Tethered Capsule Endoscopy, A Low-Cost and High-Performance Alternative Technology for the Screening of Esophageal Cancer and Barrett's Esophagus

1032 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 55, NO. 3, MARCH 2008

Tethered Capsule Endoscopy, A Low-Cost andHigh-Performance Alternative Technology for

the Screening of Esophageal Cancerand Barrett’s Esophagus

Eric J. Seibel�, Member, IEEE, Robert E. Carroll, Jason A. Dominitz, Richard S. Johnston, C. David Melville,Cameron M. Lee, Steven M. Seitz, Senior Member, IEEE, and Michael B. Kimmey

Abstract—Esophageal cancer is currently the fastest growingcancer in the United States. To help combat the recent rise in mor-bidity, our laboratory has developed a low-cost tethered capsuleendoscope system (TCE) aimed at improving early detection ofesophageal cancer. The TCE contains a resonant fiberoptic laserscanner (1.6 mm O.D.) which fits into 6.4-mm easy-to-swallowcapsule at the distal tip. The tethered portion contains a singlemode optical fiber multiplexed to three laser diodes at the prox-imal end. This design offers two main advantages over currentendoscope technology. First, because of its small size, the TCE canbe swallowed with minimal patient discomfort, thereby obviatingsedation. Second, by imaging via directed laser light, the TCE isstrategically positioned to employ several burgeoning laser-baseddiagnostic technologies, such as narrow-band, hyperspectral,and fluorescence imaging. It is believed that the combinationof such imaging techniques with novel biomarkers of dysplasiawill greatly assist in identifying precancerous conditions suchas Barrett’s esophagus (BE). As the probe is swallowed, thefiber scanner captures high resolution, wide-field color images ofthe gastroesophageal junction (500 lines at 0.05-mm resolution)currently at 15-Hz frame rate. Video images are recorded as thecapsule is slowly retracted by its tether. Accompanying softwaregenerates panoramic images from the video output by mosaicingindividual frames to aid in pattern recognition. This initial reportdescribes the rationale for the unique TCE system design, resultsfrom preliminary testing in vitro and in vivo, and discussion on the

Manuscript received February 3, 2007; revised June 21, 2007. This workwas supported in part by the National Institutes of Health (USA) under GrantR21CA110184 and by the PENTAX Corporation, Tokyo, Japan. Asterisk indi-cates corresponding author.

*E. J. Seibel is with the Department of Mechanical Engineering, Universityof Washington, Human Photonics Lab, Box 352600, Seattle, WA 98195 USA(e-mail: [email protected]).

R. E. Carroll was with the University of Washington, Seattle, WA 98195 USA.He is now with the Department of Electrical Engineering and Computer Sci-ence, University of California, Berkeley, CA 94720 USA (e-mail: [email protected]).

J. A. Dominitz is with the School of Medicine, University of Washington,VA Puget Sound Health Care System, Seattle, WA 98195 USA (e-mail: [email protected]).

R. S. Johnston, C. David Melville, and C. M. Lee are with the Universityof Washington, Seattle, WA 98195 USA (e-mail: [email protected];[email protected]; [email protected]).

S. M. Seitz is with the Department of Computer Science and Engineering,University of Washington, Seattle, WA 98195 USA (e-mail: [email protected]).

M. B. Kimmey is with the Department of Medicine, University of Wash-ington, Seattle, WA 98195 USA (e-mail: [email protected]).

Color versions of one or more of the figures in this paper are available onlineat http://ieeexplore.ieee.org.

Digital Object Identifier 10.1109/TBME.2008.915680

merits of this new platform technology as a basis for developing alow-cost screening program for esophageal cancer.

Index Terms—Biomedical image processing, biomedicalimaging, cancer, medical diagnosis, optical imaging.

I. INTRODUCTION

THE INCIDENCE of esophageal adenocarcinoma is risingfaster than for any cancer in the USA [1], and is the sixth

leading cause of death from cancer in men [2]. Esophagealadenocarcinoma is believed to arise from a condition known asBarrett’s esophagus (BE) in which the esophageal epithelium ismarked by abnormal intestinal-type cell growth, also believedto result from chronic gastroesophageal reflux disease (GERD)[3], [4]. Though it is unknown whether BE is a necessary pre-cursor to all cases of esophageal cancer, it is a well documentedand clearly recognized risk factor for esophageal adenocarci-noma [5]. BE mucosa tissue appears salmon pink in color, incontrast to the normal pearly white squamous mucosa of theesophagus (see Fig. 1). Although screening for esophagealcancer is not deemed appropriate for the general population,periodic examination of patients with BE is recommended inorder to identify dysplasia or cancer at an earlier and moretreatable stage [4]. While standard endoscopy and tissue biopsyare sufficient for the monitoring of patients diagnosed with BE,95% of esophageal adenocarcinoma develops in patients withpreviously undiagnosed BE, proving that current endoscopicscreening efforts are inadequate [6], [7].

When considering a screening strategy for such a conditionas BE, it is important to consider several factors: diseaseprogression, availability of screening resources, performanceof a particular screening test (sensitivity and specificity), ac-cessibility to treatment, the willingness of patients to undergoscreening, and the associated cost. BE is a fairly common con-dition among patients having the symptom of heartburn, withan estimated prevalence ranging from 6%–12% [8]. Currently,screening is performed using a standard GI endoscope on asedated patient to examine and biopsy any abnormal appearingmucosa. The endoscopist’s assessment of the presence of BEhas been shown to have a sensitivity and specificity of 82%(95% CI, 72–92) and 81% (95% CI, 78%–84%), respectively,

0018-9294/$25.00 © 2008 IEEE

SEIBEL et al.: TETHERED CAPSULE ENDOSCOPY 1033

Fig. 1. BE illustrated from a standard endoscopic image.

when compared to pathologic confirmation [9]. The additionaluse of topically-applied dyes for chromoendoscopy, expandedmagnification, and separate analysis from narrow-band excita-tion may improve the sensitivity and specificity for BE, thoughtheir clinical utility is currently unproven [10], [11]. Once diag-nosed, BE is treated by reducing the symptoms of GERD usingpharmaceuticals and/or surgery [8] with new highly successfultherapies being developed specifically for BE [12]. While thereare no randomized studies demonstrating that screening andsurveillance improve BE patient outcomes, retrospective cohortstudies suggest that BE patients undergoing surveillance havesignificantly improved survival compared to controls [13].

In a physician survey, 62% indicated that if unsedated en-doscopy was made available to primary care physicians in anoffice setting an increase in BE screening would result [14].Unsedated endoscopy using a thinner endoscope is an alterna-tive to standard endoscopy, but is not commonly used in theUSA, possibly due to patient lack of acceptance of the commontransnasal approach [15], [16]. Finally, there is ongoing researchaimed at finding biomarkers that identify esophageal adenocar-cinoma in its precancerous and neoplastic stages, since it is be-lieved that genetic changes precede any morphological changesfound during histologic analysis. However, at present, there is nosingle biomarker available for which a negative indicator test re-sult would warrant discontinued screening of a patient [8], [19].

Ideally, a new screening test for BE would be as sensitiveand specific as standard endoscopy, but would not require se-dation and would have low risk and low cost. The current costsfor standard endoscopy per the Center for Medicaid and Medi-care Services (CMMS) is approximately $600, excluding biop-sies. The CMMS cost for esophageal capsule endoscopy is evenhigher at approximately $750 (the single-use capsule itself costs

approximately $400). Nevertheless, screening and monitoringwith standard endoscopy followed by esophagectomy for sur-gical candidates with high-grade dysplasia or cancer, or endo-scopic therapy for cancer patients who were not operative candi-dates has been reported to be cost-effective. The price tag is cur-rently $12 140 per life-year gained compared with no screening[20].

Wireless capsule endoscopy or “pill” endoscopy is a recentalternative to standard endscopy which uses a modified cap-sule containing two cameras, battery source, and wireless trans-mitter for sending images to an external digital recorder [17],[18]. However, untethered capsule endoscopy is limited in thatit yields random views of the esophagus, produces images at asub-video frame rates ( 2 per second), and increases the overallcost of diagnosis. In contrast, a tethered capsule affords directcontrol over the camera view by the endoscopist, images at nearvideo frame rates, while reducing overall cost. For these rea-sons, it is believed that TCE is ultimately more advantageousthan other BE screening methods.

Our approach is to introduce a new low-cost device specifi-cally for BE screening based on a completely new type of endo-scope imaging technology. Instead of using passive illuminationand a camera sensor array for image capture, a single opticalfiber is used to scan a surface using laser illumination, whilethe backscattered light is recorded one pixel at a time to forma non-confocal image. Unlike fiber scanners developed for op-tical coherence tomography (OCT) that vibrate along a singleaxis at a fixed amplitude, this fiber scanner moves in two di-mensions while being amplitude modulated to form a spiral pat-tern [21]. The fiber scanner and lenses are housed within thecapsule while a single optical fiber for illumination, scannerdrive lines and six return plastic optical fibers are containedin the tether, forming a tethered-capsule endoscope (TCE) (seeFig. 2). Similar to standard endoscopy, the base station con-tains the light sources as well as the optical detectors and soft-ware to provide a machine vision software tool for clinicians.In order to judge short ( 3 cm) versus long segment BE, theclinician must measure the extent of suspected BE above thetop of the gastric folds. A mosaic of the entire esophagus isautomatically generated from the TCE images to aid the clin-ician in visualizing the extent of BE and likely sites for futurebiopsies. Virtual colonoscopy using computed tomography andthe new mosaic panorama perspective allows the radiologist toread patient’s data significantly faster than the conventional vir-tual colonoscopy perspective without detriment to detection rate[22]. Thus, both the laser-scanning imaging from a TCE probeand the application of integrated mosaicing software are ad-vanced technologies for screening and surveillance of neoplasia.

II. TCE DESIGN

The TCE is essentially a repackaged scanning fiber endo-scope (SFE) which has been invented and developed at the Uni-versity of Washington [23], [24]. Similar to the SFE, the TCEuses the same singlemode fiber scanner to scan a laser spot overtissue and also uses multimode optical fibers to capture andrecord backscatter light signals.


Fig. 2. TCE probe used for animal experiments.

Fig. 3. SFE distal end, 1.6 mm overall diameter.

The fiber scanner consists of a 420- m diameter piezoelec-tric tube through which a 4.3-mm cantilevered length of single-mode optical fiber (Nufern 460-HP) is affixed (see Fig. 3). Thepiezoelectric tube (PZT 5A) is plated with quadrant metal elec-trodes and powered by five 50 gauge wires with a limit of lessthan 30 V for operation. The piezoelectric tube, singlemodeoptical fiber, and lens system are contained in a stainless steeltube with a 1.1-mm outer diameter and 13-mm length.

The TCE was created by repackaging the fiber scanner in amedical grade plastic capsule form to aid in swallowing. Thecapsule dimensions (6.35 18 mm) are those of a standard No.2 capsule (Torpac Inc., Fairfield, NJ) and were chosen overlarger and smaller sizes for ease of swallowing and ability tohandle [25], [26]. Six 250- m diameter multimode optical fibers(ESKA, Edmund Optics) are directed to the face of the capsuleto collect backscattered light signal. The wires and optical fibersare routed back from the capsule to the base station througha thin flexible polyethylene tether with 1.4-mm diameter. TheTCE used for testing was designed to meet the following spec-ifications (see Table I).

The base station provides laser light, generates drive signals,and detects return light (see Fig. 4). Inside the base station arethree lasers: red 635 nm (FiberMax, Blue Sky Research), green532 nm (Chromalase, Blue Sky Research), and blue 444 nm

(Nichia Laser Diode and OZ Optics). The three lasers arefiber coupled and combined into one singlemode fiber using afiberoptic combiner (SIFAM RGB Combiner 40W004–001).Return light is separated into red, green, and blue using dichroicbeamsplitters and detected using a photomultiplier tube (PMT)for each color.

During operation RGB light from the laser combiner is cou-pled in the core of the singlemode optical fiber in the TCE. Thepiezoelectric tube is driven with an amplitude modulated sinewave with a 90 phase difference in the two drive axes to gen-erate a spiral scanning pattern. The frequency of the sine waveis tuned to the scanning fiber’s first mode of mechanical res-onance (currently 5 kHz). Light emitted from the end of thesinglemode scanning fiber passes through the lens system andis focused onto the tissue. This two-lens system was manufac-tured by PENTAX Corp., Tokyo, Japan, for long working dis-tance operation, having almost collimated output at 0.02 numer-ical aperture (NA). Light reflected off the tissue is collected bythe multimode optical fibers, routed through the color separationsystem and detected by the photomultiplier tubes. Two customhardware boards, containing a central field-programmable gatearray (FPGA), and five memory banks, generate and processsignals to construct the final TCE images. In addition to stan-dard RGB imaging several different imaging modes have been


TABLE ITETHERED CAPSULE ENDOSCOPE DESIGN SPECIFICATIONS

demonstrated with this technology including: florescence, po-larization contrast, and sequential color.

III. TCE OPERATION

The TCE system is operated by plugging the desired TCEprobe into the base station then powering up the PC-basedhost computer and the base station. Plugged into the PC aretwo custom PCI electronics cards. The cards each containeight analog-to-digital (A/D) converters, eight digital-to-analog(D/A) converters, five banks of parallel accessible SRAM, anda two million gate FPGA (Xilinx Spartan 3). The cards areidentical except for the programming within the FPGA. Onecard drives the resonant scanner and is used during systemcalibration. The second card controls the laser power, con-structs images from PMT data, and drives the image display.Control of the TCE system is through a LabVIEW (version8.01 National Instruments, Inc.) software interface runningon the PC. After opening the control program the user entersthe serial number of the TCE probe being used. The serialnumber identifies files stored in computer memory that containdata relating to the operation of that probe such as probe type,resonant frequency, and drive parameters to achieve the desiredfield-of-view for the application. Once the probe parametersare loaded (approximately 5 s) the probe enters imaging modeand can be used.

During imaging the TCE user interface allows the user tocapture and store single image frames or a sequence of framesforming a movie. The captured movies can be used by the mo-saicing software to create a panoramic image of the esophagus.

Additional controls allow image zooming (performed by drivingthe resonant fiber in a smaller field-of-view) and laser powercontrol. If the user desires to change performance parametersthe device can be recalibrated or rewhite balanced by placingthe probe in the calibration or white balance ports and selectingthe desired function from the user interface.

IV. TCE TESTING

In vitro testing was done to confirm image color balanceusing color charts (Gretag MacBeth Mini Color Checker,Edmund Optics), and field of view and resolution using USAF1951 photopaper resolution test target (Edmund Optics) inboth air and water. An electrical safety test was also conductedand confirmed in vitro. The safety testing protocol consistedof turning on the TCE instrument, immersing the capsule, andthe tether in a 200-mL glass beaker filled with physiologicalbuffered saline (over-the-counter NeilMed Sinus Rinse, 8 ozor 235 mL) and placing a stainless steel conducting electrodeat least 1 cm away from the probe. Imaging of test targetsplaced under the beaker commenced while current from theelectrode to ground was measured using a precision multimeter(Tenma model 72–2050). No leakage current was detected atthe detector’s 200-nA noise limit. If any measurable leakagecurrent above the noise floor of 0.2 A was detected any in vivotesting would be canceled.

In vivo testing using a porcine model was conducted at theUniversity of Washington in accordance with approved proto-cols for animal welfare. A young pig (20 kg) was fasted for 8 h,anesthetized, intubated, and placed on artificial ventilation in the


Fig. 4. SFE/TCE base station block diagram.

supine position. The esophagus–stomach junction was initiallyobserved and measured using a forward viewing flexible bron-choscope (PENTAX EB-1970K), where large amounts of bilewere observed. A suction tube was inserted to remove most ofthe bile before inserting the TCE probe. Since the animal wasanesthetized and could not be induced to swallow, a capsule in-troducer was devised. This consisted of a flexible tube with aside slit and a custom plastic saddle at the distal tip for holdingthe TCE capsule. After insertion into the pig stomach (verifiedby imaging), a wire was used to push forward and release theTCE capsule from the saddle. The insertion tube was withdrawnabout 10 cm leaving the TCE probe within the upper stomach.Together, the TCE tether and insertion tube were slowly pulledout of the pig, while TCE video images were recorded at thebase station.

A second TCE probe was fabricated and tested for leakagecurrent, cleaned with alcohol, and swallowed by a human volun-teer in the sitting position. After a few sips of water the TCE en-tered the stomach where it was slowly pulled back into the upperesophagus while recording video images. In total the testingtook about 10 min for several repeated swallow and removaliterations.

V. MOSAIC IMAGE FORMATION

A. Mosaic Software Design and Theory

To create a representation of the esophageal surface using anendoscopy video sequence two basic elements are required: asurface model of the esophagus and a camera pose estimation

for each video frame. With this knowledge it is possible toproject each frame onto the model to texture-map its surface.The texture-mapped model must then be transformed into aflat image. The surface is modelled as a cylinder, because itis generally the shape of an esophagus and because it can beeasily displayed as a 2-D image. To estimate camera motionrestricted affine transformations are computed between pairsof frames, allowing only rotations, scales, and translations.To compensate for illumination changes in the scene it isnecessary to first “neighborhood-normalize” each frame beforethe alignment is done. The resulting affine transformations areused to estimate the amount of forward motion between pairsof frames. The remaining components of the transformation,after factoring out the scale, are used to compensate for othermotion, e.g., camera rotation and non-axial camera motion.The cylinder’s vanishing point in each frame is located bydetecting a darkest spot. This information is used to determinethe camera’s angular orientation. Assuming the camera travelsalong the cylinder’s central axis, there is enough informationto project each frame onto the cylinder. From each projectedframe a strip of width corresponding to the amount of forwardmotion is extracted. These strips, concatenated together, con-stitute the texture mapped cylinder, and unwrapped it is themosaic panoramic image. As a final step to compensate for anyseaming artifacts, gradient domain blending is used.

This mosaicing technique is most closely related to that ofRousso et al. who introduced the idea of a pipe projection [27].Pipe projection allows sequences exhibiting forward motion tobe mosaiced by transforming radial optical flow into paralleloptical flow in the projected image.


Fig. 5. Pipe projection diagram. Elliptical strip centered around the vanishing point V is projected onto the pipe and added to the projections from previous frames.

The specific nature of esophageal mosaicing problem re-quires some additional assumptions about the scene. The pipeis used as an approximation for the scene structure, insteadof as a manifold that correctly transforms optical flow. In thepipe projection, as described in [27], the camera is located onthe pipe’s central axis. Although this is not guaranteed in theesophagus, the assumption of a centered camera is maintainedfor simplicity. Because of the possibly unsteady motion ofthe camera, the focus of expansion is not a reliable indicatorof the pipe’s central axis. Instead, the center of the pipe isapproximated by simply finding the dark region of the image.The centroid of all pixels under some threshold is a simple butreliable approximation for the vanishing point of the esophagus.

B. Pipe Projection and Strip Selection

Let be the vanishing point in the image plane,where is the focal length. Then, defines the centralaxis of the pipe, if the camera’s optical center is placed at theorigin (see Fig. 5). With the relative orientation of the pipe andimage plane known, it is possible to relate image coordinatesand mosaic coordinates using a perspective projection [27]. Thestrips used to make the mosaic are determined by a scanningbroom (see Fig. 6). Any line selected in the video frame willsweep over the scene as the video is played. The shape of thisbroom depends on the motion in the scene, ideally being perpen-dicular to the optical flow. In the case of forward motion, thisis an ellipse centered around the focus of expansion [27]. Thestrips are defined implicitly based on knowledge of a camera po-sition (depth in the pipe) for each frame. An elliptical scanlinein the frame is defined implicity by selecting a distance downthe pipe from a given frame’s center of projection. After deter-mining the change in depth for a particular frame, this value isadded the chosen distance to define another line in the frame.The area between these two lines is the strip to be added to themosaic. Occasionally the elliptical strip will stray outside of theframe boundaries. This issue is addressed by simply leaving thecorresponding areas of the mosaic to be filled in by subsequentframes. These areas are guaranteed to come into view becausethe camera is moving backward. This has the effect of the sam-pling strip hugging the edge of the frame when the strip wouldotherwise extend out of frame.

Fig. 6. Standard endoscopic images of the esophagus, single frame (top left).The darkened strip taken from this initial video frame is highlighted in the inputand the mosaic. Pipe projection transforms the elliptical strips in the input se-quence into vertical strips for mosaicing. The central yellow dot demarks the“dark spot” which is considered the pipe’s vanishing point. The divot in the mo-saic is a result of the scanline snapping to the input image boundary.

C. Motion Tracking

The mosaic is constructed by stacking up strips taken fromthe pipe-projected video frames. The width of the strips dependson the amount of camera motion between consecutive video


frames. To estimate the amount of motion between frames, con-secutive frames are registered using a constrained affine trans-formation that allows for uniform scales, rotations, and trans-lations (no sheers). This allows a uniform scale to be extractedthat can be used to estimate the amount of camera translation,and to correct for camera rotations. It is a four-parameter trans-formation defined as

(1)

where and are correspondent homogeneous coordinates.Finding solutions to this type of transformation is a well studiedproblem [28].

Given two images and , the four-parametertransformation that relates them needs to be estimated. Eachpixel in has an associated coordinate in such that

, where is the displacement vector. Ap-plying a first-order Taylor expansion gives

(2)

From (1)

Combining (1) and (2)

Each pixel produces one equation for each color channel. Thesystem is over-constrained and can be solved using least squarestechniques.

To obtain an accurate result, the parameters are updated iter-atively. The previous iteration’s result is used to create the warp

from . Then an update transformation between andis computed. The update transformation can be combined withthe current result with a simple matrix multiplication. The reg-istration is also run on a course to fine basis using a Gaussianpyramid, allowing it to handle large motions.

Because of the dynamic nature of the scene, it is unlikely thata single transformation will give a good result across the entireframe. However, the approximate region where the strip willbe taken is known. The registration only needs to be accuratealong the strip boundaries, so the computation can be limited toonly pixels in the neighborhood of the scanline. A ring centeredaround the dark spot of the image is a good approximation.

Once the transformation is computed it is trivial to extractthe scale factor. The change in depth in the pipe is estimated as

, where is the focal length, is the radiusof the pipe, is the scale factor, and is the distance from thescanline to the pipe’s central axis (dark spot).

D. Neighborhood Normalization and Blending

The alignment method, described in [29], relies on a fewassumptions, notably; constant lighting, small motions, andsmooth gradients. The small motion issue is addressed withcourse-to-fine alignment and the smoothness problem caneasily be fixed by blurring the images. Constant illuminationis an issue because the light source is on the camera itself, solighting changes as the camera moves. This issue is addressedusing neighborhood normalization. The mean intensity andstandard deviation are computed for a small window aroundeach pixel. By subtracting the mean from the pixel value anddividing by the deviance, some measure of a point’s actualcolor independent of the lighting conditions is obtained. Theresulting image can then be used for the pairwise alignment,satisfying the color constancy assumption.

Imperfect alignment and changing illumination result in no-ticeable seams along the strip boundaries. It is preferable to min-imize these seams without removing any details from the image.A simple blending approach like feathering usually requires alarge overlap with good registration to avoid ghosting, but inour case the registration is only likely to be good in a small re-gion along the seam. Instead, we use gradient domain blending[29].

Instead of accumulating strips of pixel color values, the colorgradients are accumulated. The gradients themselves can beblended with feathering over small area. The result is a gradientvector field for each color channel. It is then possible to solve foran image that has the associated gradient field. Since there arehave two equations per pixel, it is an over-constrained problemand a best-fit solution must be found. As in [28], each pixelgives two equations of the formand per color channel, where

and are known. Arranging into a large vector andinto a vector , gives the matrix equation

where is a sparse matrix containing two rows for each pixel(minus the boundary cases). As in [27], a least-squares fit canbe found by multiplying both sides by . The matrixgives the Laplacian of an image when represented in vectorform, so in essence the image is derived from its associatedLaplacian.

VI. RESULTS

The completed TCE system (base station and probe) meetsall design criteria listed in Table I and is shown in Figs. 2 and 7(top left). During TCE operation, the measured total laser powerat maximum power setting is 1.5 mW (B-442 nm), 2.3 mW(G-532 nm), and 3.6 mW (R-635 nm) using optical power meterand probe (Newport 1830-C and 818-ST). Since there is min-imal light loss after coupling to the singlemode optical fiberused for scanning, the total optical efficiency is expected to begreater than 10%. In comparison to standard video endoscopesand bronchoscopes, the maximum TCE optical power is 3 lessthan midrange illumination, and 40 less than full-power illu-mination when measurements are made at 532 nm responsivity


Fig. 7. TCE system (top left) being used to generate mosaic image of rolled up map, see result of 64 images stitched into mosaic (right middle). Single frameimages of test target showing spatial resolution of the 0.049 mm bar width from target 3–3 (bottom left) and color discrimination (bottom right).

of the silicon sensor. In vitro imaging of flat test targets (e.g.,Gretag Macbeth Mini Color Chart and Edmund Optics USAF1951 test target) demonstrates the high color saturation and spa-tial resolution. Target number 3–3 has a 49.0 m bar widthwhich can be resolved in the peripheral field of Fig. 7 (bottomleft). The TCE probe for animal testing measured just over 100field of view, while the TCE probe for human testing measuredjust below 100 using less than 20 V within the capsule. Amaximum field of view was recorded in vitro at 118 . Whenplaced within a 1-in tube of rolled paper and pulled slowly (2 mm/s) while remaining roughly centered in central axis of thelumen, mosaic images are generated with noticeable but accept-able levels of distortion for the purpose of identifying and map-ping regions of color variation (see Fig. 7). The TCE systemimages appear similar in both air and water mediums, and nomeasurable leakage current was detected while imaging in phys-iological buffer.

TCE testing within the live pig revealed images of the loweresophagus to mouth with a mosaic image of the upper esoph-agus (see Fig. 8). Bright yellow-green bile was present in thepig stomach and particles of bile-coated food appeared on theesophagus walls during imaging of the pig in the supine po-sition. Suction applied to a secondary tube alongside the TCEremoved much of the obscuring bile. In a sitting position, thehuman volunteer easily swallowed the TCE probe using onlysips of water and several swallows. The TCE probe revealeda clear image of the gastric folds and the important squamo-columnar junction where the stomach mucosa (red) transitionsto the esophageal mucosa (light pink) (see Fig. 9). In human

Fig. 8. Mosaic image of pig esophagus created from a 32-frame movie cap-tured during pulling of the TCE probe and individual frame numbers 8 (left), 14(middle), and 24 (right).

in vivo images, the red TCE illumination was reduced from the


Fig. 9. In vivo images from a human volunteer. Left is from stomach, middle is squamo-columnar gastroesophageal junction where BE is measured, and right isfrom mid-esophagus.

maximum in order to match the expected hues per the recom-mendations of two observing gastroenterologists. To compen-sate for darker imaging in vivo versus in vitro, TCE imagesshown in Fig. 9 were increased in brightness and contrast by10%–20% using PhotoShop (Adobe). Occasionally bubbles ob-scured the esophagus walls. The bubbles were removed from thefield of view by draining the residual water by swallowing or byadding additional water.

VII. DISCUSSION

TCE swallowability and imaging performance met all ex-pectations in this first-generation prototype. In one motivatedvolunteer, the capsule was easily swallowed with no side ef-fects. However, a capsule weighing approximately 150% moreis expected to aid in more rapid peristaltic movement into thestomach. A future study is planned for optimizing capsule sizeand weight and clinical procedures that may use simethicone toreduce bubbles. Because most of the capsule is empty space,adding weight is a minor modification. When recording videofor the mosaicing feature, the normal 15 Hz viewing frame ratewas reduced to less than 5 Hz. Nonetheless, the mosaic algo-rithm successfully captures the esophageal surface. Most of theseaming artifacts occur when the camera changes direction orpauses for an extended period of time, but are undetectable aftergradient domain blending. Lighting inconsistencies in the inputimage cause artifacts in the mosaic, especially noticeable whenthe sampling strip covers a specular reflection. Color consis-tency within the mosaic will improve as the automatic gain con-trol is made more sophisticated using gamma correction.

As a foundational technology for BE screening, the scanningfiber endoscope technology and TCE prototype must be com-pared to two very different approaches. Designed and testedspecifically for BE screening, a simple yet novel colorimetry de-vice was developed to map color along the esophagus by Dat-tamajumdar et al. [30]. By using an axially-symmetric mirrorpulled up from the stomach, a 2.9-mm ring of white light illu-mination was used to map color circumferentially with a resolu-tion determined from having 30 detection optical fibers. A 2-Dcolor map of the lower esophagus is generated at 1–2 mm of

spatial resolution with occasional patches of information loss.Although performance of this colorimetry technique proved tobe 90% sensitive with standard endoscopy for long segment BE( 3-cm length) in human subjects, the advantages (low cost)were outweighed by the disadvantages. The probe size was 6mm in diameter. A local anesthetic was needed to reduce thegag reflex as the probe shaft was just little more flexible than astandard flexible endoscope of 13 mm in diameter. Furthermore,the spatial resolution and sensitivity were not high enough to ef-fectively screen for short segment BE and tongues of BE alongthe esophagus which would severely limit usefulness of such aBE screening program.

In the highly successful approach by Ramirez et al. [31]the wireless capsule endoscope was modified by tying a stringaround the 11-mm diameter by 26-mm-long capsule. Thecapsule used for this human preclinical study was developedfor imaging the small bowel, containing only one camerawith a longer battery life and a lower image acquisition rate.Although the capsule is designated as a disposable, single-usedevice, these investigators were able to use a single capsulefor 24 esophageal exams, reducing the per patient screeningcost. By having each patient swallow the capsule three timesper procedure, the sensitivity of detecting BE was 100% for 50patients (28 short-segment and 22 long-segment BE). Of theseunsedated patients, 92% preferred string capsule endoscopyover standard endoscopy, whereas one patient was droppedfrom this study due to not being able to swallow the large-sizedcapsule. In comparison to the string capsule, the TCE capsule isalmost half the diameter at 6.35 mm and could be made smallerif necessary. However, the TCE tether is larger at 1.4 mm com-pared to the disposable string estimated to be about 0.4-mmdiameter by Ramirez [31]. A much small diameter TCE tetheris anticipated in a commercial device with RGB image sensorsplaced within the capsule, replacing the six plastic opticalfibers with much smaller diameter electrical wires. In a costcomparison, the TCE probe is expected to be cost competitivewith the camera-based capsule as all components are very lowin cost. Although the TCE probe has been designed for hospitalcleaning, the ability of the TCE probe to undergo sterilization


and/or high level disinfection to allow reusability has not beenstudied.

The TCE prototype has one major difference from all camera-based capsule endoscopes, the versatility of adding advancedimaging features and laser diagnostics while not affecting thesize or cost of the TCE probe. By electronically adjusting thescan amplitude, magnification endoscopy can be added as a fea-ture [21]. By selecting individual laser illuminations narrow-band imaging within the visible spectrum can be displayed con-currently with combined RGB imaging. Because each laser re-flectance map is generated individually, a post-processing al-gorithm can be used to enhance color differences within themosaic image beyond the visible spectrum, using light sourcesacross the ultraviolet to infrared spectrum. By turning off spe-cific laser illuminations and filtering out the higher incident il-lumination, fluorescence imaging can be an additional feature.Recently, the combination of two advanced imaging techniques,autofluorescence and narrow-band, combined with reflectanceimaging of BE, has been demonstrated to improve the sensi-tivity and specificity of detecting neoplasia compared to stan-dard endoscopy [32]. If detection of dysplasia in BE is the goal,additional information may be required which may go beyondthe capabilities of the TCE. Since there is no working channel,a separate spray catheter may be necessary to cover the mucosaof the distal esophagus with indigo carmine dye solution for en-hancing the visualization of topographical patterns that corre-late with high-grade dysplasia [33]. If elastic or light scatteringspectroscopy is desired, then illumination with only three laserbands would be insufficient. However, broadband illumination(350–900 nm) could be delivered to the esophageal mucosa byadding an additional multimode optical fiber that terminates onthe side of the capsule. A second multimode optical fiber canbe placed at a fixed separation distance (e.g., 0.35 mm) to col-lect the point measurements of elastic scattering spectra thatare a measure of dysplastic tissue features in BE [34]. Thus,the TCE with minor modifications can be considered a plat-form technology on which various combinations of multimodalimaging and diagnosis can be delivered to the upper GI tract.While the authors acknowledge the risk of information overloadwith the implementation of multimodal endoscopic imaging, thedoor is opening to doctor assistance from computer-aided pat-tern recognition and diagnosis.

The clinical value and specific role of the TCE image mo-saicing feature has yet to be determined. Possible uses are: 1) acolor printout of the endoscopy for patient-doctor counseling;2) a scaled mapping of the regions of BE to more rapidly as-sist in determining between long segments, short segments, andtongues of BE; 3) a single fused image that combines the re-sults from multiple TCE mosaics from multiple swallowings toreduce ambiguity from a single imaging pass; 4) the ability tomap regions of non-visible multimodal image data overlaid inpseudocolor and possibly select biopsy sites; 5) the ability toadd quantitative optical biopsy measures based on laser-inducedfluorescence and spectroscopies; and 6) a visual record of thepatient’s medical history which also combines multiple sensordata such as pH and sphincter pressure. Currently, the mosaicimage is generated with less than 5 minutes of post-processingof the TCE images, while real-time mosaicing is expected by

using graphic processor chips in the future. To accurately scalethe mosaic image to esophageal position a tether position sensoris needed, as previously developed for the BE colorimetry probe[30]. Future work will be calibrating the mosaic image to com-pensate for variations in lighting and color measurement so pat-tern recognition and computer-aided diagnosis algorithms canbe generated and tested. There is a growing need for diseasescreening programs in developing countries that rely on com-puter-aided diagnosis with low-cost imaging scopes using easyto follow clinical procedures on unsedated patients.

VIII. CONCLUSION

Preliminary results are encouraging that there is a new tech-nology solution for screening and surveillance for BE. Future di-rections include refining the protocol for image acquisition anddetermination of the sensitivity and specificity of the tetheredcapsule endoscopy compared to standard endoscopic imaging.

ACKNOWLEDGMENT

The first author, E. J. Seibel, would like to thank Mr. T. Hi-daka, Senior Fellow, PENTAX Corporation for his steadfastmanagerial support throughout this entire project, and the NCIIndustry-Academic Partnership with Mr. T. Hidaka, as coinves-tigator (R21 CA110184). Animal testing expertise was providedby S. Bernard and W. Lamm. Technical expertise in fabricatingthe TCE system was assisted by research engineers B. Murray, J.Crossman-Bosworth, and R. C. Bryant. Understanding the fea-tures for diagnosing dysplasia in BE was facilitated by Dr. M.Upton.

REFERENCES

[1] S. S. Devesa, W. J. Blot, and J. F. Fraumeni, Jr, “Changing patternsin the incidence of esophageal and gastric carcinoma in the UnitedStates,” Cancer, vol. 83, no. 10, pp. 2049–2053, 1998.

[2] American Cancer Society, Atlanta, GA, “Cancer facts and figures,”2006 [Online]. Available: www.cancer.org

[3] A. J. Cameron, B. J. Ott, and W. S. Payne, “The incidence of adenocar-cinoma in columnar-lined (Barrett’s) esophagus,” N. England J. Med.,vol. 313, no. 14, pp. 857–859, 1985.

[4] K. K. Wang, M. Wongkeesong, and N. S. Buttar, “American gastroen-terological association technical review on the role of the gastroenterol-ogist in the management of esophageal carcinoma,” Gastroenterology,vol. 128, no. 5, pp. 1471–1505, 2005.

[5] N. Shaheen and D. F. Ransohoff, “Gastroesophageal reflux, Barrettesophagus, and esophageal cancer: Clinical applications,” Jama, vol.287, no. 15, pp. 1982–1986, 2002.

[6] G. S. Dulai, S. Guha, K. L. Kahn, J. Gornbein, and W. M. Weinstein,“Preoperative prevalence of Barrett’s esophagus in esophageal adeno-carcinoma: A systematic review,” Gastroenterology, vol. 122, no. 1, pp.26–33, 2002.

[7] D. A. Corley, T. R. Levin, L. A. Habel, N. S. Weiss, and P. A. Buf-fler, “Surveillance and survival in Barrett’s adenocarcinomas: A pop-ulation-based study,” Gastroenterology, vol. 122, no. 3, pp. 633–640,2002.

[8] G. W. Falk, “Barrett’s esophagus,” Gastroenterology, vol. 122, pp.1569–1591, 2002.

[9] M. A. Eloubeidi and D. Provenzale, “Does this patient have Barrett’sesophagus? the utility of predicting Barrett’s esophagus at the indexendoscopy,” Amer. J. Gastroenterology, vol. 94, no. 4, pp. 937–943,1999.

[10] P. Sharma, A. Bansal, S. Mathur, S. Wani, R. Cherian, D. McGregor,A. Higbee, S. Hall, and A. Weston, “The utility of a novel narrow bandimaging endoscopy system in patients with Barrett’s esophagus,” Gas-trointestinal Endoscopy, vol. 64, no. 2, pp. 167–175, 2006.

[11] M. I. Canto, “Diagnosis of Barrett’s esophagus and esophageal neo-plasia: East meets west,” Dig. Endoscopy, vol. 18, pp. s36–s40, 2006.


[12] C. D. Smith, P. A. Bejarano, W. S Melvin, M. G. Patti, R. Muthusamy,and B. J. Dunkin, “Endoscopic ablation of intestinal metaplasiacontaining high-grade dysplasia in esophagectomy patients using aballon-based ablation system,” Surgical Endoscopy, vol. 21, no. 4, pp.560–569, Apr. 2007.

[13] D. A. Corley, T. R. Levin, L. A. Habel, N. S. Weiss, and P. A. Buffler,“Surveillance and survival in Barrett’s Adenocarcinomas: A popula-tion-based study,” Gastroenterology, vol. 122, pp. 633–640, 2002.

[14] V Boolchand, A. Faulx, A. Das, S. Zyzanski, G. Isenberg, G. Cooper,M. V. Sivak, and A. Chak, “Primary care physician attitudes towardendoscopic screening for GERD sysmtoms and unsedated endoscopy,”Gastrointenstinal Endoscopy, vol. 63, no. 2, pp. 228–233, 2006.

[15] J. G. Andrus, R. W. Dolan, and T. D. Anderson, “Transnasalesophagoscopy: A high-yield diagnostic tool,” Laryngoscope, vol.115, pp. 993–996, Jun. 2005.

[16] A. L. Faulx, S. Vela, A. Das, G. Cooper, M. V. Sivak, G. Isenberg,and A. Chak, The Changing Landscape of Practice Patterns RegardingUnsedated Endoscopy and Propofol Use: A National Web Survey, vol.62, no. 1, pp. 9–15, Jul. 2005.

[17] D. Panescu, “Emerging technologies: An imaging pill for gastroin-testinal endoscopy,” IEEE Eng. Med. Bio. Mag., vol. 24, no. 4, pp.12–14, Jul./Aug. 2005.

