tissue. This suggests that this imaging method may be useful only before the last step of the treatment via cauterization and possibly other types of thermal ablation such as radiofrequency ablation.

Fig. 6

RGB and NIR images of a cauterized area of bladder urothelium illustrating artifact.

4.

Discussion

The observed NIR AF signal arises, predominantly, from the 650 nm laser excitation and is observed at a much lower level when the laser is turned off and the excitation arises from the white light only. The system, in its current form, is not suitable for quantification of the recorded signal because there is no normalization to the excitation intensity. This is particularly important as different parts of the image are typically at different distances from the tip of the cystoscope and, therefore, they receive different amounts of photo-excitation. In addition, tumors often have a papillary form, but may also appear as sessile growths which can lead to increased excitation compared to surrounding normal tissue during in vivo measurements. One simple way to approximate normalization of the NIR autofluorescence image would be to record the corresponding light scattering image of the laser excitation, which would be a relatively simple addition to the current system.

The most important technical objective of this preliminary study, as outlined at the beginning of the previous section, was to explore whether this instrumentation design is capable of providing simultaneous and independent acquisition of the conventional RGB with the NIR AF images. The benefit of this instrumentation design is that the visual examination, which still remains the most important diagnostic method during surgery, is complemented by the spectral image(s) in the most efficient manner that allows for easy co-registration and correlation of image features and minimized acquisition time. It is also well recognized, that the NIR autofluorescence intensity is very weak compared to that of visible autofluorescence arising from excitation in the UV or near UV spectral range. This necessitates the use of longer exposure times for image acquisition. Our prototype design allows these long exposure times for the acquisition of the NIR autofluorescence images, 2 frames per second, without interfering with the acquisition of the standard video cystoscopy RGB images which are acquired at a standard video rate. The experimental results demonstrate that this technical objective was achieved, as illustrated by Figs. 2Fig. 3Fig. 4Fig. 5–6. The images were displayed on separate monitors and the operator could easily correlate the spatial location of the features of interest in real time. Although the image quality was somewhat lower due to its transmission through the image preserving fiber bundle, this is a minor technical issue that can be addressed in future applications using a fiber bundle with larger number of fibers with a smaller core diameter. Complemented with use of imaging software, this minor issue can be effectively eliminated, at least at the level of the visual perception of the operator.

A significant reduction of the power of the white light illumination transmitted through the cystoscope was the result of the passing this light through the illumination assembly for spectral filtering. However, this did not limit acquisition of RGB images at 15 frames per second using a relatively inexpensive CCD camera. One frame per second was saved in the PC, which was used to also operate the CCD camera acquiring the NIR image. This limitation was imposed by the computer’s processing speed, which was operating at it’s limit under the image acquisition and storage parameters used. The NIR autofluorescence images were acquired and saved at rate of 2 frames per second, which was a compromise to achieve a reasonable image quality that would allow features of interest to be clearly observable. As cystoscopes have a very narrow lens at the tip, it is challenging to implement an imaging method that is based on a weak signal, such as the NIR AF component. Using a higher power laser would proportionally increase the detected NIR signal.

One of the most common problems of in vivo fluorescence imaging is the appearance of artifacts arising from heme in blood, which absorbs very strongly in the visible part of the spectrum. We hypothesized that excitation with a far red light, 650 nm, would mitigate this problem. Our results, exemplified in Fig. 2, illustrate that the NIR fluorescence images acquired with this setup are relatively unaffected by the presence of blood.

While the technology described here is applicable to any type of endoscopy, the case of cystoscopic examination of bladder cancer is particularly illustrative of the need for development of additional visualization tools. 75 percent of patients with bladder cancer present with superficial disease, non-muscle-invasive, and confined to the mucosa—${\mathrm{T}}_a$ or lamina propria—${\mathrm{T}}_1$. While less than 20 percent of superficial bladder cancers progress to muscle-invasive disease, ${\mathrm{T}}_2$, about 50 percent of ${\mathrm{T}}_a$ tumors and 70 percent of ${\mathrm{T}}_1$ tumors will recur.¹¹ It has been shown that the recurrence rate of tumors with similar characteristics can vary substantially when compared by treatment institution,¹² suggesting that completeness of initial resection may play a significant role in determining the clinical outcome. Furthermore, it has become routine standard of care to bring patients with ${\mathrm{T}}_1$ bladder tumors back to the operating room for follow up resection because approximately 30 percent will show muscle-invasive disease from the second surgery,¹³ also reflecting a common tendency for incomplete initial resection of ${\mathrm{T}}_1$ tumors. In addition, carcinoma in situ, CIS, a high grade, flat, superficial bladder lesion that carries a 20 percent probability of spreading into the ureters and/or kidneys can easily be missed by white light cystoscopy because it often looks the same as bladder inflammation.

