Fluorescence microscopy acquires multicolor images using fluorescence dyes that label specific structures or compounds in biological samples. Excitation and emission spectra for such dyes are often broad and multiband, which requires complex hardware or difficult decomposition methods and complicates the choice of labels to avoid spectral overlap. We propose a deep learning algorithm for channel decomposition. We trained two supervised pixel-to-pixel generative adversarial networks (GAN) to generate separate decomposed nuclear and cytoplasmic images from a single input image containing overlapping fluorescence from both dyes, achieving high similarity between ground truth and GAN images via SSIM and MSE.
Virtual pathology techniques have demonstrated the capability to overcome surgical pathology’s critical disadvantage, time inefficiency, by imaging surgical samples ex vivo. Removing extensive tissue preparation steps allows virtual pathology to provide same-day or even within-procedure diagnosis. Structured illumination microscopy (SIM) is a wide-field fluorescent optical-sectioning microscopy technique that allows higher speed imaging as compared to similar modalities, making it a good candidate for real-time diagnostics. Although SIM image acquisition is fast, the staining procedure must also be fast to meet real-time use, considering a maximum time allotment of 20-30 minutes for staining and imaging during surgery. Thus, we require our staining procedure to take no more than 4 minutes for nuclear and extracellular matrix (ECM) staining. In this work, we evaluated multiple candidates for imaging of large surgical resection specimens based on staining time and specificity, image quality, and cost per sample. Based on phantom studies using bovine muscle, we found that SYBR Gold provides the best results for nuclear staining (compared to TO-Pro3 and DRAQ5) on samples with large volume and surface area. Also, we found an optimal ratio between concentration and staining time for Eosin as an ECM stain. We then applied these protocols to imaging of radical prostatectomy specimens immediately after surgical removal, and present here for the first-time SYBR Gold and Eosin (Sg&E) virtual pathology for whole prostate imaging. In ten cases (2 with positive margin), we were able to identify healthy and pathological structures using SG&E virtual pathology that corresponds to hematoxylin and eosin pathology.
Staining procedures in virtual H&E ex vivo microscopy is non-standardized, and image quality can be tuned by systematic parameter optimization. We are using structured illumination microscopy and image processing for an objective assessment of staining procedure. Bovine muscle tiles of 2x2 cm are used as experimental tissue. In these experiments we differ staining procedure by varying stain and rinse solvent, dye concentration, and immersion time. Quantitative measures of staining quality were compared. Our findings confirm the profound impact of the confluence of multiple factors in staining quality, and suggest general strategies for successful protocol development.
On-site pathology of the surface of resected tumors provides real-time assessments of surgical procedure and improve treatment outcomes. Tissue preparation is a major bottleneck of standard pathology procedures that can be overcome by virtual pathology imaging. Here, we use optical sectioning super-resolution structured illumination microscopy (OS-SR-SIM) with a low magnification objective. Low magnification objective provides a higher field of view, and super-resolution reconstruction retrieves lost resolution due to lower NA. We successfully resolved sub-resolution fluorescent beads in highly scattering media that were not resolvable with standard OS-SIM with an NA 0.45 objective. Human tissue testing is underway.
Neovascularization in diabetic retinopathy (DR) and age-related macular degeneration (AMD) result in severe vision-loss and are two of the leading causes of blindness. The structural, metabolic, and vascular changes underlying retinal neovascularization are unknown and, thus, there is an unmet need to identify mechanisms of pathogenesis and novel anti-angiogenic therapies. Zebrafish is a robust ophthalmological model because its retina has comparable structure to the human retina and its fecundity and life-cycle enable development of mutant phenotypes of human pathologies. Here, we perform multimodal imaging with OCT and fluorescence confocal scanning laser ophthalmoscopy (cSLO) to identify changes in retinal structure and function in a zebrafish model of vascular leakage. Transgenic zebrafish with EGFP tagged plasma protein were imaged longitudinally at six time points over two weeks to visualize vascular perfusion changes from diethylaminobenzaldehyde (DEAB) treatment. Complementary contrast from OCT-A perfusion maps and cSLO imaging of plasma protein EGFP shows vascular occlusions posttreatment. cSLO images confirm presence of vessels despite loss of OCT-A signal. Plasma protein EGFP contrast also shows significant changes in vessel structure as compared to baseline images. OCT structural volumes show empty vessel cross-sections confirming non-perfusion. In addition, we present algorithms for automated biometric identification of OCT datasets using OCT-A vascular patterns in the presence of significant vascular perfusion changes. These results establish a framework for large-scale in vivo assays to identify novel anti-angiogenic compounds and understand the mechanisms ofneovascularization associated with retinal ocular pathologies.
Ophthalmic surgery involves manipulation of delicate, layered tissue structures on milli- to micrometer scales. Traditional surgical microscopes provide an inherently two-dimensional view of the surgical field with limited depth perception which precludes accurate depth-resolved visualization of these tissue layers, and limits the development of novel surgical techniques. We demonstrate multimodal swept-source spectrally encoded scanning laser ophthalmoscopy and optical coherence tomography (SS-SESLO-OCT) to address current limitations of image-guided ophthalmic microsurgery. SS-SESLO-OCT provides inherently co-registered en face and cross-sectional field-of-views (FOVs) at a line rate of 400 kHz and >2 GPix/s throughput. We show in vivo imaging of the anterior segment and retinal fundus of a healthy volunteer, and preliminary results of multi-volumetric mosaicking for ultrawide-field retinal imaging with 90° FOV. Additionally, a scan-head was rapid-prototyped with a modular architecture which enabled integration of SS-SESLO-OCT with traditional surgical microscope and slit-lamp imaging optics. Ex vivo surgical maneuvers were simulated in cadaveric porcine eyes. The system throughput enabled volumetric acquisition at 10 volumes-per-second (vps) and allowed visualization of surgical dynamics in corneal sweeps, compressions, and dissections, and retinal sweeps, compressions, and elevations. SESLO en face images enabled simple real-time co-registration with the surgical microscope FOV, and OCT cross-sections provided depth-resolved visualization of instrument-tissue interactions. Finally, we demonstrate novel augmented-reality integration with the surgical view using segmentation overlays to aid surgical guidance. SS-SESLO-OCT may benefit clinical diagnostics by enabling aiming, registration, and mosaicking; and intraoperative imaging by allowing for real-time surgical feedback, instrument tracking, and overlays of computationally extracted biomarkers of disease.
Ophthalmic diagnostic imaging using optical coherence tomography (OCT) is limited by bulk eye motions and a
fundamental trade-off between field-of-view (FOV) and sampling density. Here, we introduced a novel multi-volumetric
registration and mosaicking method using our previously described multimodal swept-source spectrally encoded scanning
laser ophthalmoscopy and OCT (SS-SESLO-OCT) system. Our SS-SESLO-OCT acquires an entire en face fundus SESLO
image simultaneously with every OCT cross-section at 200 frames-per-second. In vivo human retinal imaging was
performed in a healthy volunteer, and three volumetric datasets were acquired with the volunteer moving freely and refixating
between each acquisition. In post-processing, SESLO frames were used to estimate en face rotational and
translational motions by registering every frame in all three volumetric datasets to the first frame in the first volume. OCT
cross-sections were contrast-normalized and registered axially and rotationally across all volumes. Rotational and
translational motions calculated from SESLO frames were applied to corresponding OCT B-scans to compensate for interand
intra-B-scan bulk motions, and the three registered volumes were combined into a single interpolated multi-volumetric
mosaic. Using complementary information from SESLO and OCT over serially acquired volumes, we demonstrated multivolumetric
registration and mosaicking to recover regions of missing data resulting from blinks, saccades, and ocular
drifts. We believe our registration method can be directly applied for multi-volumetric motion compensation, averaging,
widefield mosaicking, and vascular mapping with potential applications in ophthalmic clinical diagnostics, handheld
imaging, and intraoperative guidance.
Diabetic retinopathy (DR) and age-related macular degeneration (AMD) are two of the leading causes of blindness and
visual impairment in the world. Neovascularization results in severe vision loss in DR and AMD and, thus, there is an
unmet need to identify mechanisms of pathogenesis and novel anti-angiogenic therapies. Zebrafish is a leading model
organism for studying human disease pathogenesis, and the highly conserved drug activity between zebrafish and humans
and their ability to readily absorb small molecules dissolved in water has benefited pharmaceutical discovery. Here, we
use optical coherence tomography (OCT) and OCT angiography (OCT-A) to perform noninvasive, in vivo retinal imaging
in a zebrafish model of vascular leakage. Zebrafish were treated with diethylaminobenzaldehyde (DEAB) to induce
vascular leakage and imaged with OCT and OCT-A at six time points over two weeks: baseline one day before treatment
and one, three, six, eight, and ten days post treatment. Longitudinal functional imaging showed significant vascular
response immediately after DEAB treatment. Observed vascular changes included partial or complete vascular occlusion
immediately after treatment and reperfusion during a two-week period. Increased vascular tortuosity several days post
treatment indicated remodeling, and bifurcations and collateral vessel formation were also observed. In addition,
significant treatment response variabilities were observed in the contralateral eye of the same animal. Anatomical and
functional normalization was observed in most animals by ten days post treatment. These preliminary results motivate
potential applications of OCT-A as a tool for studying pathogenesis and therapeutic screening in zebrafish models of
retinal vascular disease.
Zebrafish have been identified as an ideal model for angiogenesis because of anatomical and functional similarities with
other vertebrates. The scale and complexity of zebrafish assays are limited by the need to manually treat and serially screen
animals, and recent technological advances have focused on automation and improving throughput. Here, we use optical
coherence tomography (OCT) and OCT angiography (OCT-A) to perform noninvasive, in vivo imaging of retinal
vasculature in zebrafish. OCT-A summed voxel projections were low pass filtered and skeletonized to create an en face
vascular map prior to connectivity analysis. Vascular segmentation was referenced to the optic nerve head (ONH), which
was identified by automatically segmenting the retinal pigment epithelium boundary on the OCT structural volume. The
first vessel branch generation was identified as skeleton segments with branch points closest to the ONH, and subsequent
generations were found iteratively by expanding the search space outwards from the ONH. Biometric parameters,
including length, curvature, and branch angle of each vessel segment were calculated and grouped by branch generation.
Despite manual handling and alignment of each animal over multiple time points, we observe distinct qualitative patterns
that enable unique identification of each eye from individual animals. We believe this OCT-based retinal biometry method
can be applied for automated animal identification and handling in high-throughput organism-level pharmacological
assays and genetic screens. In addition, these extracted features may enable high-resolution quantification of longitudinal
vascular changes as a method for studying zebrafish models of retinal neovascularization and vascular remodeling.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.