Remarkable advances in imaging, computation, and technology are rapidly moving us into an era where knowledge discovery is increasingly limited only by creativity. This is creating unprecedented opportunities to improve the diagnosis and treatment of ophthalmic disease. This talk will discuss challenges and opportunities involving artificial intelligence and data science in ophthalmology research and applications to clinical care. Specific examples will be given from the speaker’s perspective as an investigator in this area and as Director of the National Eye Institute (NEI), which directs and funds vision research in the United States. This will include discussion of challenges in the accuracy and process of ophthalmic diagnosis, as well as insights and gaps in knowledge regarding artificial intelligence research in ophthalmology. It will conclude with discussion of current NEI priorities including data sharing, data harmonization, data generation, medical education in informatics and data science, methodological innovation, and population health.

We present here a study of blood oxygen saturation (sO2) in the parafoveal region from healthy subjects using visible light optical coherence tomography. Sixteen eyes of the normal subjects were analyzed. The sO2 of arterioles was significantly higher than the venules’ (92.1 ± 7.1 (vol %) for arterioles, 48.4 ± 5.0 (vol %) for venules, p<0.001), indicating that VIS-OCT can be a powerful tool to investigate the retinal sO2 in parafoveal micro-vessels in pathological conditions.

Diffuse illumination alleviates the issue of the presence of a laser hotspot at the waist of the incident beam in laser Doppler holography for ophthalmology. The image field of view can be increased because the focal point of the imaging lens doublet can be brought closer to the cornea without compromising safety. Under these conditions, compliance with European security safety limits guidelines for ophthalmic devices ISO 15004-2:2007 is guaranteed over the entire optical path of the incident beam, and no significant change in the calculated images is observed.

We have developed a novel scanning protocol based on Spectrally Extended Line Field OCTA (SELF-OCTA) to realize 4 interscan times ranging from 0.54 ms to 5.4 ms without increasing A-scan repetitions at the same position. Based on the linear model, we further developed signal processing methods to realize high dynamic range blood flow map by concatenating the dynamic range of 3 interscan time: 0.54 ms, 1.35 ms and 2.70 ms. The differential OCTA signals provided by different interscan times help with flow dynamic range extension and blood flow evaluation.

Although elevated intraocular pressure (IOP) is considered to be a major precursor for glaucoma, up to 45% of the patients with early glaucoma show signs of disease progression despite IOP reduction therapy. Studies have shown strong clinical evidence for abnormal ocular vessel function and impaired autoregulation of blood flow in early glaucoma subjects and its role in disease development and progression. Here we present direct measure of vascular dysfunction and autoregulation in three healthy human subjects using the erythrocyte mediate angiography and adaptive optics scanning laser ophthalmoscopy line-scan techniques. These novel quantitative blood flow metrics can potentially serve as a sensitive biomarker for early diagnosis and monitoring of ocular disease.

OCT-based angiography (OCTA) enables imaging of retinal microvasculature. However, involuntary movements of subjects prevent obtaining microvasculature with a large field of view (FOV) with high-dense spatial sampling. Meanwhile, a short duration for blood flow detection results in unstable vasculature contrast. By combining a Lissajous scan and a slow shift, Lissajous OCT enables compensating eye movements, and then the FOV will be extended while the spatial sampling density is preserved by the slow shift. The Lissajous scan allows a long blood flow detection duration. In vivo human eye microvasculature imaging using the convolutional Lissajous OCTA is investigated with several disease eyes.

We propose a robust deep learning algorithm for denoising TD FF-OCT in vivo images. This algorithm does not require any clean images in its training. It specifically detects and removes residual fringes as well as other types of noise present in in vivo eye images. It can also be trained using ex vivo images as well as simulated patterns for fringe removal. Testing is performed on in vivo corneal images, but can be expanded to any TD FF-OCT images. The obtained outputs are thus easier to interpret and exploit in clinical practice as well as other image processing tasks.

In this study, we developed voxel-wise projection-resolved optical coherence tomographic angiography (PR-OCTA) using artificial intelligence (AI). Two different artificial intelligence models were developed, including a pure convolutional neural network (CNN) model and a CNN and recurrent neural network (RNN) hybrid model. Compared with the state-of-art rules-based model, the AI models were able to preserve more in-situ blood flow and suppress projection artifacts and background noise.

We present results from measurements of retina response to a chirped frequency flickering light. Such an approach facilitates faster characterization of photoreceptor response amplitude in the function of stimulus frequency in comparison to separate measurements at different frequencies. In our work, we compare responses to stimuli of various types (e.g., different flicker amplitudes) in the frequency range from 5 Hz to 45 Hz.