[18] G. M. Eisen, R. Eliakim, A. Zaman, J. Schwartz, D. Faigel, E. Ron-donotti, F. Villa, E. Weizman, K. Yassin, and R. deFranchis, “The ac-curacy of pillcam ESO capsule endoscopy versus conventional upperendoscopy for the diagnosis of esophageal varices: A prospective three-center pilot study,” Endoscopy, vol. 38, no. 1, pp. 31–35, 2006.

[19] R. N. Keswani, A. Noffsinger, I. Waxman, and M. Bissonnette, “Clin-ical use of in P53 Barrett’s esophagus,” Cancer Epidemio. BiomarkersPrev., vol. 15, no. 7, pp. 1243–1249, Jul. 2006.

[20] L. B. Gerson, P. W. Groeneveld, and G. Triadafilopoulos, “Cost-ef-fectiveness model of endoscopic screening and surveillance in patientswith gastroesophageal reflux disease,” Clin. Gastroenterol. Hepatol.,vol. 2, no. 10, pp. 868–879, 2004.

[21] E. J. Seibel and Q. Y. Smithwick, “Unique features of optical scan-ning, single fiber endoscopy,” Lasers Surgery Med., vol. 30, no. 3, pp.177–83, 2002.

[22] P. Carrascosa, C. Capunay, E. M. Lopez, M. Ulla, R. Castiglioni, andJ. Carrascosa, “Multidetector CT colonoscopy: Evaluation of the per-spective-filet view virtual colon dissection technique for the detectionof elevated leisions,” Abdominal Imag., vol. 32, no. 5, pp. 582–588,Sep. 2007.

[23] E. J. Seibel, R. S. Johnston, and C. D. Melville, “A full-color scanningfiber endoscope,” Proc. SPIE, Opt. Fibers Sens. for Med. Diagnost.Treatment Appl. VI, vol. 6083, pp. 608303-1–8, Mar. 2006.

[24] R. S. Johnston and E. J. Seibel, “Color video imaging with a scanningfiber endoscope,” in Proc. Frontiers Opt./Laser Sci. XXII Topical Meet.CD-ROM, Washington, DC, Oct. 2006.

[25] A. B. Overgaard, J. Hojsted, R. Hansen, J. Moller-Sonnergaard, and L.L. Christrup, “Patients’ evaluation of shape, size and colour of soliddosage forms,” Pharm. World Sci., vol. 23, no. 5, pp. 185–188, 2001.

[26] Paddock Laboratories, Inc., Minneapolis, MN, “Compoundingcapsules,” 2006 [Online]. Available: http://www.paddocklabs.com/im-ages/PadSec_v4n4.pdf.)

[27] B. Rousso, S. Peleg, I. Finci, and A. Rav-Acha, “Universal mosaicingusing pipe projection,” in Proc. ICCV, 1998, pp. 945–952.

[28] A. Agarwala, M. Dontcheva, M. Agrawala, S. Drucker, A. Colburn, B.Curless, D. Salesin, and M. Cohen, “Interactive digital photomontage,”in Proc. SIGGRAPH, 2004, pp. 294–302.

[29] A. Levin, A. Zomet, S. Peleg, and Y. Weiss, “Seamless image stitchingin the gradient domain,” in Proc. ECCV, 2004, pp. 377–389.

[30] A. K. Dattamajumdar, P. L. Blount, J. A. Myers, A. H. Proctor, B. H.Goldman, B. J. Reid, and R. W. Martin, “A low-cost fiber-optic instru-ment to colorimetrically detect patients with Barrett’s esophagus forearly detection of esophageal adenocarcinoma,” IEEE Trans. Biomed.Eng., vol. 48, no. 6, pp. 695–705, Jun. 2001.

[31] F. C. Ramirez, M. S. Shaukat, M. A. Young, D. A. Johnson, and R.Akins, “Feasibility and safety of string, wireless capsule endoscopy inthe diagnosis of Barrett’s esophagus,” Gastrointest. Endoscol., vol. 61,no. 6, pp. 741–746, May 2005.

[32] M. A. Kara, F. P. Peters, P. Fockens, F. J. W. ten Kate, and J. J. G. H. M.Bergman, “Endoscopic video-autofluorescence imaging followed bynarrow band imaging for detecting early neoplasia in Barrett’s esoph-agus,” Gastrointestinal Endoscopy, vol. 64, no. 2, pp. 176–185, 2006.

[33] P. Sharma, N. Marcon, S. Wani, A. Bansal, S. Mathur, R. Sampliner,and C. LIghtdale, “Non-biopsy detection of intestinal metaplasia anddysplasia in Barrett’s esophagus: A prospective multicenter study,” En-doscopy, vol. 38, no. 12, pp. 1206–1212, 2006.

[34] L. B. Lovat, K. Johnson, G. D. Mackenzie, B. R. Clark, M. R. Novelli,S. Davies, M. O’Donovan, C. Selvasekar, S. M. Thorpe, D. Pickard, R.Fitzgerald, T. Fearn, I. Bigio, and S. G. Bown, “Elastic scattering spec-troscopy accurately detects high grade dysplasia and cancer in Barrett’soesophagus,” Gut, vol. 55, pp. 1078–1083, 2006.

Eric J. Seibel, photograph and biography not available at the time of publica-tion.

Robert E. Carroll, photograph and biography not available at the time of pub-lication.

Jason A. Dominitz, photograph and biography not available at the time of pub-lication.

Richard S. Johnston, photograph and biography not available at the time ofpublication.

C. David Melville, photograph and biography not available at the time of pub-lication.

Cameron M. Lee, photograph and biography not available at the time of pub-lication.

Steven M. Seitz, photograph and biography not available at the time of publi-cation.

Michael B. Kimmey, photograph and biography not available at the time ofpublication.

Documents

Tethered Capsule Endoscopy, A Low-Cost and High-Performance Alternative Technology for the Screening of Esophageal Cancer and Barrett's Esophagus