Magnification/chromoendoscopy

Improving the quality of white-light endoscopic images and increasing their information content has been an ongoing process. It includes incremental advances in endoscope technology, particularly the development of video endoscopes that in GI endoscopy are rapidly replacing fiber optic endoscopes (based on imaging bundles), and concomitant use of high-density charge-coupled device (CCD) detector arrays that have greatly increased the spatial resolution of white-light images. A standard WLE uses a CCD with $\mathrm{100,000}\text{to}\mathrm{300,000}\text{pixels}$ . With a field of view of a few ${\mathrm{cm}}^2$ , the pixel size on the image is typically $\sim 100\mu \mathrm{m}$ . High-resolution endoscopy (HRE) utilizes $\mathrm{400,000}\text{to}\mathrm{850,000}\text{-}\text{pixel}$ CCDs. Used in conjunction with high-magnification front-end optics (magnification endoscopy or endocytoscopy), this enables imaging down to close to the cellular level $(\sim 10\mu \mathrm{m})$ ,¹³ although at the cost of a much smaller field of view, so that this may more properly be thought of as an optical biopsy technique.

Magnification endoscopy is most useful when combined with a tissue stain applied topically to the mucosal surface (chromoendoscopy).¹⁴ Such stains are generally chromogenic, i.e., optical absorbers of specific color that bind to either normal or abnormal mucosa. Examples, illustrated in Fig. 3 , are Lugol’s solution that stains normal squamous epithelium a yellow-brown color, methylene blue that is absorbed into cells and highlights specialized columnar epithelium (as in Barrett’s), and indigo carmine that accumulates in pits and valleys between cells and highlights mucosal architecture. These stains help enhance the BE-SE (squamous epithelium) contrast, but have not shown high sensitivity for the main challenge of highlighting dysplasia within BE. The best is probably crystal violet, which stains cell nuclei. This gives high sensitivity (96%) and some specificity $(\sim 66\%)$ for detecting esophageal dysplasia, at least in a subset of patients.¹⁵ An alternative to chromogenic stains is acetic acid,¹⁶ which is widely used in colposcopy to highlight lesions in the cervix. This breaks down the viscous glycoprotein surface layer and provides higher contrast of the surface epithelium by modifying the tissue refractive index (whitening). Again, this has shown sensitivity and specificity for BE versus SE of 100 and 66%, respectively, but has not improved the detectability of esophageal dysplasia. In summary, “…chromoendoscopy of the upper GI tract has failed to establish its role in diagnosing Barrett’s oesophagus and Barrett’s associated neoplasia so far.” (Kiesslich and Neurath¹⁷). This is not the same situation as in, say, the colon, where staining, particularly combined with magnification endoscopy, is of proven value in identifying flat adenomas that are otherwise difficult to see.

Fig. 3

Examples of chromoendoscopy of BE. (a) Lugol’s solution (courtesy of J. Bergman and M. Kara), (b) methylene blue, (c) crystal violet (adapted from Ref. 15, with permission), and (d) acetic acid (adapted from Ref. 16, with permission).

A recent development that, in some respects, emulates the idea of chromoendoscopy to enhance the image contrast is narrow-band imaging (NBI), illustrated in Fig. 4 . In this technique, rather than the red-green-blue channels being combined into a balanced white-light image, as in standard endoscopy, each is displayed as a separate image. This enhances the contrast in mucosal morphology and highlights the microvascular pattern of the tissue, both of which are altered in the transition from intestinal metaplasia (BE) to dysplasia. The image contrast arises from the differential attenuation of the three spectral regions, with the blue channel biased toward the most superficial layers, and the red light having the deepest penetration. Sharma ¹⁸ used NBI with WLE in an open study without a control group in 51 BE patients, and obtained very high sensitivity (100%) and specificity (98.7%) for detecting characteristic morphological changes (ridge/villous pattern) in high-grade dysplasia. NBI was not able to distinguish between LGD and nondysplastic BE. A recent exploratory study by Kara ¹⁹ in 63 BE patients (split into a learning set and a validation set) also yielded a high (94%) sensitivity [negative predictive value $\left(\mathrm{NPV}\right)=98\%$ ] for detecting HGD in BE. However, as is often the case in early stage studies of new technologies for BE, the number of patients with dysplasia was disproportionately high relative to routine surveillance groups (due to preselection). Also, the specificity was only 74% [positive predictive value $\left(\mathrm{PPV}\right)=64\%$ ], which may be due to the fact that the morphologic/microvascular changes are not universally present in all cases. (The same group has shown that $\mathrm{HRE}+\mathrm{NBI}$ is equivalent to HRE with indigo carmine staining, although this was also a nonrandomized study and only in high-risk patients.²⁰) As is discussed later, the specificity may be improved further by combining NBI with autofluorescence endoscopy.

Fig. 4

Example of narrow band imaging in BE with dysplasia (courtesy J. Bergman and M. Kara, Amsterdam University Medical Center). The left images are standard white-light endoscopy, and the right are NBI showing the enhanced microvascular and mucosal morphologic contrast.

