In Vivo Optical Detection of Dysplasia in Barrett’s Esophagus with Endoscopic Light Scattering Spectroscopy

Abstract

The incidence of adenocarcinoma of the esophagus is increasing more rapidly than any other type of carcinoma in the United States. Almost 100% of cases occur in patients with Barrett’s esophagus, a benign condition in which metaplastic columnar epithelium replaces the normal squamous epithelium of the esophagus. Although the prognosis of patients diagnosed with adenocarcinoma is poor, the chances of successful treatment increase significantly if the disease is detected at the dysplastic stage. Recently, we developed the clinical endoscopic polarized light scattering spectroscopy imaging instrument which performs rapid optical scanning and multispectral imaging of the entire esophageal surface and provides diagnoses in near real time. By detecting and mapping suspicious sites, guided biopsy of invisible, precancerous dysplasia becomes practicable.

Keywords

Dysplasia, light scattering, imaging, spectroscopy

7.1 Introduction

The incidence of adenocarcinoma of the esophagus is increasing more rapidly than any other type of carcinoma in the United States . Almost 100% of cases occur in patients with Barrett’s esophagus (BE) , a benign condition in which metaplastic columnar epithelium replaces the normal squamous epithelium of the esophagus. Although the prognosis of patients diagnosed with adenocarcinoma is poor, the chances of successful treatment increase significantly if the disease is detected at the dysplastic stage. The surveillance for dysplasia of patients with BE is challenging in three respects. First, dysplasia in nonnodular BE is not visible during routine endoscopy . Thus, numerous systematically located biopsy specimens are taken in a prescribed pattern. Although systematic, this procedure is referred to in the literature as “random,” and we will also use this term. Second, the histopathologic diagnosis of dysplasia is problematic because there is poor interobserver agreement on the classification of a particular specimen, even among expert gastrointestinal pathologists . Third, reliance on histology imposes a time delay between endoscopy and diagnosis, requiring patient recall if disease is discovered as opposed to prompt treatment during surveillance.

Once BE has been identified, most gastroenterologists will enroll the patient in an endoscopy/biopsy surveillance program, presuming that the patient is a candidate for ablative therapy should low-grade or high-grade dysplasia be detected. Although the cost effectiveness of this type of surveillance program has not been validated in prospective studies, the lack of such studies does not preclude its potential usefulness. Patients with BE who have esophageal carcinoma detected as part of such a surveillance program are more likely to have curable disease and have an improved 5-year survival as compared with those whose cancer was detected outside of a surveillance program.

Dysplasia in the gastrointestinal tract is defined as neoplastic epithelium confined within an intact basement membrane. Dysplasia in BE can be classified as low or high grade, based on the criteria originally defined for dysplasia in inflammatory bowel disease . Low-grade dysplasia (LGD) is defined primarily by cytological abnormalities, including nuclear enlargement, crowding, stratification, hyperchromasia, mucin depletion, and mitoses in the upper portions of the crypts. These abnormalities extend to the mucosal surface. High-grade dysplasia (HGD) is characterized by even more pronounced cytological abnormalities, as well as glandular architectural abnormalities including villiform configuration of the surface, branching and lateral budding of the crypts, and formation of the so-called back-to-back glands. When there is any doubt as to the significance of histological abnormalities because of inflammation, ulceration, or histological processing artifacts, the findings may be classified as “indefinite for dysplasia” (IND) in order to prevent unnecessary clinical consequences.

Not all patients with BE progress to adenocarcinoma. The majority live their entire lives without undergoing malignant or neoplastic transformation. Others demonstrate a rapid progression to carcinoma and will die of esophageal cancer if it is not diagnosed and treated in a timely manner. Several recent attempts at identifying molecular markers that can predict which patients with BE will progress to esophageal cancer have not been proven effective in clinical trials. For example, anti-p53 antibodies have been shown to develop in patients with BE and adenocarcinoma, and may predate the clinical diagnosis of malignancy .

At the present time, the standard of care for surveillance of patients with BE remains debated. Although periodic endoscopic surveillance of patients with BE has been shown to detect carcinoma in its earlier stages, surveillance has significant limitations. Dysplastic and early carcinomatous lesions arising in nonnodular BE are not visible macroscopically; therefore, surveillance requires extensive random biopsies of the esophagus and histologic examination of the excised tissue for dysplasia. Random biopsy is prone to sampling error (missed dysplastic lesions) and significantly increases the cost and risk of surveillance. There is also significant interobserver disagreement between pathologists in diagnosing dysplasia. A large 10-year observational study in 409 patients with BE published in the British Medical Journal concluded that the current random biopsy endoscopic surveillance strategy has very limited value.

