High-resolution endomicroscopy with a spectrally encoded miniature objective

Fiber bundle endomicroscopy techniques have been used for numerous minimally invasive imaging applications. However, these techniques may provide limited spatial sampling due to the limited number of imaging cores inside the fiber bundle. Here, we present a custom-fabricated miniature objective that can be coupled to a fiber bundle and can overcome the fiber bundle’s sampling threshold by utilizing the spectral encoding concept. The objective has an NA of 0.3 and an outer diameter of 2.4 mm, and can yield a maximum spatial resolution of 2 μm. The objective has been validated against a USAF resolution target and ex vivo tissue samples, and as a result yielded images with higher resolution and more details after the spectral encoding concept was employed. © 2019 Optical Society of America under the terms of the OSA Open Access Publishing Agreement


Introduction
Optical fiber bundles have been used to relay images onto a variety of diagnostic devices due to their unique characteristics.They are compact in size, ranging from hundreds of μm to a few mm in diameter [1], mechanically flexible [2], and readily available.These features have made the optical fiber a key component in many minimally invasive endoscopic systems, as it can provide a means to access various cavities in the body without causing a significant amount of tissue damage.One particular area of application includes cancer diagnostics.Cancerous lesions are commonly detected by methods such as needle biopsy and excisional biopsy [3][4][5][6].However, these methods involve the removal of tissue and can be risky, timeconsuming, and expensive [6][7][8].Thus, as an alternate method of cancer diagnostics, optical biopsy approaches that use fiber optics have been developed.Specifically, such fiber based approaches have been tested for the detection of ovarian cancer tumor nodules, Barrett's esophagus, cervical neoplasia, and oral neoplasia [9][10][11][12].Due to its aforementioned characteristics, the optical fiber bundle has also been used in numerous confocal endoscopic systems for in vivo tissue imaging applications [13][14][15][16].Fiber bundle based imaging systems, however, suffer from limited sampling.This is due to the fact that, for the fiber bundle to be flexible enough for in vivo imaging applications, there is an inherent compromise to the number of fiber cores and, thus, there is a trade-off between flexibility, sufficient sampling, and field of view (FOV).Commonly used fiber bundles (containing 10,000-60,000 cores), although they may be flexible, often under-sample morphological features of cells or subcellular architectures, making them difficult to have clinically practical FOV and resolution.
Improving sampling is important to accurately assess morphological features in applications such as cancer diagnostics, monitoring of brain activity, and others.For this reason, numerous attempts have been made in the past to overcome the sampling problems posed by the fiber bundle.Kyrish et al. shifted the sample and the fiber bundle relative to each other by a distance comparable to a sub-fiber core size at a time.This way, the sample was oversampled, which resulted in a lateral resolution improvement [17].This method, however, requires actuators at the distal end of the fiber, which can be difficult to miniaturize and, thus, may prevent minimally invasive imaging.Shin et al. used a single-mode fiber instead of a fiber bundle and employed a scanning mechanism using a two-axis scanning mirror at the fiber's distal end [18].However, using a scanning method can slow down the image acquisition process, which may not be ideal for in vivo imaging applications.Lee et al. superimposed multiple images that were obtained by laterally moving the fiber bundle in x and y directions [19].The method yielded an image that was free of fiber bundle patterns, but no evidence of resolution improvement was shown.Languirand et al. rapidly dithered the fiber bundle to obtain information in the gap space between fiber cores [20].However, this method can yield inconsistent resolution improvement results, as the motion of the fiber bundle resulting from the dithering process can be random.Chang et al. attached a two-axis piezoelectric scanner at the distal end of the fiber bundle and scanned the fiber laterally [21].The resulting images after scanning were stitched together to yield a reconstructed image.As a result, the fiber bundle pattern artifact was reduced in the reconstructed image, but no evidence of resolution improvement was demonstrated.Vyas et al. shifted the fiber bundle in a precise manner using a miniaturized piezoelectric scanner to collect images that are spatially offset from each other [22].The images were then combined together to yield a super resolution image.While the resolution improvement was clearly demonstrated in the paper, the size of the distal optics (~5 mm) is more suitable for large instrument channels of endoscopic systems.
Besides the methods mentioned so far, which rely on a mechanical means to increase sampling of a fiber bundle, there have been attempts to use computational means to overcome the fiber bundle's sampling as well.For instance, Shao et al. used an inverted forward model by which they computed low resolution images using fiber core mapping calculated from the point spread function of a single fiber [23].Then, a maximum a posteriori estimate of a high resolution image using conjugate gradient descent was calculated.Ravì et al. used a method of training deep neural networks (DNNs) [24].Specifically, the authors created models that were trained using a pair of estimated high resolution images and synthetic low resolution images.Then, using such models, a high resolution image with improved quality was obtained.This approach may be, however, limited to the applications that can provide a database of images for the models to be trained on.
In this paper, we present a custom-designed imaging configuration that takes advantage of the aforementioned strengths of the fiber bundle, but does not suffer from under-sampling.The configuration combines a custom-fabricated miniature objective with an optical fiber (FIGH-30-650S, Fujikura Ltd.) and employs the spectral encoding method [25] to exceed the sampling potential of the fiber bundle.To the best of our knowledge, this is the first configuration that includes an integrated miniature probe utilizing dispersion to improve the lateral resolution of a fiber bundle based imaging system without requiring any moving components.The details of the spectral encoding concept are carefully described in our previous proof-of-concept publication [25].While using this concept, in this article, we focus on integrating the method into an endoscope format, system integration, and testing.The custom-fabricated, miniature objective has a diameter of 2.4 mm and provides a spatial resolution of 2 µm in the direction of dispersion when coupled with the optical fiber bundle.The objective contains a prism, which disperses an image of the sample onto the fiber bundle's distal face.As a result, each fiber core inside the fiber bundle captures spectrally encoded signals, each of which is from a different spatial location.The optical fiber works as an imaging conduit and relays these wavelength-coded signals onto the Image Mapping Spectrometer (IMS), which outputs data that is later reconstructed into a 3D data cube (x,y,λ).The 3D data cube is subsequently broken down into λ number of spectral channels, which are ultimately combined into a single high-resolution image using a custom MATLAB reconstruction algorithm.
The proposed optical system has three major advantages compared to other competing imaging modalities.First, it does not require any scanning apparatus, which simplifies its design and enables it to be more easily miniaturized.Secondly, the proposed system includes a custom-designed miniature objective as opposed to a commercial gradient index (GRIN) lens.While GRIN lens based systems [26] suffer from strong spherical and chromatic aberrations [27], the proposed system can be corrected against such aberration errors through customized design optimization step.Lastly, for systems that are fiber bundle limited, the sampling (and also resolution) can be improved by increasing magnification, although this comes at a price of FOV loss [28].However, the proposed configuration can improve the system sampling by a factor of two without the need to change magnification and, therefore, can maintain a relatively large FOV.

