2.1.

DF-HRME System Design

The optical design of DF-HRME is shown in Fig. 1(a). To achieve scanning darkfield imaging without complex optomechanical components, we utilize structured illumination provided by a digital light projector (DLP) (LightCrafter 4500, Texas Instruments, Dallas, Texas) and digital line-scanning detection enabled by the rolling shutter of a complementary metal-oxide semiconductor (CMOS) image sensor (FL3-U3-120S3C-C, Teledyne FLIR, Wilsonville, Oregon). The spatiotemporal design of illumination and detection aperture sequences is shown in Fig. 1(b). For non-scanning widefield imaging, no illumination pattern is used and the entire field of view (FOV) is illuminated. The detection aperture collects the specular reflection generated at fiber surfaces, which significantly reduces contrast and image quality. When performing scanning darkfield imaging, a series of line pairs with different spatial locations are programmed as illumination patterns and projected sequentially from the DLP. A spatiotemperal offset between illumination and detection (rolling shutter) is ensured by adjusting and synchronizing the aperture sequence. Compared to non-scanning widefield imaging, scanning darkfield imaging avoids the spatial overlap between illumination and detection. Direct specular reflection is effectively rejected and only scattered signals are captured, which substantially improves contrast and the visibility of tissue features.

Fig. 1

DF-HRME implements scanning darkfield imaging to reduce internal reflection and enhance image contrast. (a) Optical schematic of DF-HRME. To enable scanning darkfield imaging, a spatiotemporal offset is introduced between illumination and detection by programming and synchronizing the DLP illumination and CMOS camera rolling shutter. Arrows show the scanning direction. A thin, flexible fiber bundle is placed in contact with tissue epithelium stained by MB for real-time imaging of cell nuclei. The dashed square indicates the optical system enclosure. (b) Spatiotemporal design of illumination and detection aperture sequences. Overlapped illumination and detection aperture in non-scanning widefield imaging captures the strong specular reflection generated at fiber surfaces, degrading image quality. In scanning darkfield imaging, a sequence of illumination line pair patterns is designed to avoid overlap with the detection aperture and suppress specular reflection background. DF-HRME, scanning darkfield high-resolution microendoscope; DLP, digital light projector; CMOS, complementary metal-oxide semiconductor camera; BF, bandpass filter; LP, linear polarizer; L1, collimating lens; L2, tube lens; PBS, polarizing beamsplitter; M1, M2, mirrors; OBJ, microscope objective; WF, non-scanning widefield imaging; DF, scanning darkfield imaging.

In the previous study, the green LED of the DLP and a green light bandpass filter centered at 525 nm were used in DF-HRME for label-free imaging of microvasculature. Hemoglobin absorption provided the endogenous contrast for microvessels.²³ To extend the capability of DF-HRME for imaging cell nuclei with MB staining, we replaced the green light filter with a bandpass filter centered at 605 nm (605/64 nm, FF01-605/64-25, AVR Optics) and enabled both green and red LEDs of the DLP for illumination. The wavelength is selected to match the absorption peak of MB and enhance nuclear contrast. DF-HRME achieves a seamless transition between these two imaging applications via a simple filter change and software control. Through a collimating lens (AC254-100-A, Thorlabs) and a 10× objective (#86-818, Edmund Optics), projected illumination patterns are focused onto the proximal end of a thin and flexible fiber-optic imaging probe ($790\mu \mathrm{m}$ circular FOV; FIGH-30-850N, Myriad Fiber Imaging). The fiber probe, which is attached externally to the DF-HRME, relays the structured illumination onto the tissue epithelium stained by MB and collects scattered darkfield signals. Real-time darkfield images of tissue are focused on the image sensor through the objective and a tube lens (AC254-100-A, Thorlabs). We used a linear polarizer (LPVISE100-A, Thorlabs) and a polarizing beamsplitter (CCM1-PBS251, Thorlabs), as well as background subtraction, to further suppress residual background signal. A graphical user interface (GUI) is programmed for software control and real-time data are displayed at a video frame rate (seven frames per second). The DF-HRME was assembled into a portable encasement (dashed square) to facilitate application in different clinical settings and the total cost of goods is less than $5500.

2.2.

Ex Vivo Imaging of Animal Specimens

We performed ex vivo imaging on fresh porcine tongue specimens obtained from a local abattoir to evaluate the performance of DF-HRME imaging with MB staining for characterization of nuclear morphology. We compared images obtained in three modes: scanning darkfield reflectance imaging with MB staining, non-scanning widefield reflectance imaging with MB staining, and non-scanning widefield fluorescence imaging with proflavine staining. For a direct comparison of all three modalities, the distal end of fiber probe was controlled using a translation stage so that the same FOV was imaged with each modality. We first topically applied proflavine solution (0.01% w/v in phosphate-buffered saline) to the specimen using a cotton-tipped applicator and performed imaging using a fluorescence microendoscopy system that has been previously described.²⁴ The distal end of fiber probe was affixed to a translation stage, allowing it to be translated away from the tissue surface in the vertical direction with a fixed lateral position after fluorescence imaging. This created space to wash out proflavine and stain the same tissue area with MB solution (0.3% w/v in deionized water, Biopharm). We performed DF-HRME imaging immediately after the topical administration of MB solution. We placed the distal end of the fiber in contact with MB-stained tissue epithelium and connected the proximal end of fiber to the DF-HRME for reflectance imaging. When imaging with DF-HRME, we programmed the system to record successive image frames from the same FOV using scanning darkfield imaging and non-scanning widefield imaging.

Microendoscopy images were processed to improve the visibility of nuclear morphology; we first applied a Gaussian filter to remove the intrinsic honeycomb pattern of the fiber bundle in microendoscopy images and then linearly adjusted the brightness of the images to a consistent level. Image contrast was enhanced using adaptive histogram equalization. The same image processing procedure was applied to all fluorescence, widefield reflectance, and darkfield reflectance images.

2.3.

In Vivo Imaging of Healthy Volunteers

To further demonstrate the potential of DF-HRME imaging in combination with MB staining for in vivo microscopic evaluation of nuclear features, we performed scanning darkfield imaging in the oral cavity of healthy adult volunteers at Rice University. The study was approved by the Institutional Review Board of Rice University. Prior to each imaging experiment, written informed consent was obtained from the participant, and the imaging probe underwent standard high-level disinfection procedure. Tissue areas of interest were stained by 0.3% (w/v) MB solution using cotton-tipped applicators. We performed DF-HRME imaging immediately after the topical administration of MB solution. The fiber probe was placed in gentle contact with different stained tissue sites in the oral cavity. Microscopic images of epithelial cell nuclei were displayed on the GUI in real time and saved for further evaluation. DF-HRME imaging was conducted for 10 to 15 min to examine tissue areas of interest. The quality of MB staining and DF-HRME images was consistent during the imaging experiment. Obtained in vivo images were processed using the same image processing procedure described in Sec. 2.2 to improve the image quality.

3.1.

Ex Vivo Imaging of Animal Specimens

Figure 2 shows images acquired at representative sites of porcine tongue specimens using non-scanning widefield reflectance imaging and scanning darkfield reflectance imaging when staining with MB, as well as fluorescence imaging with proflavine staining. Due to optical absorption, cell nuclei stained with MB appear as dark dots when using reflectance imaging. Strong reflection background is observed in widefield images, which compromises the visibility of stained cell nuclei. Only a few cell nuclei are visualized, and it is difficult to discern morphology. In comparison, scanning darkfield imaging consistently improves image contrast and signal-to-noise ratio by effectively suppressing specular reflection. Darkfield images captured at the same tissue sites reveal the size, shape, and distribution of epithelial cell nuclei with high spatial resolution throughout the entire FOV.

Fig. 2

Comparison of representative images acquired from porcine tongue specimens using non-scanning widefield imaging, scanning darkfield imaging, and fluorescence imaging. The image contrast of MB stained cell nuclei is significantly improved by darkfield imaging compared to widefield imaging. Cell nuclei are clearly resolved with comparable image quality in darkfield images and corresponding fluorescence images. (Scale bar: $100\mu \mathrm{m}$)

As shown in Fig. 2, darkfield imaging with MB staining has comparable ability to image nuclear morphology as compared to fluorescence imaging with proflavine staining. The appearance of cell nuclei in darkfield images differs slightly from that in the corresponding fluorescence images of the same sites. The observed difference can be attributed to the rotation of the fiber proximal end and the intermediate tissue processing steps when imaging the same tissue region with these two imaging modalities. Nevertheless, similar nuclear patterns can be discerned with high spatial resolution and image contrast by both imaging modalities. We calculated the number of nuclei in 25 regions of interest with identical size from all collected image data. On average, darkfield images reveal similar nuclear count ($91.4\pm 11.4$) as fluorescence images ($92.0\pm 11.0$). These results demonstrate that DF-HRME imaging with MB staining is an effective approach for real-time microscopic assessment of nuclear morphometry.

3.2.

In Vivo Imaging of Healthy Volunteers

Figure 3 shows representative DF-HRME images obtained from various MB-stained sites within the oral cavity of healthy normal volunteers. Stained cell nuclei with uniform size and shape are readily distinguishable in darkfield images. Each individual cell nucleus is well-isolated, and nuclei are regularly spaced in all in vivo images. These results indicate that DF-HRME imaging in combination with MB staining can reveal epithelial nuclear features with high spatial resolution and high contrast on healthy volunteers. This approach has the potential to enable a comprehensive evaluation of nuclear atypia associated with malignant progression in suspicious lesions, such as increased nuclear size, increased nuclear density, and nuclear pleomorphism.

Fig. 3

Representative DF-HRME images acquired in the oral cavity of healthy volunteers. The morphology of cell nuclei stained by MB is clearly resolved by in vivo DF-HRME imaging. (Scale bar: $100\mu \mathrm{m}$)