In recent years, there have been vast interests in probing photoreceptor dynamics using optical coherence tomography (OCT). Most of the successful demonstrations implemented adaptive optics or digital adaptive optics to resolve individual cones or rods in human subjects. Here we use phase information to trace the photoreceptor response in rodents using an ultrahigh-resolution, phase-sensitive, spectral-domain OCT. As a result, two types of nanoscale signals (monophasic and biphasic) were detected with a clear separation in depth. The monophasic signal is less susceptible to stimulus intensity and saturated from a 3% breach rate.

Two-photon vision is a newly discovered mechanism of the perception of pulsed near-infrared laser beams as a color stimulus corresponding to approximately half of the laser wavelength (PNAS 111(50), E5445–E5454). Based on this phenomenon, a new visual field test instrumentation – two-photon microperimetry, has been developed (BOE 10(9), 4551–4567). This study shows that two-photon perimetry gives more reproducible results than one-photon perimetry for standard threshold finding strategies. This unquestionable advantage of the nonlinear vision-based visual field testing technique may benefit the clinical assessment of retinal disease progression and treatment efficiency.

Measurement of visual function plays a critical role in assessing the health of the retina. Optoretinography (ORG) is an emerging technique for noninvasive measurement of retinal neural function. Recent efforts have demonstrated the feasibility of the ORG using advanced OCT systems which track single cells in the retina and measure the stimulus-evoked movement of their subcellular features. Here we demonstrate a novel velocity-based approach in three healthy subjects. The resulting responses were reproducible, exhibited expected dependence on dose and retinal eccentricity, and could be related to earlier position-based methods through numerical integration.

Phase-sensitive OCT can be used for imaging the photoreceptor deformations in response to the light stimulus or optoretinography (ORG). Here, we propose a phase-restoring subpixel image registration method and an automated signal extraction algorithm for optoretinography using phase-sensitive OCTs. We validated these methods in simulations, phantom experiments, and in-vivo optoretinogram imaging. Our image registration method yields better amplitude stability and higher phase accuracy compared with conventional approaches, and we found two types of signals (one monophasic and the other biphasic) simultaneously in rodent ORG imaging. These results can be beneficial to the ongoing preclinical/clinical ORG studies.

Optoretinography serves as an effective biomarker for light-evoked retinal activity owing to its sensitive, objective, and precise localization of retinal function and dysfunction. We developed a coarse-scale optoretinography instrument based on a line-scan spectral domain OCT. We demonstrate its ability to acquire robust and repeatable ORG signals rapidly over a 5° field of view in a single acquisition, without adaptive optics. The high repeatability, good agreement with cellular-scale ORGs, and non-AO operation are of promise for its widespread clinical application to retinal diseases such as AMD and inherited retinal degenerations.

In this work, we implement a pupil conjugated DM-based wavefront sensorless approach in our FFOCT setup. Given the high DM stroke and precision, as well as the Zernike mode-based wavefront optimization using the high-speed DONE algorithm, we showed that higher resolution can be achieved in foveal imaging, as well as obtaining an improved SNR when imaging photoreceptors and NFL when compared to our previous work. Thanks to this SNR improvement, we were able to visualize inner retinal features not previously observed in FFOCT.

The photoreceptor (PR) – retinal pigment epithelium (RPE) – choriocapillaris (CC) complex is an extremely important group of layers in the outer retina. We demonstrate resolution of the CC vascular network across the macula, as well as the methodology to extract and quantify structural metrics from all three layers from averaged AO-OCT volumes. In diseased eyes, small changes in CC structure may portend the initiation of disease and therefore the investigation of CC structural changes may aid early disease diagnosis for many diseases, both prevalent and rare, that begin in the outer retina.

Microscope-integrated optical coherence tomography (MIOCT) is an emerging multimodal imaging technology in which live volumetric OCT (“4D-OCT”) is displayed simultaneous with standard stereo color microscopy. 4D-OCT provides ophthalmic surgeons with many visual cues not available in microscopy, but it cannot serve as a replacement due to lack of color features. In this work, we demonstrate progress toward a unified solution by fusion of data from both modalities, guided by segmented 3D features, yielding a more efficient visualization combining important cues from both modalities.

Intraoperative OCT (iOCT) has been shown to help guide ophthalmic surgical management by allowing for verification of the completion of surgical objectives. However, iOCT systems are limited by static OCT fields-of-view (FOVs) that require manual alignment to surgical regions-of-interest, preventing real-time volumetric imaging of surgical maneuvers. Here, we demonstrate an automated instrument-tracking method that leverages multimodal ophthalmic imaging and deep-learning-based object detection for localization of multiple surgical instruments within the OCT FOV. We present video-rate 4D imaging of mock surgical maneuvers in ex vivo porcine eyes at 16 volumes/second, which can be used to provide real-time feedback on instrument-tissue dynamics.