Optical design of the objective
The miniature objective was custom designed with the optical design software, ZEMAX (Radiant Zemax, LLC., Redmont, WA, USA), following the specifications listed in Table 1 below.An object space NA of 0.3 yields a diffraction limited resolution of around 1.1 μm at the central wavelength of 543 nm.The objective is designed to work with the fluorescent dye proflavine, which stains cell nuclei and whose emission full width half maximum spectrum covers the range of 515-570 nm [25].The objective's working distance of 150 μm allows it to be used in close proximity to tissue and to work through 0 thickness coverslips.The optical elements are designed to be fitted inside a hypodermic tube with an outer diameter of 2.41 mm and an inner diameter of 2.16 mm.The objective's clear aperture of 1.69 mm is smaller than the inner diameter of the tube.
The objective's lens prescription data is listed in Table 2 and its optical layout is shown in Fig. 1.The objective consists of two Poly(methyl methacrylate) (PMMA) singlet lenses, one doublet lens consisting of polystyrene and PMMA, and one zinc sulfide (ZnS) prism.PMMA and polystyrene were chosen as lens materials since they provide high transmission and low autofluorescence in the proposed wavelength range [27].They are also easily machinable, especially for fabricating aspheric surfaces, which effectively correct for spherical aberrations.ZnS, a polycrystalline material, was chosen as the prism material due to its high refractive index (n = 2.39 at 543 nm) and high dispersion.The lenses making up the doublet lens were fixed together with NOA 61 optical glue (Norland, New Jersey, USA).The front surface of the prism has 4-degree angle, allowing it to disperse light over a distance of approximately 8 μm.The dispersion distance is long enough to cover at least the core-to-core distance of the fiber bundle.The miniature objective is designed to be immersed in saline solution.Note that the objective design was performed with seawater (see Table 2), a standard    wn in Fig. 2, th gative edge of A decrease in p ance depicted n in Fig. 3, the or the sagittal r dge of the fie ally dispersive on the y-axis.
-axis field, wh was evaluated prior to the d .Thus, the obj ifically, the mi nd Monte Carl