Fluorescence endoscopy

Fluorescence endoscopy has an interesting history.²¹ It was revived in the early 1980s as part of photodynamic therapy (PDT) to image the uptake and localization of the photosensitizers, particularly in treating bronchial tumors. Earlier imaging studies of fluorescent compounds as a diagnostic tool go back to the 1920s. Several decades later, an attempt to improve the diagnostic (i.e., fluorescence) characteristics of one of these agents, hematoporphyrin, led serendipitously in the 1960s to the discovery of hematoporphyrin derivative (HpD), the first PDT agent to be approved for human use in 1993.²² Using a fiber optic endoscope modified for fluorescence detection, Lam, Hung, and Palcic then performed a series of clinical bronchoscopic studies, successively reducing the dose of HpD to minimize skin photosensitivity.²³ Unexpectedly, they found that the highest detection rate for early stage lesions (dysplasia, carcinoma-in-situ) was with no HPD, i.e., utilizing the natural, endogenous fluorescence of the tissue. This autofluorescence broncho-scopy was then developed further and was recently approved by the FDA as an adjuvant to white-light bronchoscopy: the corresponding commercial system (Xillix Technology Corporation, British Columbia, Canada) is illustrated in Fig. 5 . In its most recent implementation, it comprises blue-light excitation with combined imaging of green fluorescence, detected by an intensified CCD camera, and red (diffuse) reflectance, detected by a nonintensified CCD.²⁴ In this mode, early neoplastic lesions show as (false color) reddish on a blue-green normal tissue background.

Fig. 5

Autofluorescence endoscopy system, showing block diagrams of the optical functions under white-light and autofluorescence imaging modes. In fluorescence mode, a source filter is inserted that selects blue light plus a small fraction of the red light from the lamp, the latter serving to normalize the green fluorescence signal that is detected on an ICCD camera (courtesy Xillix Technologies Corporation, British Columbia, Canada).

There is reasonable understanding of the underlying mechanism for the fluorescence contrast seen at endoscopy in the bronchus, namely mucosal thickening that causes preferential increased attenuation of the green fluorescence from submucosal collagen, together with loss of extracellular matrix that reduces the collagen signal.²⁵ The red channel here effectively serves to normalize the fluorescence image. At least four companies now have autofluorescence endoscope systems currently in clinical trials.

There have been several clinical studies to evaluate auto-fluorescence endoscopy (AFE) in the GI tract,²⁶ particularly in the colon to differentiate benign (hyperplastic) from premalignant (adenomatous) polyps²⁷ and to increase the detection of flat adenomas, and in BE.^{28, 29, 30} As in the bronchus, this is always done in combination with WLE, not as a replacement for it. In the colon we have found significant increase (from 58 to 81%) in the ability to differentiate premalignant (ademomatous) from benign (hyperplastic) polyps.²⁷ There is also a good understanding from confocal fluorescence microscopy of biopsies of the underlying tissue changes that enable this increase.³¹ Unfortunately, the situation in BE has been less encouraging. It has certainly been possible to visualize dysplasias that were missed on WLE and/or that are invisible under WLE, even after they have been localized by fluorescence, as illustrated in Fig. 6 . However, a recent randomized, crossover study comparing AFE and standard white-light videoendoscopy showed equivalent sensitivity and a lower specificity with fluorescence imaging.²⁹ Note, however, that in practice these techniques would be used in combination rather than as alternatives, so that the combined accuracy is still higher than that of either single technique. For example, another study³⁰ by the same group showed an increase in sensitivity for detecting high-grade dysplasia in BE from 23% for WLE alone to 33% for the combination, with several dysplastic lesions detected by AFE only. Nevertheless, it is clear that a significant problem with autofluorescence endoscopy in BE (at least with blue-light excitation) is the high level of autofluorescence in the nondysplastic BE mucosa itself and, particularly, the high point-to-point and patient-to-patient variability in this (see section on point spectroscopy). Primarily, these effects increase the false-positive rate to a level that is presently unacceptably high using white light and autofluorescence endoscopy alone. For example, a PPV of only 49% was recently reported by Kara, ³⁰ even in a selected group of patients. We have recently shown by confocal fluorescence microscopy that dysplasia in BE is not accompanied by any significant alteration of the intrinsic fluorescence signature (i.e., in the endogenous fluorophores) of the epithelial layer compared with nondysplastic BE, at least with blue-light excitation.³² Hence, any diagnostic algorithm would need to exploit other tissue changes, such as altered tissue morphology or vascularity.

Fig. 6

Example of autofluorescence endoscopy (right), showing the high visibility of the dysplastic lesion in BE compared with the white light image (left). (courtesy M. Kara and J. Bergman, Amsterdam University Medical Center).