OCT angiography (OCTA) is an extension of optical coherence tomography (OCT) that identifies motion contrast from moving red blood cells to map retinal vasculature in vivo. We propose to use robotically-aligned optical coherence tomography (RAOCT) to acquire OCTA data at multiple illumination angles on the retina in order to reduce shadowing artifacts and enhance vessel visualization. Using RAOCT, retinal volumes were automatically acquired from consented subjects at various pupil entry positions and processed to generate en face OCTA images. These OCTA images were compared to identify areas of changed visualization and reduced shadowing artifact when varying illumination angle.

The choroid appears to play a role in the regulation of eye growth and may respond to defocus cues. We developed a combined OCT and video optotype stimulus system wherein the optotype may be adjusted for defocus and incidence angle at the retina. We imaged an initial single subject with OCT before and after watching a 40-minute video under two test conditions: no defocus and +5D defocus. This system offers the potential to investigate choroidal changes over the entire macula and under multiple stimulus conditions.

We present a flexible optical coherence tomography (OCT) imaging platform that allows for interchangeable imaging modules for specific target tissues of interest while meeting the requirements for robotically aligned OCT including integrated pupil tracking cameras. Our OCT imaging platform consisted of a fixed scan head (analogous to an SLR camera body) mounted to the robot and interchangeable anterior chamber (AC) and retinal imaging modules with their own integrated pupil cameras. We validated our system in both phantom and ex vivo porcine eyes. This flexible design enables the ability to develop new imaging modules for new robotically aligned applications.

Corneal biomechanical weakening presumably precedes keratoconus (KC), an ocular disease that leads to vision loss. Cross-meridian swept-source OCT coupled to air-puff excitation was used to induce corneal deformation to investigate biomechanics in Forme Fruste (FF)/subclinical (n=10), KC I (n=10) and healthy (n=12) eyes. Shape and asymmetry deformation parameters were analyzed in two meridians, and the tangent modulus was calculated using Finite Element modeling (FEM). Compared to healthy eyes, the asymmetry parameter decreased 0.32±0.05% (FF/subclinical), and 0.66±0.18% (KC I). The shape parameter increased 0.91±0.32% (FF/subclinical) and 1.47±1.2% (KC I). Significant differences between groups were observed mostly on the vertical meridian. Inverse FEM showed ∼30% localized stiffness reduction in KC eyes, compared to healthy eyes. Our results show that the additional vertical meridian allows more significant use of deformation parameters as biomarkers of biomechanical changes.

Presbyopia is a loss of the dynamic accommodation response of our vision and affects everybody as they age. Despite many static treatment options, we still do not address the biomechanical cause of lens stiffness. Novel therapies are limited by no ability to monitor in vivo biomechanics. To address this unmet need, we developed a combination OCE/Brillouin system capable of measuring co-located Brillouin spectra and elastography information to derive depth-dependent elastic moduli. The system specifications were quantified for evaluating a human lens and testing performed in vivo. This technique has the potential for patient-specific predictive models of presbyopia.

The detection of subclinical keratoconus in human corneas remains a challenging task. We propose to use a wave-based air-coupled ultrasonic optical coherence elastography system to map the elasticity of the corneas of 15 patients with a clinical diagnosis of keratoconus (KC) in one eye, and subclinical keratoconus (SK) in the fellow eye. Two biomarkers are proposed: Spatial Anisotropy of Wave Speed (SAWS), and the Speed-Thickness Index (STI). Our results show important biomechanical differences between normal, subclinical, and advanced stages of keratoconus, suggesting SAWS and STI as potential biomarkers to identify “at-risk” corneas before changes in topography and pachymetry become evident.

This conference presentation was prepared for the Ophthalmic Technologies XXXIII conference at SPIE BiOS, SPIE Photonics West 2023.

Mucin secretive conjunctival goblet cells (CGCs) in the eye play important roles in ocular surface homeostasis by forming the mucous layer of the tear film. CGC information is also an important biomarker for diagnosis because CGC loss or dysfunction is observed in various ocular surface diseases. In this study, we developed moxifloxacin-based extended depth-of-field (EDOF) microscopy with surface tracking for non-invasive CGC imaging in awake human subjects. The system had a DOF of 0.8 mm, a field of view (FOV) of 1.3mm x 1.3mm, and imaging speed of 15 fps. The phase detection method was used for real-time surface tracking. Moxifloxacin ophthalmic solution was topically instilled for CGC labeling. Repeated large area imaging of the same conjunctiva in a human subject was demonstrated. MBFM might have the potential for non-invasive CGC examination in patients.