Optical imaging offers great promise for the detection and characterization of precancers in the esophagus. Endoscopic laser-induced fluorescence (LIF) spectroscopy based imaging is believed to measure the abnormal concentrations of certain endogenous fluorophores such as porphyrins in malignant tissue. Promising results for esophagus have been obtained by Vo-Dinh et al. , von Holstein et al. , Stepp et al. , Messmann et al. , Braichotte et al. , and Georgakoudi et al. . Another native contrast optical imaging approach, optical coherence tomography (OCT), is a method that provides two-dimensional cross-sectional images of the gastrointestinal tract. Like endoscopic ultrasound, OCT provides true anatomic images corresponding to the layers of the gastrointestinal tract , however, by using light the resolution of OCT is nearly 10-fold finer than that of high-frequency endoscopic ultrasound. Recently, Tearney et al. used high-speed OCT technology, termed optical frequency domain imaging (OFDI), to demonstrate the feasibility of a large area imaging of the entire distal esophagus in patients . Fujimoto et al. demonstrated the feasibility of using endoscopic ultra-high resolution optical coherence tomography (UHR OCT) to locating esophageal adenocarcinoma. Wax et al. developed an interferometric light scattering technique called angle-resolved low-coherence interferometry (a/LCI) capable of measuring nuclear morphology as a function of depth in epithelial tissue and demonstrated its ability to detect nuclear atypia in esophageal dysplastic epithelial tissues. Bigio et al. developed a minimally invasive diagnostic light scattering technique called elastic scattering spectroscopy (ESS) and applied it to detect early cancer in colon , bladder , and esophagus . The study in esophagus demonstrated that ESS has the potential to target conventional biopsies in Barrett’s surveillance saving significant endoscopist and pathologist time.

Over a decade ago, we pioneered a new diagnostic optical technique, biomedical light scattering spectroscopy (LSS), and demonstrated that LSS can measure subcellular nuclear morphology in esophageal tissue in vivo and be directly correlated with histopathology . Clinical feasibility of LSS for detection of epithelial precancers (dysplasia) in several organs was later demonstrated and then extended to large field-of-view imaging . Recently, we demonstrated that LSS can be employed to scan an entire esophagus during routine endoscopy, successfully guiding biopsy and detecting and mapping sites of invisible dysplasia missed by the current standard of care .

This article reviews the principles of LSS and describes application of endoscopic LSS for detecting invisible dysplasia in BE during routine clinical procedures.

7.2 Light Scattering from Cells

An electromagnetic wave, interacting with a dielectric or conducting particle (by a particle we will mean any bounded region with a complex refractive index different from the refractive index of the surrounding medium), induces oscillations of bound and free charges in that particle, which in turn generates electromagnetic waves inside and outside of the particle. Thus, the problem of light scattering by a single particle can be formulated in the following way: given a particle of known structure, illuminated by a plane wave of particular polarization, find the electromagnetic field inside and outside the particle. Usually, however, the simplified formulation of the problem suffices: find the approximate values of electromagnetic field at large distances from the particle.

For example, a well-known van de Hulst approximation enables one to obtain the wavelength and size-dependent scattering cross section for a large particle of an arbitrary shape

σs(λ,a)≈2πa2{1−sin2xx+(sinxx)2}σs(λ,a)≈2πa2{1−sin2xx+(sinxx)2}
σ s ( λ , a ) ≈ 2 π a 2 { 1 − sin 2 x x + ( sin x x ) 2 }

where

λλ
λ
is a wavelength,

aa
a
is a size of the particle,

mm
m
is the relative refractive index averaged over the volume of the particle, and

x=2π(m−1)a/λx=2π(m−1)a/λ
x = 2 π ( m − 1 ) a / λ
is the size parameter. The above expression demonstrates that scattering spectra of large particles are quite different from the monotonic

1/λ41/λ4
1 / λ 4
dependent Rayleigh scattering spectra of particles with sizes much smaller than a wavelength. Here the spectra are nonmonotonic functions exhibiting oscillations with the wavelength, with the frequency of these oscillations increasing with the particle size and its refractive index. This fact is quite important for understanding light scattering spectra of individual cells and cellular layers.

Though there are hundreds of cell types, the subcellular compartments in different cells are rather similar and are limited in number . Any cell is bounded by a membrane, a phospholipid bilayer approximately 10 nm in thickness. Two major cell compartments are the nucleus which has a size of 7–10 μm and the surrounding cytoplasm. The cytoplasm contains various other organelles and inclusions. One of the most common organelle (and the largest after the nucleus) is a mitochondrion which has the shape of a prolate spheroid. The large dimension of a mitochondrion may range from 1 to 5 μm and the diameter typically varies between 0.2 and 0.8 μm. Other smaller organelles include lysosomes which are 250–800 nm in size and of various shapes and peroxisomes which are 200 nm to 1.0 μm spheroidal bodies of lower densities than the lysosomes. Peroxisomes are more abundant in metabolically active cells such as hepatocytes where they are counted in the hundreds.