Among the novel types of imaging that have been developed to address these problems to date, the most successful has been fluorescent imaging involving either intravenous or topical application of 5-aminolevulinic acid, 5-ALA, or its ester derivative, hexyl ester hexaminolevulinate (HAL) which has increased uptake into tumor tissue than aminolevulinic acid (ALA).¹⁴ Synthesis of ALA occurs naturally and is the rate limiting step in the production of heme. Exogenous ALA or HAL provided in large excess cause bioaccumulation of protoporhyrin IX (PpIX) the last molecular precursor to heme prior to enzymatic addition of iron. PpIX is a potent fluorophore when excited by blue light, 380 to 480 nm, and causes tumors to fluoresce more than normal tissue with red emission in the 625 to 725 nm region. This method has been studied in several large series including prospective, multicenter, randomized trials in Europe.^15–20 The largest of these studies showed that fluorescence diagnosis (FD) resulted in a nearly 20 percent increase in detection of residual tumors following initial resection using white light cystoscopy (WLC) alone, and statistically significant increases in recurrence free survival up to eight years after the initial transurethral resection of bladder tumor. Despite the successful outcomes associated with this technique, there was a reduced specificity for cancer detection with a 37 percent false positive detection for HAL versus 26 percent for WLC.¹⁶ Inflammation is the major source of false positive fluorescence signals.²¹ However, there are a number of significant risks and limitations currently identified when ALA is used systemically in concentrations that allow detectable tumor fluorescence with current instrumentation.*† Protoporphyrin IX is induced by the instillation of the bladder with the contrast agent that is held for one hour before the fluorescence light cystoscopy is performed. Although using ester derivatives of 5-ALA as the contrast agent has shown increased tissue penetration depth, this remains on the order of 1 mm and, therefore, the induced fluorescence does not allow tumor visualization after resection for evaluation of the deep margins. Therefore, a second transurethral resection is indicated in T1 tumors to rule out muscle invasion.²²

Compared to using ALA or HAL, the method we describe in this paper to detect bladder tumors has several potential advantages. First, the use of long wavelength, red, excitation light minimizes the absorption of heme which can lead to image artifacts while still allowing detection of a near infrared fluorescence signal without necessarily using an exogenous fluorophore. Interestingly, the NIR is considered to be part of the spectrum in which there is relatively little native tissue absorption and autofluorescence, which has helped generate a large interest in developing contrast agents that fluoresce in the NIR for biological imaging.²³ The ability for the system to qualitatively generate a NIR fluorescent signal that corresponds with presence of visible tumor reflects the high sensitivity that this combination of excitation and fluorescence wavelengths allow, without the loss of signal that would typically be associated by absorption of excitation light by heme.

The detection and imaging of cancer, using administration of ALA or HAL discussed above, can also be implemented using the current instrumentation maybe with only a minor modification on the laser wavelength to operate at 630 nm. This is because, although PpIX exhibits its maximum absorption in the blue and near UV spectral region, it also exhibits absorption extending to about 650 nm. With the ultrasensitive detection system used in this work, the emission from PpIX would be detected using concentrations that are multiple orders of magnitude lower to those used previously. This in turn means that much smaller concentration of ALA or HAL can be used that will help alleviate some of the issues of that method.

The instrumentation design presented in this work could be interfaced with NIR contrast agents that bind selectively to transitional cell carcinoma (TCC) such as via an antibody–antigen interaction. This technological combination would theoretically increase the sensitivity and specificity of the method. Such NIR contrast agents are in the process of being developed.²³

The convenience of not needing to switch imaging modes during surgery increases the potential utility of including fluorescence imaging as a surgical tool in a way that has not yet been realized and increases the flexibility of the instrument to use in a wide variety of endoscopic applications. In addition, the use of relatively long wavelength excitation light increases the probe depth compared to what is possible using UV or blue light for fluorophore excitation. The anticipated advantage is the ability to visualize tumor that is more extensive than superficial lesions that have so far been studied using fluorescence methods in endoscopy.

The finding of variations in NIR AF intensity in neoplastic bladder tissue with some tumors exhibiting higher AF than normal tissue is consistent with our previous experience with more advanced bladder tumor specimens studied in vitro as illustrated in Figs. 1 and 2 in Ref. 9. In these ex vivo studies, the images of more advanced tumors were from cross sectioned specimens following bladder resection. The higher NIR AF intensity regions were typically observed in the superficial layer of the tumors. The in vivo images obtained in this study represent a different viewing perspective, top view of the bladder surface as opposed to a cross section of the bladder tissue. We presumed that the areas that exhibited strong fluorescence were necrotic tissue, and it seems likely that devitalized tissue undergoes a chemical change that increases its NIR AF properties. In addition, as discussed above, there is an increased porphyrin production in the bladder tumor in response to an external stimulus such as ALA administration. It has been shown before, in various reports, that tumors originating in various parts of the body, but not in bladder, exhibit increased porphyrin content which would probably increase the NIR fluorescence signal. This effect has been shown in one of our previous reports using ex vivo specimens.¹⁰ It is possible that bladder tumors at some stage of their development may start exhibiting this behavior and, therefore, the observation of higher intensity autofluorescence emission in bladder tumors may be related to their stage of development. Additional experiments will be needed to elucidate the cause for the variation in behavior we see for different tumors. We also do not yet have a set of experiments that included patients with possible carcinoma in situ, so it is unclear how these lesions would exhibit NIR AF compared to normal tissue.

The most important limitation of the device we describe here, given its current configuration, is the fluorescence artifact seen in cauterized and necrotic tissue. The images, seen in Fig. 3, confirm the presence of a higher fluorescence signal from even mild cautery compared with the normal tissue. The reasons for these high artifactual signals are not yet clear since to our knowledge there have never been similar images presented in the literature. While this phenomenon should be studied further in a clinical environment, ultimately the best application for the NIR AF imaging approach may be used in a non-surgical setting such as an ambulatory clinic environment, prior to use of cautery.

5.

Conclusion

A novel instrument which utilizes long wavelength visible excitation and near-infrared fluorescence in combination with conventional white light endoscopy has been developed. The images we present here suggest high qualitative sensitivity given that tissue autofluorescence in the NIR part of the spectrum occurs typically at low amplitude. Cancer tumors were qualitatively visible with NIR autofluorescence in vivo. Necrotic, calcified, and cauterized tissue exhibited artifactual increased NIR autofluorescence.