Imaging
The optical se spectral encod target and the bundle.The prism result of the d spectrally enc Mapping Spe spectrometer that yield low in the 515-57 cores, 30,000 The IMS r a 3D data cub are used off demonstrates mensions, upp wercase counte mplex conjugat product, F(u,v correlation, an fset, informatio 570 nm is com h spans the e core distance o hat exhibits res des 21 spectra l images, the e olation method n in Fig. 10 an shown in Fig.The result also used a co of the tissue s Fig. 14.Diffe microscope, r optical setups

Discussio
Based on the Fig. 7 tion, the reconstruction algorithm will be further optimized to reduce its computational cost and time to ultimately be used for in vivo real-time imaging applications.

Conclusions
In this study, a high-resolution, spectrally encoded miniature objective has been designed, fabricated, and assessed.The objective has an outer diameter of 2.4 mm and can easily be integrated with a system that utilizes a fiber bundle for minimally invasive imaging applications.The performance of the miniature probe was validated against a USAF resolution target and ex vivo tissue samples.As a result, the probe successfully yielded images with higher resolution and more details after the spectral encoding concept was employed without requiring any scanning components.Since the objective can obtain an image in a snapshot fashion at a video rate, optimization of the current reconstruction algorithm may allow the system to be used for real-time imaging applications in the future.As a next step, we aim to develop an in vivo minimally invasive imaging modality after further miniaturizing the distal optics and fabricating an objective with an outer diameter of approximately 1 mm.The objective will include a disperser allowing two-dimensional encoding, which will enable us to increase sampling in both vertical and horizontal directions.The objective will also work in the broadband wavelength range, which will allow it to simultaneously work with multiple dyes.
Fig. 1.Opti performance o ction (MTF) p system is limi mpling criteria).e core-to-core d different than 5 ures will chan ich would low e spectral enco ted imaging th ikely resolvabl formance of t Also, due to th s and y-axis w Fig. 2 for the Fig. 4 ensure comeThe fabric with an outer of the assemb placed on a U elements insi objective to p Fig. 6 prism on a U 2.4.Objectiv Lateral resolu Air Force) hig is shown in F used and a 10 used to relay Canada).Chr bandpass filte assessment ar smallest featu proposed wav

11 .
Fig. 1 system spectr and 4 vertic visual Based on single channe sampling limi reconstruction element 3 (32 target feature pattern.The c using the spec due to the fac improvement theoretically e the optical se bundle.In gen bi-directional bundle as see 10, which we predominantly cores were ea 3.2.Imaging Porcine tissue Meat Co, Spr oral cheek tis which proflav proflavine dy (PBS) solutio samples.

Fig. 11 ,
Fig. 11, up to el image.The r it of the fiber n step using a 22.5 lp/mm) v es in Fig. 11(b cross section in ctral encoding ct that the disp shown betwe expected 2 tim tup shown in F neral, the samp improvement en in our prior ere based on y one direction asily visualized g results with e e samples (J a ring, TX) were ssues were im vine readily bo e and any exce on.The optica