The main options to improve this situation are:

Approach 2. is considered later. For 1. there are several studies of fluorescence spectroscopy of various tissues using UV excitation (see Ref. 32), rather than the blue light that is employed in current AF endoscopic imaging systems. The rationale is that this provides access to additional endogeneous fluoropores,³³ particularly collagen and NADH that are often altered in malignancy. However, initially, AF endoscopy used fiber optic endoscopes, for which the transmission at short wavelengths is relatively poor. With the introduction of video endoscopes with AF capability, this may be less of a limitation. In addition, there is a concern for mutagenesis with UV light, although this may not be significant in practice at the light exposures typically used in optical diagnostics, at least in point spectroscopic mode.³⁴

There is also, in principle, the option to use multiple fluorescence excitation and/or emission wavelengths to improve the diagnostic power of AFE. Most work on these approaches to date have concentrated on point spectroscopy (see next), since implementation of such schemes becomes technically complex and, hence, prohibitively expensive in endoscopic imaging mode. However, since technology advances may alleviate some of the limitations of the multispectral approach, including swept laser sources and fast tunable filters, this may be a future option if there is compelling evidence that using more wavelengths would significantly improve the diagnostic accuracy. Note that this would not necessarily have to be justified on the basis of the BE application, if other endoscopic scenarios would benefit enough to drive the development of clinical systems: improved BE diagnosis would then be a side beneficiary. A compelling reason for such developments may be in providing spectral matching for multiplexed contrast-agent imaging (see next), more than for imaging autofluorescence.

In AFE clinical systems to date, there have usually been two detection channels, primarily to reduce the confounding effects of factors such as varying endoscope-to-tissue distance and angle and tissue shadowing, to which single-channel detection is very vulnerable. Thus, for example, in the original system used in the lung and GI tract,²⁵ the green fluorescence signal was normalized by the red fluorescence in each image pixel, so that when the geometric factors changed, their effects (partially) cancelled out. In the latest implementation,²⁴ the red fluorescence has been replaced by the red diffuse reflectance signal, both to reduce cost (since reflectance imaging does not need an intensified CCD) and, for colonoscopy, to make the signal less disrupted by the strong red fluorescence (porphyrins) from fecal matter. There is now active research into whether there are other, or additional, reflectance channels that would further improve the sensitivity and/or specificity of AFE, particularly exploiting possible alterations in the diffuse reflectance spectrum due to differences in tissue blood content or oxygenation. Again, in imaging mode, the number of detection channels that can be used, even for reflectance, is restricted by complexity, speed, and especially cost, at least using conventional optical technologies.

Note that all the studies discussed have used steady-state fluorescence: the option of time-resolved measurements are considered under point spectroscopy later, since it poses major technology challenges for endoscopic imaging.

For approach 3. in the simplest conceptualization, fluorescent contrast agents are analogous to chromogenic stains for WLE. However, given the history of autofluorescence endoscopy, as outlined before, an initial focus was on using fluorescent PDT agents.^{21, 23} In particular, an appealing option has been to use aminolevulinic acid (ALA). This is a rate-limiting precursor in heme biosynthesis such that, when given in excess, it leads to increased levels of the fluorescent photosensitizer protoporphyrin IX (PpIX).³⁵ Many studies in different organs, such as the bladder, cervix, and lung, have shown that ALA-PpIX can give high fluorescence contrast in early malignancy,^{21, 35} most likely due to increased metabolic rate and/or to reduced ferrochelatase, the enzyme responsible for converting PpIX to heme. ALA-PpIX has been studied extensively for photodynamic therapy for numerous early stage lesions, including dysplasia in BE, as discussed later. There have been several clinical studies of ALA-PpIX as a fluorescence endoscopy contrast agent in BE. For example, in a dose-ranging study in 47 BE patients, Endlicher ³⁶ found a sensitivity of $>80\%$ for dysplasia detection, but the specificity was only 27 to 56%. A recent study by our group³⁷ has confirmed that there does not seem to be a combination of ALA dose or time interval between (systemic) administration and imaging that gives adequate reliability. The problem again is that there is high, and highly variable, PpIX synthesis in the nondysplastic BE itself that gives a confounding fluorescence background.

An alternative to PDT sensitizers is to use a fluorophore linked to a targeting moiety that has specificity for the diseased tissue. In this approach, the optical properties of the fluorophore can be selected independently of its delivery to the tissue. This falls into the category of molecular imaging, which is being explored for many applications.³⁸ There are essentially three elements in such a contrast agent: the fluoropore, the targeting moiety, and the linker between them.

Fluorophores

Possible fluorophores include organic dyes and the recently introduced quantum dot (Qdot) semiconductor nanoparticles. The former have been used widely in many preclinical studies of tumor detection/localization (see Ref. 38), mainly using far-red or near-infrared emitting dyes, to achieve maximum light penetration in tissue. For BE, Polglase ³⁹ and Kiesslich ⁴⁰ used intravenous fluorescein (not linked to any delivery or targeting moiety) as a fluorescent contrast agent during confocal endoscopy (see next), with excitation at $488\mathrm{nm}$ , since deep light penetration is not required. Wróblewski ⁴¹ have used a water-soluble copolymer drug delivery system targeted to the surface glycoproteins that are expressed on BE mucus-secreting cells, and labeled this with the fluorophore fluorescein isothiocyanate (FITC) for fluorescence microscopy of human BE biopsies. To our knowledge this has not been reported to date in vivo and the question remains as to whether this would also differentially target dysplasia within BE.