Since the size of the majority of cellular organelles is comparable to the wavelength of light and their refractive index ranges from 1.38 to 1.42 with the refractive index of the cytoplasm being within 1.34–1.36 range, the relative refractive index of the cellular organelles, m
m
, is close to unity. At the same time, the phase shift across the particle 4πa(m−1)/λ
4 π a ( m − 1 ) / λ
is small. The above two conditions mean that for the majority of small organelles the Rayleigh–Gans approximation is satisfied , which means that total intensity of light scattered by these small organelles increases with the increase of its refractive index as (m−1)2
( m − 1 ) 2
and with its size as a6
a 6
.

Unfortunately, the Rayleigh–Gans approximation cannot be applied to the cell nucleus, whose size is significantly larger than that of optical wavelengths. Here we can either use the intuitively transparent but approximate van de Hulst solution from Eq. (7.1) or cumbersome but exact Mie solution.

Sizes and refractive indices of major cellular and subcellular structures are presented in Fig. 7.1 . On this figure, we also provide the information on the relevant approximations that can be used to describe light scattering from these objects when interacting with light in the visible range.

7.3 Light Transport in Superficial Tissues

Light transport in biological tissue is dominated by elastic scattering. The primary scattering centers are thought to be the collagen fiber network of the extracellular matrix, the mitochondria, and other intracellular substructures, all with dimensions smaller than the optical wavelength. However, larger structures, such as cell nuclei, typically 5–15 μm in diameter, also scatter light. Because of the ubiquity of light scattering, its effects are enmeshed with those of absorption, making interpretation of tissue spectra difficult. On the other hand, this feature can actually enrich the information provided by spectroscopic techniques. For example, the strong dependence of the scattering cross section (7.1) on size and refractive index of the scatterer, such as the cell nucleus, as well as on the wavelength suggests that it should be possible to design a spectroscopic technique which can differentiate cellular tissues by the sizes of the nuclei. Indeed, the hollow organs of the body are lined with a thin, highly cellular surface layer of epithelial tissue, which is supported by underlying, relatively acellular connective tissue. There are four main types of epithelial tissue: squamous, cuboidal, columnar, and transitional, which can be found in different organs of the human body. Depending on type, the epithelium consists of either a single layer of cells or several cellular layers. Here, to make the treatment of the problem more apparent, we consider epithelial layers consisting of a single well-organized layer of cells, such as simple columnar epithelium or simple squamous epithelium. For example, in healthy columnar epithelial tissues, the epithelial cells often have an en-face diameter of 10–20 μm and height of 25 μm. In dysplastic epithelium, the cells proliferate and their nuclei enlarge and appear darker (hyperchromatic) when stained . LSS can be used to measure these changes. The details of the method have been published by Perelman et al. and will only be briefly summarized here.

Consider a beam of light incident on an epithelial layer of tissue. A portion of this light is backscattered from the epithelial nuclei, while the remainder is transmitted to deeper tissue layers, where it undergoes multiple scattering and becomes randomized before returning to the surface. Epithelial nuclei can be treated as spheroidal Mie scatters with a refractive index which is higher than that of the surrounding cytoplasm . Normal nuclei have a characteristic size of 4–7 μm. In contrast, the size of dysplastic nuclei varies widely and can be as large as 20 μm, occupying almost the entire cell volume. In the visible range, where the wavelength is much smaller than the size of the nuclei, the van de Hulst approximation (7.1) can be used to describe the elastic scattering cross section of the nuclei. Eq. (7.1) reveals a component of the scattering cross section which varies periodically with inverse wavelength. This, in turn, gives rise to a periodic component in tissue reflectance. Since the frequency of this variation (in the inverse wavelength space) is proportional to the particle size, the nuclear size distribution can be obtained from that periodic component.

However, single scattering events cannot be measured directly in biological tissue. Because of multiple scattering, information about tissue scatterers is randomized as light propagates into the tissue, typically over one effective scattering length (0.5–1 mm, depending on the wavelength). Nevertheless, the light in the thin layer at the tissue surface is not completely randomized. In this thin region, the details of the elastic scattering process are preserved. The total signal reflected from tissue can be divided into two parts: single backscattering from the uppermost tissue structures such as cell nuclei and a background of diffusely scattered light. To analyze the single scattering component of the reflected light, the diffusive background must be removed. This can be achieved either by modeling using diffuse reflectance spectroscopy or by other techniques such as polarization background subtraction and coherence gating method .

There are several techniques that can be employed to obtain the nuclear size distribution from the remaining single scattering component of the backreflected light, which can be called the LSS spectrum. A good approximation for the nuclear size distribution can be obtained from the Fourier transform of the periodic component as described in Ref. . A more advanced technique introduced by Fang et al. and described here is based on linear least squares with a nonnegativity constraints algorithm.

The experimentally measured reflectance spectrum consists of a large diffusive background plus the component of forward scattered and backscattered light from the nuclei in the epithelial layer. For a thin slab of epithelial tissue containing nuclei with size distribution (number of nuclei per unit area (mm ²) and per unit interval of nuclear diameter (μm), the approximate solution of the transport equation for the backscattered component is a linear combination of the backscattering spectra of the nuclei of different sizes: