Fluorescence imaging has often been employed clinically to map retinal vasculature. An example is fluorescein angiography (FA), which has been a standard for diagnosing conditions such as diabetic retinopathy and macular degeneration—both major causes of preventable blindness and visual impairment [1–5]. Typically, FA is performed at blue/green wavelengths with a fundus camera, but can also be accomplished with confocal and ultra-wide methods [6,7]. A similar technique is indocyanine green (ICG) angiography, which can be performed at near-infrared wavelengths to identify abnormalities in choroidal circulation [8]. More recently, fundus autofluorescence imaging has entered clinical use as a complementary procedure to FA. The technique involves imaging intrinsic macular fluorophores (i.e., lipofuscin) at green/orange wavelengths and can be performed with similar hardware to diagnose many retinal conditions, including age-related macular degeneration [9]. However, image quality in fundus cameras can vary widely, and can be affected by factors such as system design and photographer skill [10–12]. These issues indicate a need for a universal standard to enable stronger quality control on fluorescence imaging devices and their implementation. Another imaging modality lacking in well-established performance test methods is retinal oximetry, which is also emerging as an effective method for diagnosing diseases linked to abnormal retinal oxygen saturation [13,14].

Microscale performance test methods are needed to provide objective, quantitative, and consistent evaluation of ophthalmic imaging devices in a manner similar to emerging consensus approaches for intraoperative fluorescence systems [15]. Advancements in additive manufacturing (or “three-dimensional [3D] printing”) have enabled the fabrication of test objects that can provide insights into imaging system performance, including biomimetic phantoms [16,17]. For example, Ghassemi et al. used a stereolithography (SLA) 3D printer to generate a fillable-channel hyperspectral reflectance phantom from a vascular model segmented from a patient fundus image, achieving channel diameters of 750 µm in the retinal phantom and 450 µm in a straight channel model [18]. Another spectral imaging phantom was fabricated using laser micromachining to engrave channels in adhesive tape, reaching a minimum channel diameter of 160 µm [19]. Both phantoms, however, featured channel diameters significantly larger than the 10 µm diameters of retinal capillaries [18,19].

Direct laser writing (DLW) is a 3D manufacturing technology based on two-photon (or multi-photon) polymerization that offers unique means for fabricating microstructures with 150 nm feature resolutions [20], and thus a pathway to phantoms with biologically relevant diameters. Previously, the Sochol group introduced an in situ DLW (is DLW) technique for printing microfluidic structures directly inside—and attached to—enclosed microchannels [21,22]. In particular, the use of devices composing cyclic olefin polymer (COP) enabled the fabrication of a microfluidic system consisting of two interwoven microvessel-inspired components with diameters of 8 µm each [22]. DLW has also been applied to achieve micrometer-scale features in optical phantoms, including a 50 µm thick structure to simulate an arterial plaque and micrometer-scale cylindrical posts representing photoreceptors [23–25]. In addition, the fluorescent properties of the photoresist used as a DLW printing substrate present a unique opportunity to fabricate label-free fluorescent imaging targets. In this Letter, we demonstrate that DLW can be used to fabricate high-resolution fluorescent image quality targets and biomimetic vascular phantoms.

Both a simple bifurcation model and a biomimetic model of retinal vasculature were fabricated as hollow channel phantoms via a COP-based DLW fabrication process [22]. This technique involved printing microchannels within a mold pattern inscribed on a COP sheet bonded to a silicon wafer. Computer-aided designs of a circular center attached to two linear regions [Figs. 1(A) and 1(D)] were created using a software package (DeScribe, Nanoscribe GmbH, Stutensee, Germany). The dimensions for the two linear regions were 140 µm in width, 3000 µm in length, and 22 µm in height while the circular center was 422 µm in diameter for the bifurcation [Fig. 1(A)] and a width of 422 µm, length of 1400 µm, and height of 22 µm for the linear regions, and the same dimensions for the circular center for the retina [Fig. 1(D)]. The models were printed using DLW with IPS-photoresist (Nanoscribe) on the polished surface of a silicon wafer using the ${25\times }$ objective. The prints were then developed via submersion in propylene glycol methyl ether acetate (PGMEA) to dissolve the uncured photoresist. The prints of these geometries in a completely solid form on a silicon wafer acted as a mold for the channels within the COP device.