Compared to organic fluorophores, Qdot semiconductor nanopartices have high brightness, low photobleaching, a broad excitation spectrum, and relatively narrow and size-dependent (and hence, selectable) fluorescence emission spectra.⁴² As summarized in Table 1 , this confers several potential advantages for their use in fluorescence endoscopy, some of which could increase the efficacy of fluorescence-based diagnostics, including BE endoscopy. However, Qdot-based contrast agents have two main potential limitations for in vivo use. First, there is evidence that the maximum size of Qdots that can be administered systematically may be quite limited, since larger particles tend to be sequestered in organs such as the liver, spleen, and kidney, and hence do not reach the site of interest in adequate concentration. For example, renal filtration of Qdots shows a sigmoidal dependence on size with the 50% point at around $5\mathrm{nm}$ (hydrodynamic diameter, including surface modifications).⁴³ This would be a severe constraint in many in vivo fluorescence imaging situations. For any given Qdot material and structure, the emission wavelength increases monotonically with size. Thus, for example, $10\text{-}\mathrm{nm}$ CdTeZnS Qdots emit at around $750\text{to}800\mathrm{nm}$ . Hence, a possible solution is to switch materials, such as InS Qdots, that emit longer in the near-infrared. However, for endoscopic detection of early lesions, such as dysplasia in BE, this may not be an issue, since relatively short wavelength (blue-green) excitation gives adequate mucosal penetration. Indeed, if there is less than perfect target specificity of the contrast agent, then this may actually be preferable to longer wavelengths, for which the mucosal contrast is decreased by contributions from deeper layers. Conversely, the autofluorescence background is generally higher at shorter wavelengths (as exploited in autofluorescence endoscopy). Hence, there are several counteracting factors in the optimal emission wavelength range for targeted Qdot (and organic fluorophore)-based fluorescence endoscopy, and this is the subject of current preclinical investigations,⁴⁴ an example of which is shown in Fig. 7 , where Qdots conjugated to an antibody for targeting were applied in an animal model of BE.

Fig. 7

White-light (upper) and fluorescence (lower) images of the luminal surface of the esophagus in a BE rat model (immediately postmortem), at $24\mathrm{h}$ following intravenous application of a conjugate comprising $660\text{-}\mathrm{nm}$ -emitting Qdots conjugated to an antibody targeted to villin that is overexpressed in BE tissue (courtesy R. da Costa and W. Chan, University of Toronto).

Table 1

Potential advantages of Qdots compared to organic fluorophores for fluorescence endoscopy

The second concern is the possible toxicity of Qdots, due to their heavy metal content (e.g., cadmium).^{42, 45} The in vivo lifetime and integrity of the coatings of Qdots are major factors in ameliorating this toxicity, while Qdots of less toxic elements will also help to overcome this limitation. For endoscopic applications, such as in BE, applying the agent topically may mitigate some of this concern, depending again on the stability of the Qdots as they then pass though the GI tract and/or are cleared systematically.

Targeting moieties

The second element in developing fluorescence contrast agents for detecting dysplasia in BE is perhaps the most challenging, since it requires that there be alteration in one or more cell surface markers that are associated with transformation to dysplasia and that are expressed at a much lower level in the Barrett’s tissue itself. Although numerous markers have been identified at different stages along the cancer initiation and progression pathway in the esophagus, until very recently none were known that have this specific property. At least one marker (TRAIL/Apo2L) has been identified that is down-regulated in dysplasia.⁴⁶ Unfortunately, this would yield negative fluorescence contrast, which is not a good feature for high-sensitivity imaging. In 2006, Dorer and Odze⁴⁷ reported an enzyme, $\alpha$ -methylacyl-CoA racemase (AMACR), that showed positive immunohistological staining (using an anti-AMACR antibody) with 81% sensitivity for high-grade dysplasia in BE, which is very encouraging. However, this is cytoplasmic rather than being expressed on the cell surface, which may make it difficult to target the fluorophore in vivo. On the other hand, as an active enzyme, it could be suitable for molecular beacon-based contrast (see next). A second potential marker is Survivin, a member of an apoptosis inhibitor gene family, which has shown increased $m\mathrm{RNA}$ expression in dysplasia compared to Barrett’s,⁴⁸ although there was a large overlap in the ranges of expression. Other options besides antibodies for targeting include specific peptide sequences, aptamers, or antisense sequences. There are no ${\text{dysplasia}}^{+}∕{\mathrm{BE}}^{-}$ examples at the present time.

Linkers

The third element may be simply a passive linker that couples the fluorophore and target molecule without interfering significantly with the properties of either. Examples include covalent bond formation between carboxylic acids and biomolecules, steptavidin-biotin linking and cross-linking agents that are commonly used to couple carboxyl groups to primary amines. The optimum linker depends strongly on the specific physicochemical properties of the fluorophore and targeting moiety. It is important to minimize fluorescence quenching, especially with organic dyes. Alternatively, the molecular beacon concept could be used, in which the fluorophore is deliberately quenched through Forster resonant energy transfer until the linker is cleaved (e.g., by a disease-specific enzyme) or opened (in the case of a target-specific mRNA antisense sequence). These approaches have been demonstrated in various preclinical models, both for fluorescence imaging⁴⁹ and, more recently, for photodynamic therapy.⁵⁰

Fluorescence

Clearly, autofluorescence spectroscopy (AFS) is an option and indeed has been used often as a means to determine the optimal spectral range(s) to apply in AF imaging. As an alternative to autofluorescence endoscopy, it has the advantages of being low cost and relatively easy to implement in the clinical setting. Typically a fiber optic probe is used, placed within the instrument channel of the endoscope. This is, however, not without its own problems.

