Urothelial carcinoma in situ (CIS) is a precursor of invasive bladder cancer, which if left untreated, will likely progress to more aggressive disease. Approximately 50% of CIS lesions are missed on routine cystoscopy owing to their flat architecture. Furthermore, many benign but abnormal-appearing areas may be biopsied owing to lack of cellular resolution of cystoscopes. Multiphoton microscopy (MPM) is an optical imaging technique that generates subcellular-resolution three-dimensional images from unfixed tissue without using exogenous dyes.
To assess the diagnostic potential of MPM in identifying and differentiating benign from malignant flat bladder lesions, especially CIS.
Seventy-eight specimens (benign = 46, CIS = 23, invasive = 9, as diagnosed on histopathology) were obtained from flat bladder mucosa via transurethral resection of bladder, cold cup biopsy, or cystectomy, imaged fresh with a commercial benchtop MPM, and submitted for routine histopathology. Multiphoton microscopy and hematoxylin-eosin diagnoses were compared.
In 77 of 78 specimens (99%), accurate MPM diagnoses (benign/malignant) were given on the basis of their architectural and cytologic features (nuclear to cytoplasmic ratio, pleomorphism, polarity/organization of urothelial layers, etc). The sensitivity and specificity were 97% and 100%, respectively, with positive (malignant) and negative (benign) predictive values of 100% and 98%, respectively. The interobserver agreement, κ, was 0.93.
Our study demonstrates the capability of MPM to identify and differentiate benign from malignant flat bladder lesions, especially CIS. With the advent of MPM endoscopes, we foresee their potential as a biopsy guidance tool for early detection and treatment of CIS, thus reducing the rate of biopsies with benign diagnoses and their associated complications.
The American Cancer Society estimated that 72 570 new cases of bladder cancer were diagnosed in 2013 in the United States alone, accounting for 15 210 deaths.1 Almost 70% of these cases represent early-stage superficial disease, that is, non–muscle invasive (stage pTa, pT1, or pTis).2,3 Among the superficial cancers, carcinoma in situ (CIS) is often the most difficult to detect and if left untreated progresses to muscle-invasive disease in 80% to 100% of the cases, with associated mortality.4,5
Also see p. 719.
Urine cytology followed by white light cystoscopy (WLC) constitutes the current standard for the detection and diagnosis of CIS in the United States. Although urine cytology findings are positive in more than 90% of patients with CIS, it cannot determine the extent and location of the disease.6 White light cystoscopy accompanied by histopathologic diagnosis of transurethral resection of bladder (TURB) specimens from the suspicious lesions thus typically follows positive urine cytology findings. Unfortunately, it has been estimated that approximately 50% of CIS lesions are missed on WLC alone.6 This is mainly due to flat architecture of these lesions, accompanied by lack of cellular resolution of WLC to distinguish CIS from benign flat lesions. Thus, patients with positive urine cytology findings but no obvious tumor during cystoscopic examination may be subjected to repeated cystoscopies and numerous biopsies (including many with benign diagnoses) to rule out carcinoma. Furthermore, biopsies are not without potential complications, such as bleeding, infection, and occasionally, bladder perforation, that negatively impact the patient's quality of life. To add to this scenario, routine histology processing and examination cannot be done during the same office visit, resulting in patient anxiety and delay in treatment.
New technologies, such as narrow band imaging (NBI),7–9 photodynamic diagnosis (PDD),10–15 optical coherence tomography (OCT),16–19 confocal endomicroscopy (CLE),20–22 and endocytoscopy23 have been used for better in vivo visualization and characterization of bladder cancers, especially CIS. However, each of these techniques has its own limitations. Multiphoton microscopy (MPM) is a nonlinear imaging approach that generates high-resolution images from fresh (unprocessed) tissue by exciting intrinsic tissue emissions.24,25 These images reveal morphologic features similar to those seen in histologic images.24,25 The benchtop MPM system has been used successfully to identify benign and malignant lesion in ex vivo tissues from various organs, including the urinary bladder, in human subjects.26–28 While miniaturized MPM endoscopes (MPEs)29,30 are being developed and tested in animal models, we aim to determine the potential of MPM to detect malignant flat lesions, especially CIS, using ex vivo human bladder samples. The ultimate goal of this project is to integrate MPM during cystoscopy for the detection of malignant cells in superficial urothelial carcinomas. Efficient detection and biopsy/resection of such lesions at first TURB will not only prevent recurrence and progression of CIS, but may also reduce the rate of unnecessary biopsies with benign diagnoses and their associated complications.
