Design, fabrication, and preclinical testing of a miniaturized, multispectral, chip-on-tip, imaging probe for intraluminal fluorescence imaging of the gastrointestinal tract

Gastrointestinal cancers continue to account for a disproportionately large percentage of annual cancer deaths in the US. Advancements in miniature imaging technology combined with a need for precise and thorough tumor detection in gastrointestinal cancer screenings fuel the demand for new, small-scale, and low-cost methods of localization and margin identification with improved accuracy. Here, we report the development of a miniaturized, chip-on-tip, multispectral, fluorescence imaging probe designed to port through a gastroscope working channel with the aim of detecting cancerous lesions in point-of-care endoscopy of the gastrointestinal lumen. Preclinical testing has confirmed fluorescence sensitivity and supports that this miniature probe can locate structures of interest via detection of fluorescence emission from exogenous contrast agents. This work demonstrates the design and preliminary performance evaluation of a miniaturized, single-use, chip-on-tip fluorescence imaging system, capable of detecting multiple fluorochromes, and devised for deployment via the accessory channel of a standard gastroscope.


1
Introduction 28 The clinical use of endoscopic imaging devices remains critical to the discovery of malignancies that 29 arise in internal luminal organs, such as those of the gastrointestinal (GI) tract. As endoscopes 30 continue to improve in resolution, sensitivity, and operability within tight spaces, so, too, have 31 clinicians' ability to visualize subtle morphological and functional changes indicative of pathogenesis 32 miniaturized, multispectral, distal chip probe 2 in a variety of tissues, diseases, and procedures. Endoscopic devices are now commonly used in 33 combination with surgical tools to perform complex laparoscopic operations that can improve clinical 34 outcomes in oncologic surgery. [

1] 35
Endoscopy of the GI tract is generally performed with broadband illumination. However, current 36 implementation of white light endoscopy (WLE), which has proven less sensitive than other methods 37 of imaging, such as narrowband and chromoendoscopy, may not be sensitive enough to resolve 38 changes at the molecular scale present in precancerous lesions.[2,3] The sensitivity of WLE is limited 39 by the operator's ability to detect subtle changes in color, texture, and form, all of which may present 40 differently between patients. Additionally, WLE is not specific for demarcating tumor margins, 41 which is critical for facilitating the complete resection of malignant tissues that can seed future 42 recurrence of GI cancer. Therefore, the development of advanced imaging technologies, such as 43 those which implement fluorescence imaging, could greatly improve clinical practice by providing 44 tumor-specific contrast with high sensitivity. Although most endoscopes do not feature this imaging 45 modality, clinical gastroscopes are designed to support the addition of modular accessories through a 46 working channel to expand the operator's capabilities during different procedures. Moreover, the 2.8-47 mm diameter channel of diagnostic gastroscopes has traditionally posed challenges for introducing 48 imaging devices, which often exceed this diameter. However, the miniaturization of chip-on-tip 49 imaging systems has simplified the development of endoscopes of a compatible size. The clinical 50 value of these miniature, accessory-port imaging instruments is multifold: reductions in outer 51 diameter have enabled endoscopes to traverse narrow anatomical regions that were previously 52 inaccessible, and when combined with standard reusable gastroscopes, single-use accessory-port 53 endoscopes can supplement high-resolution gastroscope images with other imaging modalities, 54 including fluorescence. 55 Novel miniaturize endoscopic systems have provided access to millimeter-scale lumens for collecting 56 data using autofluorescence, reflectance, and optical coherence tomography modalities.
[4] On the 57 forefront of advanced miniaturized endoscopy, sub-millimeter scanning fiber endoscopes pave the 58 way for detecting and imaging cancer in regions of the human anatomy that were previously 59 inaccessible via endoscope. [5,6] While fiber endoscopes provide powerful imaging capabilities, 60 optical fibers are both expensive and prone to damage from crushing and bending, which can result in 61 a reduction in image quality. As a result, chip-on-tip endoscopy as a miniature alternative to fiber 62 endoscopy has gained significant traction, with some camera module footprints as small as 0.55 mm 63 x 0.55 mm.

Design and Fabrication 101
A miniature, biopsy-enabled, chip-on-tip imaging assembly was created to detect fluorescent 102 emissions in four unique wavelength bands without the requisite of changing front-end optics 103 between varying wavelengths. The probe was designed to detect emissions from several fluorophores 104 common in research and clinical settings, including fluorescein and quinine. The design 105 specifications are listed in Table 1. Mechanical specifications of size constrained the endoscope to 106 fitting within the working channel of a commercial articulating gastroscope. 107

Illumination Assembly 108
The onboard illumination assembly was designed to operate in four distinct wavelength bands to 109 target the absorption peaks from fluorescent compositions with peak excitation wavelengths of 405, 110 488, 561, and 640 nm (Figure 1, A). To provide illumination at each excitation wavelength, four 111 fiber-coupled LED light sources (M405FP1, M490F3, MINTF4, M625F2; Thorlabs, Newton, NJ) 112 were connected via a 400 μm core 1-to-4 fanout fiber bundle (BF44LS01; Thorlabs) to an aspheric 113 condensing lens (ACL2520U-A; Thorlabs). To prevent light leakage into detection bands, each LED 114 was filtered with a corresponding unique 20 nm bandpass filter (FB405-10, FL488-10, FB560-10, 115 FB640-10; Thorlabs) with full-width half-max wavelengths corresponding to the peak absorption 116 wavelengths of a subset of fluorescent probes. 117 These bandpass filters were mounted in a filter wheel (Thorlabs FW2A) and focused with a second 118 aspheric condensing lens (Thorlabs ACL2520U-A) onto a 1x2 multimode fiber optic coupler 119 (Thorlabs TH400R5F1B), which pairs with two multimode fibers (Thorlabs M136L03) which run 120 through the length of the imaging probe shaft and exit at the distal tip of the device. 121

Imaging Assembly 122
A miniature, monochromatic camera, Osiris M (OptaSensor GmbH, Nürnberg, Germany), was 123 selected to perform image acquisition. The 1mm x 1mm x 2mm unit captures images sized at 320 px 124 x 320 px with a pixel size of 2.4 μm.
[19] To interface with a computer, signal from the Osiris M is 125 first routed through an external image processing system. To remove eliminate signal from 126 illumination sources, a 1.5 mm x 1.5 mm multiband interference filter (Iridian, Ottawa, ON, Canada) 127 was custom designed to meet the specifications in Table 1 and fabricated via thin film deposition, 128 impeding the transmission of excitatory illumination wavelengths with an optical density of 5 while 129 simultaneously permitting transmission of desired fluorescence emission wavelengths (Figure 1 B). 130

Biopsy Assembly 131
In the histological characterization of tissues in the luminal GI tract, it is necessary to collect bulk 132 tissue biopsies with forceps (as opposed to needle aspiration) to preserve tissue architecture, a critical 133 feature in histopathological sample analysis. To maintain a miniature footprint, 3 Fr Piranha® biopsy 134 forceps (Boston Scientific M0065051600, Marlborough, MA), shown in Figure 1 C, were 135 incorporated through a backloading channel and were both introduced and secured using a 136 Hemostasis Valve Y connector (Qosina 97380, Ronkonkoma, NY), featuring a Tuohy Borst adapter 137 with rotating male luer lock and an angled female luer injection side port. 138

Operator Interface & Device Assembly 139
To secure components of the imaging system, a cylindrical housing of 2.7 mm in diameter by 10 mm 140 in length with four through-holes was designed in SolidWorks (Dassault Systèmes, Vélizy-141 Villacoublay, France) and fabricated with stereolithography 3D printing (Elegoo US-SO-3D-110,  142 Shenzhen, Guangdong, China). The terminal ends of the optical fibers were mounted in the lateral 143 through-holes of the cylinder, and Osiris M camera module was positioned in the central through-144 hole of the printed cylinder such that the multiband filter could then be placed directly in contact with 145 the front of the camera lens, facing outward from the distal tip. The housing was designed so that the 146 fiber channels possessed a subtle curvature to direct the two output beams into a single overlapping 147 illumination region, aligned with the imaging field of view. The biopsy forceps were mounted 148 through the chamber located below the camera module. These components were housed in a 149 stainless-steel shell measuring 3 mm x 14 mm (Microgroup, Medway, Massachusetts with a stand was designed to secure the device to allow for stationary imaging (Figure 1, c). 168 Fluorescence imaging may be toggled by switching between illumination sources. Reflectance 169 imaging can be achieved using broadband illumination or narrow band illumination in wavelengths 170 within the transmission ranges accepted by the distal tip multiband filter. 171 To collect a biopsy, two hands are required in instances in which the device is not mounted and 172 stationary. In this configuration, the biopsy forceps are run through the Touhy Borst at the proximal 173 end of the Qosina Y connector. Next, the user tightens the extended forceps such that they are both in 174 contact with tissue and sufficiently visible to the camera. The imaging device is handled by the user's 175 dominant hand, and the biopsy mechanism is activated by the user's non-dominant hand. In instances 176 in which the imaging system is mounted and stationary, this process may be performed with one 177 hand. Collected tissue samples may then be retrieved by loosening the Touhy Borst and retracting the 178 closed biopsy forceps from the forceps subchannel. image. This is a common effect seen in imaging systems with large fields of view such as in 195 endoscopic imaging. Correction measures can be implemented during postprocessing, provided that 196 the degree and type of distortion are well-characterized. Distortion of the system was measured using 197 a grid target (ThorLabs, R2L2S3P4). To calculate distortion, the target grid was placed such that the 198 pattern filled the field of view of the camera. Images were captured and imported into ImageJ where 199 the distance from the image center of diagonal grid points was measured and the deviation from 200 linearity was recorded. 201

Fluorescence Sensitivity 202
To measure fluorescence sensitivity, serial dilutions were performed from a stock solution of a 203 known concentration of Fluorescein and the resulting dilutions were transferred to standard 1 cm x 1 204 cm spectrometry cuvettes. The cuvettes were illuminated with the device using the 488 nm LED and 205 images were collected for each concentration. Images were analyzed by calculating the mean image 206 brightness for each concentration in a stationary region of interest to detect the illumination level at 207 which the sensor is sensitive to the presence of fluorescent emission. 208

Verification of Biopsy Quality 210
Tissue samples were collected from the lumens of excised murine duodenums using Piranha® biopsy 211 forceps tracked through the fluorescence probe subchannel. Samples were fixed for 30 minutes in 4% 212 paraformaldehyde, dehydrated overnight in 30% D-sucrose, and then embedded in O.C.T. (optimal 213 cutting temperature) compound. Tissue was cryosectioned to a 10-micron thickness and stained 214 according to a standard hematoxylin and eosin, or H&E, protocol. Following staining, tissue quality 215 assessment was performed by the identification of crypts and villi, fragile but crucial landmark 216 structures within duodenal tissue, via light microscopy. 217 proximal duodenum were removed, opened along the greater curvature to expose the lumen, and 226

Detection of Fluorescence, Ex vivo, in Gastrin
flushed with ice-cold phosphate buffered saline (PBS) prior to imaging. 227 The imaging system was configured with the distal tip 2-3 cm above samples of ex vivo murine 228 stomach and duodenal tissue. An external light source (Lambda LS, Sutter Instruments, Novato CA) 229 was used to illuminate sections of mouse GI tissue in a petri dish, and images were captured with 230 both broadband unfiltered light and 488 nm (Thorlabs, FB560-10) filtered illumination for excitation 231 of ZsGreen. Images were captured and fluorescence detection data were digitally enhanced and 232 overlayed upon the broadband light images to create a composite highlighting the signal-producing 233 regions of the GI tract. 234

Examination of Tissue Autofluorescence Capabilities 235
To assess if the device is sufficiently sensitive to collect autofluorescence, a portion of the GI tract 236 from forestomach to duodenum, brain, lung, and pancreatic tissues was excised from healthy, 237 unlabeled wild type mice via necropsy. These sections were briefly placed on ice in a petri dish with 238 PBS and fluorescent images were collected with both the multiband fluorescence imaging probe and 239 a previously-reported Multispectral Fluorescence Imaging System (MSFI) at four distinct 240 wavelengths: 400nm, 490nm, 543nm, and 560nm, wherein 560nm serves as reflectance imaging 241 wavelength and 400nm, 490nm, and 560nm serve as fluorescence emission imaging wavelengths 242 [22]. 243 3 Results 244

Device Construction and Characterization 245
Analysis of the slanted-edge MTF method yielded the line spread function and the MTF in Figure 2  246 (A). The MTF50 of the system is 0.2 cycles/pixel. The cutoff frequency for this system was 247 determined to be 0.35 cycles/pixel. Due to the inverse relationship between cutoff frequency and 248 resolution, the resolution of the system can be estimated as approximately 6.72μm. 249 The image captured in Figure 2

Biopsy Quality Verification 262
The resulting H&E slides were imaged under a light microscope, and analysis of bright-field images 263 confirmed that duodenal biopsies collected with the Piranha biopsy forceps preserved tissue 264 architecture for histological observations. Crypts, villi, and other structural features retained their 265 anatomical features, Figure 3  invaluable for assessing biopsied tumor tissue. 267

Ex vivo Fluorescence Detection in Gastrin CreERT2; ZsGreen Models 268
Images captured of ex vivo murine GI tissue indicate the presence of fluorescence in the antrum of 269 the stomach in transgenic models, where gastrin-secreting G cells are predominantly located. The 270 expression of ZsGreen was additionally confirmed by fluorescence microscopy of cryosectioned 271 antral tissue (Figure 3 (B)), but the signal was not recorded in wild type mice. 272

Examination of Tissue Autofluorescence Capabilities 273
Detectable levels of autofluorescence were recorded by the miniature multiband fluorescence 274 imaging system in the forestomaches of unlabeled mice (Figure 3 (C)). This finding was 275 corroborated by high resolution images captured by the external MSFI system. Fluorescent artifacts 276 were noted in images captured with both systems. Tissues were further analyzed using fluorescence 277 microscopy to eliminate the possibility of contamination with exogenous fluorescent moieties, with 278 no evidence of contamination found. 279 4 Discussion 280

Device Design & Configuration 281
At less than 3 mm in diameter, the component that spaces and houses each piece posed a 282 manufacturing challenge, as the scale of the part, 2.7mm in diameter, requires very fine tolerances 283 that are difficult to achieve without resorting to costly and time-consuming methods like wire 284 Electrical Discharge Machine (EDM). One of the simplest alternatives to manufacture designs in 285 swift succession is through 3D printing. Yet, most commercial FDM 3D printers do not possess 286 sufficient resolution to reliably print pieces with through-holes less than one third of a millimeter in 287 diameter. In contrast, high precision, small-scale printers like those created by Nanoscribe, are 288 specially designed to print 3D structures that are far smaller than 2.7mm. Remarkably, it was through 289 a widely available consumer-grade stereolithography (SLA) resin 3D printer (Mars 2 Pro, Elegoo), 290 that this housing component was successfully manufactured. 291