These considerations apply to any endoscopic site, but the esophagus is particularly difficult because of the limited angle from which one can approach the tissue surface and its high motility due to the heart beat. In addition to these technical challenges, the fundamental problem with point fluorescence spectroscopy in BE, as illustrated by Fig. 8, is the large variability in the spectra from point to point in a given patient and between patients. Hence, even though on average there may be systematic differences with disease state, the variability makes it difficult to establish a robust spectral algorithm to detect dysplasia without the context of the autofluorescence appearance of the surrounding tissue that is obtained in imaging mode. It is not clear whether this variability will apply also with UV excitation, which Bourge-Heckly ⁵¹ have reported in a small patient series in vivo (using $330\text{-}\mathrm{nm}$ excitation and, interestingly, ratios of NIR fluorescence intensities rather than the more conventional visible range). Other advances, such as the use of time-resolved fluorescence, may also increase the diagnostic accuracy of point autofluorescence spectroscopy in BE, either in full time-resolved mode or using time gating,⁵² since this provides complementary information (on the tissue microenvironment) to steady-state spectroscopy. This approach has been implemented in the past in endoscopic imaging mode in the frequency domain.⁵³ However, the technology was very expensive, since it required use of multiple high-frequency modulated ICCD cameras.

Fig. 8

Autofluorescence point spectra (blue-light excitation) in BE patients taken in vivo during routine endoscopy. Left: patient-averaged spectra for normal (squamous), gastric metaplasia, intestinal metaplasia (BE), and dysplasia. Right: individual measurements (courtesy N. Marcon and R. DaCosta, University of Toronto).

Is there a role for point spectroscopy combined with exogenous fluorophores? In other applications, this seems likely, especially using molecularly targeted contrast agents. A prime example is in fluorescence-guided tumor resection, where the strong signal with exogenous fluorophores, coupled with highly disease-specific targeting, could be used to probe the surgical margin for residual tumor (or tumor-associated) tissue. This might be used in combination with fluorescence imaging⁵⁴ providing fast scanning of the resection surface, while point probing is used to check for subsurface extension of tumor and/or to enable quantification of the fluorescence signal, thereby providing an objective threshold for the presence or absence of tumor. In the case of dysplasia detection in BE, the approach faces two challenges: first, the general limitation of point spectroscopy versus imaging, and second, the need for a dysplasia-specific biomarker.

Reflectance

There is some confusion of terminology in this field between the terms (diffuse) reflectance spectroscopy (DRS), elastic scattering spectroscopy (ESS), and light scattering spectroscopy (LSS). The key factor that separates the techniques into two distinct classes is the extent to which there is an attempt to suppress or subtract the diffuse, i.e., multiply scattered, photons from the measured spectra to enhance the contribution of single- or few-scattering photons. If this can be achieved, either by using cross-polarization to select the true backscattered light or by numerical fitting and subtraction of the more slowly varying diffuse spectrum, then the fine spectral structure can be determined and depends on the details of the Mie-like scattering by the surface layer(s) of the tissue. The term LSS are used here for this technique. Diffuse reflectance spectroscopy (DRS) refers to measurements of the total spectrum of light from the tissue surface: in some cases, this can be (partially) separated into the absorption and elastic scattering (transport coefficient) components.

For DRS, many of the same considerations discussed for point fluorescence spectroscopy also apply. In an ex vivo study using tissue biopsies, Lovat ⁵⁵ found 92% sensitivity but only 60% specificity for detecting high-risk disease (high-grade dysplasia or cancer). Georgakoudi ⁵⁶ found reasonable sensitivity (86%) and very high specificity (100%) in the transport scattering spectra ${\mu }_s^{\prime }\left(\lambda \right)$ between HGD and $\mathrm{LGD}+\mathrm{BE}$ . Recently, Sterenborg⁵⁷ used differential path-length spectroscopy in the esophagus (reported previously, for example, in the optical biopsy of the breast⁵⁸) and found some evidence for differences in the hemogobin saturation with this semiquantitative technique, although the diagnostic accuracy for detecting dysplasia in BE is not known at this time. Recent work from our own laboratory supports this finding. Hence, in general there is likely diagnostic information in both the absorption and/or scattering spectra in DRS, although the differences are fairly subtle and it remains to be seen how robust any algorithms based on these would be in routine clinical practice.

LSS has been applied to a variety of epithelial tissues. By fitting the Mie theory to the detailed spectra, with the scattering particle size distribution as a free parameter in the fit, it has been possible to show differences in these distributions,⁵⁹ likely associated with altered nuclear size, which is used as one histopathological marker of dysplasia due to increased cell ploidy and chromatin. Georgakoudi ⁵⁵ found a sensitivity of 100% and specificity of 91% for HGD versus $\mathrm{LGD}+\mathrm{BE}$ using this technique. A significant advance in LSS was reported recently by Yu at MIT,⁶⁰ in which the backscatter spectra were measured at the tissue surface for both large and small values of the azimuthal angle between the incident light polarization and the scattering plane. This further isolated the contribution of nuclear scattering from that of other smaller organelles in the epithelial cells, as demonstrated by in vivo measurements in the colon in a small number of patients. As shown by Gurjar, ⁶¹ LSS can be extended to imaging, using a CCD camera and a set of narrow-band filters to sample the reflectance spectrum and, subsequently, extracting the nuclear size parameters at each image pixel point. This has not been reported to date for BE. There are significant technical challenges to achieve adequate spatial resolution and speed to make this a true endoscopic imaging technique that can be implemented at reasonable cost, and detailed studies will be needed to determine what spatial and spectral resolutions are needed to distinguish dysplasia in BE. Nevertheless, the importance of the method is that it exploits micromorphological features of tissues that are well established in clinical pathology as valid markers of dysplasia.