MATERIALS AND METHODS
Study Cohort
Sixty-two adult men and women who underwent TURB (n = 35), cold cup biopsy (n = 11), or cystectomy (n = 16) for symptoms or treatment of urinary bladder cancer were consented to participate under an institutional review board–approved study. For this study, our goal was to collect specimens from every patient who underwent the abovementioned procedures and presented with suspicious lesions on the flat mucosa. However, not all consecutive samples are included, since some subjects did not consent to participate, and in some other cases, the microscope was unavailable during the time of the procedure.
Acquisition of the Specimens
Specimens from both normal- and abnormal-appearing flat (nonpapillary) bladder mucosa that were taken for diagnostic purpose were included in the study. A total of 78 specimens were obtained for MPM (TURB [n = 36], cold cup procedure [n = 18]), and cystectomy [n = 24]). These specimens were collected in 0.9% phosphate-buffered saline and immediately transported to the MPM suite for imaging. Imaging was completed within an hour of tissue arrival in the suite. Immediately after imaging, samples were transferred to 10% formalin and returned to surgical pathology for routine histopathologic tissue processing.
MPM Imaging
For our study, we used the Olympus FluoView FV1000MPE upright microscope (Olympus Imaging America Inc, Center Valley, Pennsylvania). Bladder specimens were placed in a Petri dish with mucosal surface facing upward and overlaid with a coverslip to flatten the imaged surface. Images were acquired first at low magnification by using ×4 /0.28 numerical aperture (NA) dry objective (total magnification of ×24) to get an overview of the tissue architecture and identify areas of interest. These areas were then explored at high magnification by using ×25 /0.9 5NA water immersion objective (total magnification of ×150) to confirm the diagnosis at cellular level (benign versus malignant). To achieve water immersion for high-magnification imaging, a drop of saline was placed on the coverslip. Throughout the imaging session, the specimens were kept hydrated with saline to prevent drying of the tissue. Specimens were excited by using 780-nm light from a tunable femtosecond-pulsed Ti-Sapphire laser (Mai Tai DeepSee, Spectra-Physics, Newport Corporation, Santa Clara, California). Laser power used varied between 10 and 30 mW under the ×25 objective, depending on the intrinsic brightness of the specimen. Two distinct intrinsic tissue emission signals were collected by using photomultiplier tubes in nondescanned configuration: (1) second-harmonic generation (360–400 nm, color-coded red), a nonlinear scattering signal originating from tissue collagen (lamina propria and basement membrane), and (2) autofluorescence (420–490 nm, color-coded green), originating in part from reduced nicotinamide adenine dinucleotide (NADH) and flavin adenine dinucleotide in cells (benign and malignant cells as well as lymphocytes), elastin, and smooth muscle fibers. Nuclei did not fluoresce but appeared as signal-void regions in a background of fluorescent cytoplasm, thus allowing the determination of their shape and size for making the diagnosis. The signals in individual channels were collected as separate gray-scale images that were then color coded and merged to produce final Z-stacks. Typically, the collection of a single frame as shown in the figures took 1 to 3 seconds, and acquisition of an entire image stack of 20 to 40 images took 1 to 2 minutes. Imaging at low magnification provided a field of view (FOV) of approximately 3.1 mm2 with an imaging depth of 500 μm from the surface of the tissue. At high magnification, the FOV was 0.5 mm2 and we could image as deep as approximately 250 μm. Minor image post processing, such as adjustments to brightness and color balance, were carried out by using Adobe Photoshop CS4 (Adobe Systems Incorporated, San Jose, California) for better visualization. However, images were readily interpretable during acquisition by the study pathologist who was present during the image acquisitions. We thus note that real-time diagnosis with MPM is possible without further manipulation of the images.
Blinded Analysis
Representative MPM stacks from each bladder specimen were compiled by the research pathologist, and this set was shown independently to both attending uropathologists in our study. Reviewing uropathologists were blinded to the clinical diagnosis (eg, normal versus tumor) and to the final histopathologic diagnosis based on hematoxylin-eosin (H&E) staining. They were then asked to categorize lesions as benign or malignant (including CIS and invasive high-grade urothelial carcinoma) by their cellular and architectural details and in conjunction with World Health Organization/International Society of Organization/International Society of Urological Pathology (2004) classification.31 The same uropathologists made histopathologic diagnoses from the corresponding H&E slides, again blinded to the clinical as well as MPM diagnoses. Multiphoton microscopy diagnosis was then compared with H&E diagnosis and a standard 2 × 2 contingency table was used to determine the diagnostic test operating characteristics (accuracy, sensitivity, specificity, positive predictive value, and negative predictive value). Cohen κ coefficient was also calculated to obtain a statistical measurement of interobserver agreement.