 

Fig. 1. Conceptual illustrations of the COP device, channel phantom, and perfused channels of bifurcation and retina models. (A)–(C) Microchannel mold for the bifurcation, completed bifurcation model printed in situ within the COP device, and IR700 perfused within the channels. (D)–(F) Microchannel mold for the retina phantom, retina phantom printed in-situ microchannels within the COP device, and IR-700 perfused within the channels. (G) Wide angle view of the completed device, including a square COP sheet placed on COP film. The input/output ports are not shown.

Download Full Size | PDF

To manufacture the COP device, a 3 mm thick COP sheet (ZEONOR 1060 R, Zeon Corp., Japan) was first immersed in acetone, then placed in a sonicator for 1 h, and dried with ${{\rm N}_2}$ gas. The wafer with the circular mold was then placed in a thermal press for 2 min at 182°C. A COP sheet was placed on top of the wafer, and a polydimethylsiloxane (PDMS) slab with glass on top was placed on top of the COP sheet. The COP sheet was hot embossed for 5 min at 182°C to inscribe the microchannel design from the mold. Heat to the thermal press was turned off for 10 min before releasing the pressure on the COP sheet. Holes were then drilled in the molded COP device to create inlets and outlets to the channels. The inscribed COP sheet was then bonded to a 100 µm thick COP film to create a sealed microfluidic device. This was accomplished by heating cyclohexane for 1 h at 30°C before exposing the surface of the COP film to the vapors for approximately 2 min. After removing the cover, the COP film was placed tacky side up on a smooth, metal block, and the molded COP device was pressed microchannel side down onto the COP film for 1 min at room temperature (RT) (20–25°C) to facilitate bonding.

To begin the is DLW process for the COP-COP device with enclosed microchannels, IP-L 780 photoresist (Nanoscribe) was vacuum loaded into the microchannels through the inlet with a syringe. The channels were then printed in situ (i.e., within the microchannels of the COP device) using the ${63} \times$ objective lens in an oil-immersion mode via a ceiling-to-floor, point-by-point, layer-by-layer writing path. After print completion, the COP device was submerged in a PGMEA bath overnight to dissolve the uncured photoresist in the device. A microfluidic pump (Fluigent MCFS-EZ, France) was then used to perfuse the device with air for 90 sec at 50–100 kPa of pressure to remove any remaining solvent. The device was then flushed with isopropyl alcohol and cleared with air. An illustration of the completed channel phantom in the COP device is shown in Fig. 1(G). This procedure was used to fabricate two models: (i) one with two bifurcations [Figs. 1(A)–1(C)] and (ii) one with channels derived from retinal vasculature [Figs. 1(D)–1(F)].

A 50 µg/mL solution of fluorescent dye, IR700 (IRDye700, LI-COR, Inc., Lincoln, NE), was injected into the hollow channel devices using the microfluidic pump (Fluigent MCFS-EZ, Chelmsford, MA) to assess channel perfusion. Completed phantoms were imaged before and after dye infusion using a laser scanning confocal microscope (Zeiss LSM710, Carl Zeiss, Thornwood, NY). Fluorescence images of the phantoms were acquired at 405 nm excitation and 415–487 nm emission, as well as at 633 nm excitation and 693–758 nm emission to detect IR700. The 60 nm gap between excitation and emission was implemented to minimize detection of photoresist fluorescence.

Strong emission by the photoresist across a wide wavelength range also presented an opportunity to create fluorescence targets. In an initial test, the dimensions and spacing of elements in groups 2–7 of the standard USAF 1951 spatial resolution target were replicated in SolidWorks (Dassault Systemes, Vélizy-Villacoublay, France) and extruded to create a 10 µm tall volumetric model. The resolution target was printed using IPS and the ${25\times }$ objective, then imaged with a brightfield microscope and with the confocal microscope at 405/415–487 nm excitation/emission. In addition, a set of two biomimetic vascular geometries adapted from [18] was printed with the same material and settings as the resolution targets. In the first model, the channels were resized to 17.5 µm in diameter, and the bottom was flattened in SolidWorks to produce a “positive” model where the vasculature was fluorescent. The second model was generated by subtracting the first model from a ${550} \times {550} \times {10}\;\unicode{x00B5}{\rm m}$ block in SolidWorks, thus creating a “negative” image and a resultant phantom where the entire sample, except for the vessel regions, was fluorescent. This approach was used to simulate fundus autofluorescence. Both phantoms were imaged with the confocal microscope at six excitation/emission configurations: (i) 405/415–487, (ii) 458/467–514, (iii) 488/492–544, (iv) 514/517–588, (v) 561/574–656, and (vi) 633/640–656 nm. Images were edited for presentation in ImageJ (NIH, Bethesda, MD).

Imaging at 405 nm shows a minimal signal from the channels in both the bifurcation and biomimetic models, indicating that the channels were sufficiently cleared of liquid photoresist [Figs. 2(A) and 2(E)]. Imaging prior to IR700 perfusion showed minimal background fluorescence from the phantoms [Figs. 2(C) and 2(G)]. Following IR700 injection, imaging at 633 nm produced fluorescence in all channels, indicating that all channels were dye-perfused [Figs. 2(D) and 2(H)]. There was, however, a substantial signal in the cavity surrounding the printed channels and in the stitching between the blocks forming the printed channels. The DLW build area for the ${63} \times$ objective lens is limited to approximately ${180} \times {180}\;\unicode{x00B5}{\rm m}^2$—significantly smaller than the size of our printed channels. To bypass this limitation, DLW enables serial printing of adjacent blocks with some overlap to print larger models—a process termed “stitching”—resulting in the rectangular pattern seen in Figs. 2(D) and 2(H). The smooth gradation of the signal radiating from the stitching blocks suggests that the signal may be due to scattering of emission light within the otherwise transparent phantom. The presence of the signal with significantly higher intensities in the vicinity outside the printed channel phantom, however, suggests that dye has leaked into the gaps between the COP device and the printed structures. A 3D measurement tool was used to determine channel diameters from stacked confocal images (Zen Blue, Carl Zeiss, Thornwood, NY). The lateral and axial diameters were measured to be 9.96 and 10.61 µm for the bifurcation model and 9.96 and 9.85 µm for the biomimetic model [Figs. 2(D) and 2(H), insets].

 

Fig. 2. (A), (E) Digital view and (B), (F) and confocal imaging of photoresist and (C)–G), (D)–H) IR700 in hollow channel phantoms. All fluorescence images were cropped. The insets in (D) and (H) display cross-sectional images, where tick marks indicate channel boundaries.

Download Full Size | PDF

The DLW-printed spatial resolution target demonstrated strong fluorescence at 405 nm excitation (Fig. 3). The smallest element in the model, Group 7 Element 6 (221.8 line pairs/millimeter spatial frequency, 2.19 µm line width) was also successfully printed (Fig. 3, inset). Most groups showed uniform spacing, indicating that the model printed accurately. Nonuniformity in fluorescence images is most likely due to the excitation being stronger towards the center of the image—a limitation of the imaging system.

 

Fig. 3. Digital model (left), brightfield images (top right), and confocal fluorescence images (bottom right) of the resolution target. The inset on the bottom right displays Group 7 Element 6 (the smallest element). The confocal fluorescence image of Groups 6–7 was cropped.

Download Full Size | PDF

Both printed models exhibited fluorescence with all six excitation/emission configurations, but signals were somewhat lower at 458/467–514 and 514/517–588 nm (Fig. 4). All vessels can be distinguished in both the solid and negative versions of the biomimetic models at 458 nm excitation, but are not as clearly visible with 488 and 633 nm excitation. While the negative models showed successfully printed channels, in some cases the stitched regions exhibited somewhat higher fluorescence than the center of the blocks. It can also be noted that beyond the stitching the signal is consistent, most likely because all retina models were consistently 10 µm in height. This differs from the nonuniformity in clinical images, in which the center of the image contains a dark spot from the optic disc.