Raman

Rather than using such structural features of the tissue as the diagnostic parameter, Raman spectroscopy (RS) probes the biochemistry of the tissue. In RS, the spectrum of light that is scattered inelastically from tissues is measured, which reveals the vibrational or rotational energy states of common molecular bonds, such as those in proteins, lipids, and nucleic acids.⁶² Hence, the spectra are typically much richer than (auto)fluorescence spectra that show only the electronic excited states. The practical challenge comes from the very low Raman signal, which is several orders of magnitude weaker than the autofluorescence. As a result, the background fluorescence from the tissue, and Raman and fluorescence contamination from the optical components themselves, particularly the fiber optics in endoscopic applications, are high relative to the true tissue Raman signal. In vivo endoscopic near-infrared Raman spectroscopy became technically possible only in the late 1990s with the availability, at reasonable cost, of compact diode lasers, high-efficiency notch filters, high-throughput holographic spectrometers, high-sensitivity CCD detectors, and specially designed fiber optic probes with in-line spectral filters. These enabled systems, such as that shown in Fig. 9 , that can collect a tissue spectrum with good signal-to-noise in a few seconds at safe light exposure levels.^{63, 64}

Fig. 9

In vivo NIR Raman spectroscopy system showing (a) schematically the laser, fiber optic probe, and spectrograph/CCD detector; (b) the details of the probe tip (central light delivery fiber, six circumferential collection fibers) with in-line spectral filters and beveled ends to increase the overlap between the illumination and collection light cones in the tissue; (c) the system in use with the probe placed within the instrument channel of an endoscope and in contact with the tissue surface; and (d) a typical NIR Raman spectrum of tissue with some of the main peak assignments (adapted from Ref. 62, with permission).

As has been seen in other sites, the changes in the NIR Raman spectra with dysplasia are fairly subtle, as illustrated in Fig. 10a . It is generally not the case that simple analysis, such as comparing the intensities of individual peaks in the Raman spectra, provides statistically robust diagnostic algorithms.⁶⁵ Rather, the diagnostic information is spread throughout the whole spectrum, so that chemometric approaches^{66, 67} have proven necessary (and sufficient), primarily principal component analysis (PCA), followed by linear discriminant analysis. This is perhaps not surprising, since the Raman spectra do not, except in a few very special cases (e.g., tissue calcification), correspond to specific biomolecules. Rather, it is the alteration in factors such as the total protein-to-lipid ratio or the DNA content that provides the diagnostic information.

Fig. 10

NIR Raman in BE. (a) Average in vivo spectra averaged in 65 BE patients. (b) Component histograms from decomposition of Raman spectra measured in different regions on an ex vivo tissue biopsy (regions 1 and 2 -SE, regions 3 and 4-HGD). (c) Principal component map of the biopsy (adapted from Ref. 70, with permission).

NIR Raman spectroscopy has been found to have high diagnostic accuracy in a variety of tumor sites. In the case of BE, an ex vivo (biopsy) study in 44 BE surveillance patients showed 88% sensitivity and 92% specificity for HGD versus a consensus pathology classification model.⁶⁷ It also highlighted an important issue for RS and any other optical techniques that have to be trained against the gold standard of histopathology, in demonstrating that multiple independent pathologists (at least three according to that study) are needed to account for the high interobserver variability. Recently, we have obtained sensitivity and specificity for differentiating high-risk (i.e., HDG or adenocarcinoma) from low-risk (LGD or BE) tissues in the 90% range using the in vivo endoscopic Raman system shown in Fig. 9, which is consistent with the ex vivo findings. (Note that we had earlier reported⁶⁸ the tissue-handling conditions under which ex vivo RS measurements are valid, an issue that has often been overlooked in assessing new optical techniques using tissue samples.)

Continuing evaluation and endoscopic applications of RS are hampered by the nonavailability of suitable fiber optic probes that provide adequate filtering of the fiber fluorescence and Raman signals and of the tissue autofluorescence, since the original commercial unit (Visionex, Atlanta, GA) is no longer available. Alternative probe designs have been explored⁶⁹ to improve the performance and/or get around the original patents, but again these are not available to the general community. To circumvent this problem, Koljenovic in Rotterdam⁷⁰ recently suggested using the $2400\text{to}3800\text{-}{\mathrm{cm}}^{-1}$ wavenumber range, where the silica Raman background is much lower than the standard fingerprint region of $400\text{to}1800{\mathrm{cm}}^{-1}$ . Equivalent diagnostic efficacy for the two wavenumber regions was demonstrated in brain tumor and bladder tissues ex vivo, even though the actual spectral information content may be different. This is a promising new approach that bears exploration for application to BE.