RESULTS
All MPM images were collected as a set of optical sections, starting from the tissue surface, and going several hundred microns deep (maximum imaging depth depended on the magnification and the objective used). For the figures shown here, however, single XY planes from within these stacks are shown, chosen by the study pathologist as the ones that best illustrate the morphologic features required for diagnosis. The MPM images shown in this study are morphologically similar, but not necessarily identical, to the images shown in their corresponding H&E sections, since it is not always possible to orient the tissues in both procedures in precisely the same way, so as to obtain the exact angle and plane of section in both cases.
All specimens imaged with MPM revealed flat (nonpapillary) architecture of the bladder mucosa at low magnification. Further characterization of these flat lesions as benign or malignant (including CIS and invasive high-grade urothelial carcinoma) was done at high magnification, based on their architectural (polarity/organization of the urothelial layers) and cytologic features (nuclear to cytoplasmic [N:C] ratio, nuclear pleomorphism, etc).
The biopsy specimens diagnosed as benign on H&E (Figure 1, A through D) showed multilayered urothelium along with surface umbrella cells (cellular autoflourescence; color-coded green) as identified on MPM. The N:C ratio of the umbrella cells appeared low with abundant cytoplasm. The underling lamina propria was composed mainly of collagen (second-harmonic generation signal; color-coded red) along with a few elastin fibers (autoflourescence; color-coded green). Many of the specimens that appeared cystoscopically suggestive of CIS either due to erythema or irregularity of the surface mucosa, were found to have benign pathologic features (Figure 2, A through F) such as florid proliferation of von Brunn nests or cystitis cystica et glandularis. In samples that were diagnostic of CIS on H&E, high-grade cytologic features of CIS were detected on MPM. These CIS specimens showed varied appearance of surface urothelium ranging from largely denuded to irregularly thickened and/or heaped-up urothelium at low magnification. In the high-magnification images shown in Figure 3, A through F, we could identify the anaplastic cytologic features characteristic of CIS (high N:C ratio and marked nuclear pleomorphism). In addition, the inset of Figure 3, C, shows some mononuclear inflammatory cells (confirmed as lymphocytes on H&E) in the lamina propria, which were distinguishable from surface malignant cells of CIS by their small size and regular shape. Specimens that were found to have invasion on H&E staining (Figure 4, A through D) showed malignant cells invading collagen fibers of lamina propria on MPM and could be classified as invasive high-grade urothelial carcinoma. Since MPM can only image to a depth of 250 μm at high magnification, we could only detect superficial invasion in the lamina propria (not muscle invasion).
In the blinded analysis (Figure 5), 78 bladder biopsy specimens were obtained from flat lesions of bladder via TURB, cold cup procedure, or cystectomy. Of these 78 samples, 46 were diagnosed as benign, 23 as CIS, and 9 as invasive high-grade urothelial carcinoma on H&E. Since the purpose of this study was to identify surface high-grade lesions and differentiate them from benign flat lesions, MPM diagnoses were clustered under 2 broad categories: (1) malignant lesions, including CIS and invasive high-grade urothelial carcinoma; and (2) benign lesions, which included all other flat lesions such as florid proliferation of von Brunn nests, cystitis cystica et glandularis, atypia, hyperplasia, and others.
In 77 of 78 samples (99%), accurate MPM diagnoses (benign or malignant) were given by the cytologic and architectural features (N:C ratio, pleomorphism, urothelial layer polarization and/or organization, etc). Correct (benign) diagnoses were given in all (46 of 46) benign samples (100%).
A correct diagnosis of malignancy was given in 31 of 32 specimens (97%) imaged. Of the 23 of 32 malignant cases that were CIS, 22 (96%) were correctly diagnosed on MPM. One was given a diagnosis of atypia and was classified under the benign category. The remaining 9 of 32 malignant lesions were of invasive high-grade urothelial carcinoma (as diagnosed on H&E), for which malignancy was correctly identified in all 9 samples. However, invasion was missed in 2 cases on MPM. The sensitivity and specificity of MPM in our analysis were 97% and 100%, respectively. A malignant diagnosis on MPM had a high positive predictive value of 100%, and benign diagnosis had a negative predictive value of 98% (Figure 5). The interobserver agreement, Cohen κ, was calculated to be 0.93.
COMMENT
In this study, we have demonstrated the capability of MPM to generate high-resolution images that are comparable with H&E images, from fresh, unfixed, and unstained ex vivo bladder specimens. Under WLC guidance or from cystectomy specimens, samples from both normal- and abnormal-appearing flat (nonpapillary) bladder mucosa, acquired for diagnostic purpose, were included in the study. In all the specimens imaged, flat architecture of bladder mucosa was apparent on MPM. Carcinoma in situ lesions were accurately diagnosed with high sensitivity and specificity from their cytologic (ie, high N:C ratio and marked nuclear pleomorphism) and architectural features (loss of polarity and organization of the urothelial layers). Many lesions that appeared abnormal cystoscopically showed benign inflammatory changes of cystitis cystica et glandularis and/or florid proliferation of von Brunn nests. Moreover, inflammatory cells, especially lymphocytes, were clearly distinguishable from malignant cells by their relatively small size and regular shape.
Multiphoton microscopy has been previously used to image ex vivo bladder biopsy specimens from rats and human subjects.26–28 Cicchi et al28 imaged paired cold cup biopsy specimens from 5 patients to differentiate between healthy bladder mucosa and CIS. Although they could differentiate the 2 lesions from their morphologic and spectroscopic differences, their sample size was too small to determine diagnostic accuracy. Jain et al26 conducted a larger blinded analysis of 65 ex vivo human bladder biopsy specimens. They reported an accuracy of 88% in diagnosing benign and malignant lesions with an overall sensitivity and specificity of 90% and 77%, respectively, for the MPM diagnosis of bladder cancer. A positive (neoplastic) diagnosis on MPM had a high predictive value (94%), and negative (benign) diagnoses were sustained on histopathology in two-thirds of cases. However, cytologic grade was accurate in only 38 of 56 cases (68%). A major limitation of that study was that the sample included both flat and papillary bladder lesions, and thus the representation of CIS lesions was very small (n = 14). Our current report thus builds on the previous study,26 with a focused goal of assessing the ability of MPM to distinguish CIS from other flat lesions of the bladder mucosa.
As a limitation of the current study, we acknowledge that interobserver reproducibility could pose a diagnostic difficulty of distinguishing moderate dysplasia from CIS, similar to the difficulty encountered on H&E.31 In the future, we propose to address this problem by quantitatively assigning a grade of dysplasia, based on parameters such as nuclear cytoplasmic ratio and nuclear pleomorphism, through morphometric analysis. Such an analysis, however, would require a large sample size of cases with dysplasia, which is beyond the scope of this current study.
As discussed earlier, CIS is often missed on initial diagnostic workup, affecting its early management, which leads to recurrence and subsequent progression to muscle-invasive disease.9 This is primarily due to the flat nature of the lesion and lack of cellular resolution provided by WLC in differentiating benign from malignant lesions. Recently, macroscopic techniques such as fluorescence cystoscopy/PDD and NBI have been used to scan large areas of bladder mucosa and flag suspicious lesions. Although these techniques have significantly improved the sensitivity of CIS detection to up to 98% and 90%, respectively, compared to 50% to 70% sensitivity alone with WLC,14,32 they have high false-positive rates of nearly 30% to 36% with consequent increase in the rate of unnecessary biopsies with benign diagnoses. The reason for these high false-positive rates appears to be that inflammatory lesions often have similar optical signatures at low magnification as those of CIS. New high-resolution optical imaging techniques, such as OCT, CLE, and endocytoscopy,16,18,20–23 are being explored to bring down the high false-positive rates associated with the abovementioned macroscopic techniques. These optical imaging techniques can characterize flat lesions at cellular level during in vivo imaging. However, each of these techniques has its own inherent limitations. For example, OCT can image to a depth of 2 to 3 mm and has been used to grade and stage bladder cancer; however, owing to a lower lateral resolution of 10 to 20 μm16 (in comparison to ~1-μm resolution of MPM), it cannot provide the same degree of cellular and subcellular details as MPM. Confocal endomicroscopy and endocytoscopy produce high-resolution images reminiscent of histology; however, they have shallow depth of imaging (at most 50 μm). In addition, both CLE and endocytoscopy require exogenous contrast agents such as fluorescein and methylene blue, respectively.33 Multiphoton microscopy, on the other hand, generates high-resolution images revealing features similar to those observed in histologic images, without the need for exogenous contrast agents, and has an imaging depth of up to 500 μm.