Characterization 292
The low system resolution can be attributed to several limitations imposed by both the physical 293 system and test method which, were not representative of diffraction-limited conditions. First, the 294 camera module, which pushes the limits of present manufacturing capabilities, is generated through 295 the adhesion of a series of optical components that likely result in a degree of aberration due to both 296 coma and spherical aberration. As a consequence of performing this MTF analysis in a setting 297 designed to mimic general use cases, parameters like low light and a broad range of illumination 298 wavelengths serve to reduce contrast and induce chromatic aberration, further reducing measured 299 values of resolution. Resulting from the nature of the fluorescence emission of the backlit slide, the 300 imaged interface does not exhibit a sharp, well-contrasted transition between surfaces. The MTF 301 values gained from this analysis serve as conservative estimates of device functionality in dimly 302 emissive environments and represent the capabilities of the system in a typical use setting while also 303 highlighting room for improvement. As miniaturized sensors continue to advance, we can anticipate 304 higher resolution in the same or smaller sensor footprints with specific improvements to performance 305 in low-light settings. 306 In instances of geometric distortion, images captured do not reflect the reality of the proportions of 307 objects imaged as a result of a relationship between magnification and height within the optics of the 308 camera module. In this wide-angle camera module, barrel distortion is evident, indicating that objects 309 captured near to the center of the field of view occupy a larger proportion of the overall frame than 310 objects imaged at the periphery. Once quantified, geometric distortion can be corrected during image 311 post-processing. 312 Although the imaging system successfully detected highly dilute levels of fluorescein during 313 experimental measurements of fluorescence sensitivity, few resources exist to quantify the level of 314 sensitivity a system must possess to adequately perform in biological fluorescence applications, thus 315 additional verification in additional biological models are requisite. 316

Preclinical Testing 317
Experimental results from biopsy collection confirmed that miniature bite biopsy performed with 3 318 Fr Piranha forceps yields meaningful structural information that could be used to identify 319 hyperplastic and dysplastic tissues in regions identified as potentially cancerous, supporting this 320 method as a useful accessory to be applied in tandem with fluorescence imaging. 321 Images captured of fluorescent ZsGreen signal localized to the antrum support that the miniature 322 multiband fluorescence imaging system is capable of accurately detecting concentrations of 323 fluorescent moieties within labeled tissue. From fluorescent microscopy images, it is evident that 324 only a small number of cells in the antrum, presumably G cells, were producing the fluorescent 325 protein, suggesting that the device is highly sensitive to ZsGreen. This can be further corroborated by 326 the lack of signal detection in the corpus and duodenum, which are not modified to express the 327 fluorescent reporter. 328 Preliminary animal testing has highlighted the system's sensitivity to artificial fluorescent markers as 329 in the studies involving ZsGreen expression; however, results from autofluorescence imaging studies 330 have proven the imaging system to be sensitive to the emissions of endogenous fluorophores. 331 Autofluorescence imaging has been clinically employed as a diagnostic pathway for flat GI tumors 332 through endoscopic autofluorescence imaging, supporting its utility in a miniaturized chip-on-tip 333 imaging system.
[24] When comparing images captured by both the high-resolution MSFI and the 334 miniature multiband fluorescence imaging system, the similarity in tissue autofluorescence pattern is 335 marked. However, images captured by the miniature system are more than 200x smaller, and the 336 overall bit depth of the system is much lower, resulting in a significant loss of detail. Between the 337 two systems, images from the miniature imaging device possessed more visual artifacts and had a 338 lower signal-to-noise ratio; although, the miniature system also performed with a much shorter 339 exposure window, at just a fraction of a second. With additional adjustments for gain and 340 thresholding, fluorescent regions could be made more discrete. Despite the high optical density, light 341 leakage around the edges of the multiband filter appear to occasionally interfere with the signal 342 quality. Signal artifacts were also caused by the ice bath in which tissues were imaged to preserve 343 tissue quality for subsequent histopathology; consequently, the irregular ice particles refracted 344 fluorescence emissions such that the ice itself appeared to emit a signal. 345

Conclusion 346
We present the design and characterization of a miniature, handheld, chip-on-tip fluorescence 347 imaging probe with a front-end multiband filter for the imaging of multiplexed fluorophores to aid in 348 the localization of labeled cancerous lesions.  with a composite image at the right, as captured by the miniature multispectral system. Row 452 (II) depicts the stomach and duodenum of a transgenic mouse that produces ZsGreen in