An important advance in understanding the biochemical basis for the observed spectral changes has been recently reported by Shelly ⁷¹ using microscopy to map the full Raman spectrum across tissue biopsy specimens. PCA was then used, and maps of the principal component values were generated [Fig. 10c]. In addition, by fitting the basis spectra of known tissue constituents to the measured spectra, the relative concentration of each constituent was determined for each image pixel [Fig. 10b]. Signatures for DNA, oleic acid, collagen 1, and actin were associated with HGD compared to normal squamous esophagus, in which glycogen showed significant peaks. Further studies on a larger number of samples, including the full range of esophageal pathologies, are needed, but this study has shown the value of such detailed spectral mapping. It would, of course, be a significant advance if such maps could be produced in vivo during endoscopy, since this could increase even further the diagnostic power of the Raman characteristics.

With spontaneous Raman, as used to date, imaging or even crude mapping is impracticable during endoscopy: the signal is simply too weak ( $\sim {10}^{-8}\times \text{elastic}$ scattering, ${10}^{-6}$ to ${10}^{-4}\times \text{autofluorescence}$ ). There are two possible ways in which this might be overcome. The first is to use coherent anti-Stokes Raman spectroscopy (CARS) a nonlinear version of RS that has an intrinsic signal strength $\sim {10}^2$ to ${10}^4$ times higher than spontaneous Raman, for which video-rate in vivo CARS microscopy has been reported,⁷² although not for BE tissues to date. An interesting challenge will lie in the spectral analysis in this case, since the signal depends quadratically on the molecular concentrations, so that linear chemometric algorithms will not be applicable. The laser sources required are also currently fairly complex and expensive for clinical endoscopic use. The second possibility would be to use surface-enhanced Raman spectroscopy (SERS), exploiting the huge (up to about $\sim {10}^{15}$ fold) increase in the Raman signal from molecules that are close to a metal surface, particularly metal nanoparticles. One could envisage two ways in which this might be applied: by administering targeted metal (e.g., gold) nanoparticles topically or systemically, which would accumulate preferentially in dysplasia, or by using a metal-coated fiber optic probe placed in contact with the tissue. A bioprobe based on the latter approach has been recently reported for in vivo glucose monitoring,⁷³ one of the other pivotal challenges in biophotonics. Each of these approaches raises many issues. For example, in the first case, the same problem of targeting applies as in fluorescence contrast agents, although toxicity is much less of a problem with gold NPs than with quantum dots. In addition, since the SERS enhancement factor varies inversely as the sixth power of the distance between the metal surface and the Raman-active molecule, the microlocalization of the targeted material will be critical. This is also a factor in the second approach, since intimate contact with the tissue surface is required.

Endocytology

Microendoscopy, or endocytology, is the extreme extension of high-magnification endoscopy, in which, using a contact endoscope with high-magnification optics, real-time in vivo images of the tissue surface ( $\sim 5\text{to}15\text{-}\mu \mathrm{m}$ depth) can be obtained with close to cellular resolution $(\sim 2\text{to}4\mu \mathrm{m})$ over a submillimeter field of view $(\sim 100\text{to}300\mu \mathrm{m})$ . For example, Inoue ⁷⁴ reported initial studies in the esophagus, using methylene blue staining of the mucosa, with optical magnification $>1000$ .

Confocal endoscopy

The principle of confocal imaging, used widely in microscopy, has been transferred into endoscopy in the past few years, with several different technological approaches reported. As in microscopy, the concept is to perform point scanning across a specific depth in tissue (up to a few hundred microns) by focusing a laser beam at this depth and rejecting out-of-plane photons by a pinhole placed in front of the detector. This can be done either in reflectance or fluorescence mode, the latter being achieved simply by using narrow-band excitation and filtered detection. In a typical confocal microscope, the optics themselves are macroscopic and must be down-scaled by a large factor for endoscopic use. Figure 11 shows a system with direct miniaturization of the optics,³⁹ which is the approach that is most developed as a clinical system (Optiscan/Pentax), together with examples of the image quality and resolution with this system, using fluorescence contrast enhancement. This system has a $0.7\text{-}\mu \mathrm{m}$ depth resolution over a focal range up to $200\text{-}\mu \mathrm{m}$ depth, with $1024\times 1024$ images, generated at approximately 1 frame per second. With this device, Keisslich ⁴⁰ obtained $>90\%$ diagnostic accuracy for detection of neoplasia in BE patients. Two examples of alternative technological solutions, as reviewed in Ref. 75, are illustrated in Fig. 12 , namely the use of MEMS at the distal end of the endoscope to perform the light beam scanning or placement of the scanning outside the body (proximally) with a digital micromirror (DMD) array to select specific fibers in a coherent fiber bundle.

Fig. 11

High resolution confocal endoscopy (endomicroscopy) based on miniaturization of the distal scanning optics at the tip of a probe placed through the endoscope biopsy channel (from Ref. 39, with permission). The lower images show in vivo frames of Barrett’s and dysplastic glands, demonstrating near-histological resolution, taken under fluorescence with intravenous fluorescein contrast enhancement. Field of view $500\times 500\mu \mathrm{m}$ (adapted from Ref. 40, with permission).

Fig. 12

Two alternative technologies for confocal endoscopy. Top: MEMS-based system. Bottom: DMD-imaging fiber bundle configuration (adapted from Ref. 75, with permission).