Miniaturized MPM endoscopic probes have been built and tested in vivo in rats to image different organs such as colon, liver, kidney, and bladder.29,30 These studies have shown the feasibility of performing MPM in vivo. Thus, from our encouraging results and the availability of miniaturized MPM probes, we foresee MPM as a potential tool for the detection and differentiation of CIS from other benign flat lesions during cystoscopy.
It should be noted, however, that image acquisition using currently available commercial MPM systems (such as the one used in this study) is quite slow (~1 frame/s). This may become limiting for translation to in vivo investigation in human subjects, owing to longer procedure time. The primary reasons for such slow image acquisition in traditional MPM systems are that (1) each 2-D image is constructed by raster scanning a single laser point over the entire field of view, and (2) a 3-D stack of images is acquired by physically moving the sample and the objective relative to one another in a stepwise fashion (using a mechanical device called stepper motor). However, several independent approaches are currently under investigation to overcome these limitations.34 Using one such approach (termed multiphoton multifocal microscopy), it is now possible to image live sperms at video rate (30 frames/s) and to obtain a Z-stack including twelve 2-D images in 1.2 seconds.35 Furthermore, some of these developments are being translated for use with miniaturized fiber-based multiphoton systems, which will allow the development of multiphoton endoscopes with high-speed scanning.36,37 Scanning speeds of up to 4 frames/s, with images simultaneously collected in 3 focal depths, has been reported when using these miniaturized devices.36
As such, in its current state, MPM would be most useful in assessing areas red-flagged by WLC and/or other low-magnification large-FOV techniques such as NBI or PDD. In one such approach, Schmidbauer et al38 used a multimodal technique to overcome the problem of small FOV, by using PDD to flag the suspicious lesions and then OCT for high-resolution confirmation. However, while PDD has been reported to have a higher sensitivity in flagging CIS lesions,17 it does require the use of an exogenous PDD dye, such as hexaminolevulinate. Thus, NBI offers a potentially useful alternative, especially for patients with hypersensitivity to the PDD agent,33 since NBI does not require the instillation of any exogenous agents. Another challenge could be acquiring high-resolution images from naturally nonflat biological tissue, since MPM requires the tissue to be at right angles (or at least at a fixed angle) relative to the imaging probe. However, such problems have been addressed to a point where such imaging is now clinically feasible with approaches such as CLE and OCT. This has been achieved in part by using contact probes, which can be gently pressed down on the tissue. If this stabilization and flattening of the tissue is inadequate for MPM, we can consider local application of low-intensity vacuum suction (just around the imaging tip). All commercial endoscopes are equipped with vacuum suction capabilities in any case, so this approach should be technically highly feasible.
Finally, to bring MPM from bench to bedside, extensive in vivo imaging in live anesthetized large animals needs to be performed, followed by phase 1 clinical trial in humans. As technology advances and similar challenges are addressed by the entire optical biopsy community, both faster imaging and increased FOV are expected in future-generation devices. Also, as multimodal approaches are further refined, it is likely that these optical biopsy techniques will be increasingly combined with point-of-care therapeutics (the developing field of theranostics). A recent study by Palmer et al39 has suggested that technologies such as MPM can be combined with photosensitizer-free tumor ablation, potentially yielding a point-of-care diagnostic monitoring and therapeutic device, which may reduce procedure time, improve outcomes, and lower cost. Various types of localized high-precision ablation techniques including cutting/vaporizing lasers, freezing, and high-frequency ultrasonography could be visualized as useful in such a context. Thus, it is likely that these techniques will continue to evolve to become more useful in patient care in the near future.
CONCLUSION
In conclusion, MPM is a potential diagnostic tool which, when integrated into miniaturized probes, could be used cystoscopically to accurately identify and differentiate CIS from other benign flat lesions of bladder. This may not only reduce the rate of repeated unnecessary biopsies with benign diagnoses and their associated complications, but may also allow earlier and accurate detection to reduce the rate of recurrence and progression to muscle-invasive disease, thus improving overall patient prognosis and survival.
References
Author notes
The authors have no relevant financial interest in the products or companies described in this article.