 

Fig. 4. Brightfield and confocal fluorescence images of solid-form positive and negative vascular phantoms. All images were cropped.

Download Full Size | PDF

Our results show that DLW can generate hollow channel phantoms, as well as solid fluorescent resolution targets and retinal phantoms. The hollow channel phantoms represent, to the best of our knowledge, the only retinal vascular phantoms to reach the true diameter of retinal capillaries (10 µm) while achieving biomimetic geometric complexity. In addition, the photoresist used in this Letter possesses fluorescent properties that have not yet specifically been applied to fluorescent imaging phantoms. The resolution targets and solid retina phantoms demonstrate a novel application of DLW to generate label-free fluorescent models for microscopy and clinical imaging.

Through the application of isDLW techniques, we were able to achieve patent channels with diameters of 10 µm in both a simple bifurcation design and a biomimetic vascular geometry. Fluorescence seen in the cavity may indicate that the seal between the print and the COP mold was not secure, enabling dye leakage outside the channels. This issue could potentially be resolved by adjusting scan speed and laser power of the DLW device to create a more robust seal between the splitting blocks. Emission light scattering off regions of turbidity in the cavity may provide another possible explanation for the signal surrounding the stitching between blocks. Advances in DLW—namely, the recent release of a ${10} \times$ objective lens capable of ${1} \times {1}\;{{\rm mm}^2}$ print areas—may eliminate the need for stitching and the related problems that degrade phantom quality. In addition, these phantoms are limited in height because the microchannels are sealed prior to printing in situ. However, recent work has demonstrated that it is possible to print in an unenclosed channel, removing the limit on phantom height [26].

Our demonstration of a resolution target shows that DLW has the potential to produce test methods suitable for a range of high-resolution biophotonic imaging systems. The strong signals seen at multiple wavelengths may enable applications in fluorescein angiography (peak excitation at 496 nm) and fundus autofluorescence (peak excitation at 464 nm) [27,28]. Resolution targets could be used as a performance standard over a wide range of fluorescence microscopy techniques, as the photoresist shows a wide spectrum of fluorescence excitation and emission. Hollow channel phantoms have the potential to incorporate near-infrared fluorophores such as ICG, which has also been used for retinal imaging. Such approaches could also be used to flow blood at various oxygen levels to test saturation measurement accuracy in systems based on multi- or hyper-spectral imaging.

The phantoms described here have numerous potential roles in objective, quantitative performance assessment for fluorescence microscopy and clinical retinal imaging. Specifically, these applications include assessment of spatial resolution, uniformity, and geometric measurement accuracy, all of which are essential to image quality [15]. By placing a phantom in a model eye that simulates ocular focusing, it may be possible to directly evaluate systems in clinical settings to compare performance or for single-device constancy testing. Biomimetic models can be further customized for clinical task-based evaluation and training for physicians who use retinal imaging, with the addition of pathological features such as microaneurysms characteristic of diabetic retinopathy. Future work could also include printing phantoms on a curved surface to better mimic real anatomy.

Overall, this Letter demonstrates that high-resolution additive manufacturing using DLW holds significant promise for the fabrication of microscale biophotonic phantoms. The solid phantoms hold potential as tools for fluorescence microscopy, FA, and fundus autofluorescence over a wide range of wavelengths. The hollow channel phantoms fabricated via COP-based is DLW represent, to the best of our knowledge, the only phantoms of 3D retinal vasculature to reach the true channel diameters of retinal capillaries. Specifically, the 10 µm diameters demonstrated here are nearly 10 times smaller than laser-micromachining-based methods and 43 times smaller than SLA-based 3D printing efforts [18,19]. The microscale channels in our phantoms can also be perfused with other liquids for other modalities. By achieving a higher degree of biomimetic architectural complexity, these 3D microscale phantoms can serve as a high performance standard for fluorescence imaging.