The confocal approach has the advantage over direct ultrahigh-magnification endoscopy of allowing depth scanning, and is probably more amenable to fluorescence implementation. However, improvements are required to increase the frame rate and to achieve greater depth of penetration: the latter will probably require moving from the visible into the NIR wavelength range. Another possibility, at least conceptually, would be to use two-photon excitation: as in microscopy, this would have the advantage of greater depth of imaging and reduced photobleaching, but is likely to be cost prohibitive in the near term, because of the need to use a fs laser source, and dispersion in the fiber optics could also be a problem.

Optical coherence tomography

We⁷⁶ and others^{77, 78} have developed endoscopic OCT, the optical analog of (high-frequency) ultrasound, with which it has been compared in BE.⁷⁷ A system is illustrated in Fig. 13 that provides both structural and Doppler microvascular images. Ex vivo and in vivo studies have demonstrated the capability of OCT to image the changes in tissue microarchitecture, including the loss in BE of the well-ordered layered structure of the normal squamous epithelium and the increased microvasculature in the latter, consistent with observations made by histology, reflectance spectroscopy, and confocal microendoscsopy. The advantages of OCT compared to confocal endoscopy are the higher speed (which allows video-rate imaging with good signal-to-noise ratio), the greater imaging depth (up to $\sim 2\mathrm{mm}$ compared with $<500\mu \mathrm{m}$ ) and, with Doppler, direct microsvascular blood flow imaging. The deeper (cross sectional) imaging may be particularly valuable for staging of neoplastic lesions, if it is possible to see submucosal invasion, since this is a critical determinant of whether or not local endoscopic ablation can be used. At present the imaging depth is borderline for this, but should be improved by recent technological advances (see later). A recent study in 34 BE patients showed a 62% sensitivity and 82% specificity for detecting dysplasia (50 and 72% for HGD alone).⁷⁹ This system did not include Doppler capability, so that the diagnosis was based solely on the structural features of the OCT images. The authors concluded that further improvement in the spatial resolution ( $25\text{-}\mu \mathrm{m}$ lateral, $10\text{-}\mu \mathrm{m}$ axial) and in the characterization of OCT features in dysplasia are required. The intraobserver variance was also high, which may speak to the difficulty of interpreting the OCT structural images.

Fig. 13

Endoscopic Doppler OCT system. Left: schematic of the system and the side-looking fiber optic probe that is inserted through the endoscope biopsy channel. Right: in vivo Doppler images superimposed on the structural OCT images: (a) normal squamous and (b) BE. (c) and (d) show the corresponding histology.

A possible further limitation of OCT compared with other techniques is that it cannot operate in fluorescence mode, which restricts the use of contrast agents to those such as gold nanoparticles^{80, 81} that enhance the optical backscattering. The pros and cons of this approach are discussed later. Until recently, endoscopic OCT also suffered from the same fundamental limitation as the other point techniques in being able to scan only a very small field of view, thus restricting its use as a rapid surveillance tool, albeit with very high potential information content to allow in situ diagnosis. A major breakthrough has been the replacement of the established time-domain (TD) OCT by frequency-domain OCT.⁸² In TD-OCT, as the reference mirror is scanned (see Fig. 13), the tissue is sampled along the depth (A-scan) axis point by point. At any instant, light from all other depths along the A-scan is rejected by the coherence gate (that provides the high depth resolution). Thus, most of the light is wasted, which sets the tradeoff between imaging speed and signal-to-noise in the images. Two methods have been demonstrated that utilize the full back-reflected signal: spectroscopic OCT, in which the backscattered light is spectrally dispersed (through a spectrograph onto a CCD detector), or swept-source OCT, in which a single point detector is used but the wavelength is swept rapidly through the spectra bandwidth of the system. These Fourier-domain techniques, while technologically challenging and conceptually complex, are both feasible, with various advantages and limitations in terms of cost, ease of implementation, and performance. From the clinical perspective, the critical point is that they can be used to speed up OCT by a large factor, so that large-area scanning becomes possible, as illustrated in Fig. 14 , where several centimeters of esophagus has been imaged in a few seconds, while retaining comparable subsurface depth and resolution to TD-OCT.⁸³ Clinical trials of these technologies are in progress, and it is conceivable that this approach could become an important component in BE endoscopy. Recently, Doppler capability has also been reported with this system.⁸⁴ Technical challenges for routine clinical use include: 1. how to make this compatible with biopsy (since at present a balloon is used to center the OCT probe in the esophagus), 2. how to balance the speed versus imaging depth, and 3. how to process and display to the endoscopist the huge digital image dataset that is generated and to coresister it with standard white-light endoscopy/biopsy.

Fig. 14

Swept-source OCT images of the normal (pig) esophagus in vivo in the region of the squamous-columnar junction, generated using a rotating probe that generates cross sectional images that are then translated along the axis of the esophagus. The 3-D rendering is shown in (a) and the corresponding 2-D cross sections in (b), (c) and [plus histology in (d)] (e) (reprinted from Ref. 83, with permission from the American Society for Gastrointestinal Endoscopy).

Table 2

Advantages and limitations of endogenous versus exogenous contrast for dysplasia detection in BE.