The Oxford nomenclature was developed to distinguish the type of cataract based on the optical signal discontinuity (OSD) zones. The OSD zones were used to distinguish sections within the crystalline lens in our study using OCT images. A swept-source OCT was used to observe in-vivo age-related changes in the crystalline lens from the 50 healthy eyes with the age range of 9 to 78 years. The C3 layer (from oxford nomenclature) of the cortex was identified as the section of the crystalline that contributes highly to the age-related changes. We compared age-related degradation of the optical quality of the crystalline lens measurements using commercial VAO and OQAS systems with a custom-built SS-OCT system.

The signal-to-noise ratio of an optical coherence tomography system is inversely proportional to the spectral bandwidth of its input light source. In this work, we have modified a conventional 1 μm swept-source retinal imaging system to have two input laser states to exploit this trade-off. The states have constant average power but are concentrated over two wavelength ranges - one with a 75 nm bandwidth (optimized for high-resolution imaging of the retina) and a second with a 25 nm bandwidth (favoring increased penetration depth in the choroid, sclera and lamina cribrosa.) A fusion of the data from these two modes provides enhanced images of the whole posterior eye.

Temperature measurement is demanded for retinal laser treatment with pulse durations of 5µs and longer to take pigmentation variations of the retinal pigment epithelium and choroid into account. An optoacoustic technique to determine the maximum temperature at the end of an irradiation with a single microsecond laser pulse was investigated. A 15W diode laser with 514nm and variable pulse duration was used to irradiate porcine RPE. Gradients of the laser pulse flanks >10W/µs excite pressure transients. Optoacoustic temperature measurements were possible. Cell damage was found at maximum temperatures around 80°C at pulse durations of 35µs and to 50µs.

We report on a novel mice imaging system based on the Spatio-Temporal Optical Coherence Tomography (STOC-T) technique. The contribution describes the translation of the STOC-T technique, initially developed for human eye imaging, into the field of experimental small animal imaging. We present images of retinal and choroidal tissue from a B6 albino wild type mouse acquired at a volumetric rate of 112 Hz.

Direct access to retinal coordinates and fine spatial and temporal resolution are clear advantages of the retinal tracking methods over widely used video-based eye trackers. Several approaches to the retinal eye tracking have been proposed so far, including frame-to-frame scanning laser ophthalmoscope (SLO) tracking, reference-frame based methods using sub-sampled frames or our MEMS-scanner-based tracker capable of measuring both fixational and saccadic eye movements. In this work we present a novel approach to the design of the retinal MEMS-based tracking system, taking advantage of two combined high-speed MEMS mirrors to form a Lissajous scanning pattern with adjustable density and framerate.

Simultaneous multi-channels offset imaging provides isotropic images of retinal microstructures free of directionality artifacts and enables phase imaging in the living eye with enhanced visualization contrast. Phase imaging is widely used in microscopy and biomedical imaging to reveal structures not visible in standard imaging due to their low scattering properties. We introduce the technique in high-resolution retinal imaging to visualize microstructures in the living eye with enhanced contrast, that were not visible with other modalities, and without adding contrast agents. Phase imaging has the potential to be quantitative, may have diagnostic value for retinal diseases, and may enable monitoring treatment.

Most point-scanning retinal imaging modalities use raster scan patterns, which can appear in the subject’s vision as a rapidly moving line and cause subject eye motion, resulting in motion artifacts that require a dedicated fixation target to mitigate. In our recent development of a spiral-scanning confocal scanning laser ophthalmoscope, we noticed that the spiral scan pattern is visible to the subject as a bullseye-like shape and hypothesized that it could function as a virtual fixation target. A pilot study confirms this hypothesis by showing that images acquired with spiral scan show less eye motion than images acquired with raster scan.

Here we present a new approach to quantify macular pigments and importantly, localize them in depth within the human retina in vivo. The approach utilizes visible light Optical Coherence Tomography (OCT) imaging with multiple combined superluminescent diodes, with energy concentrated at discrete red, green, and blue wavelength bands. Imaging simultaneously with red and blue wavelengths, we reveal the expected distribution of macular pigment optical density with a peak at or near the foveal center. Imaging simultaneously with red and green wavelengths, we localize macular pigments in depth to the region beneath the foveal pit, inner to the photoreceptors.

Longitudinal chromatic aberration (LCA) and transverse chromatic aberration (TCA) adversely affect retinal image quality. Thus, one expects improved visual performance when chromatic aberrations are minimized or eliminated. Systematic evaluation of the impact of LCA and/or TCA correction under broadband illumination is needed. Thus, we developed a system, called the Binocular Varichroma and Accommodation Measurement System (BVAMS) that can be used to measure and correct the eye’s LCA and TCA and to perform vision tests with custom corrections. We demonstrate a measurable benefit in visual acuity only with both LCA and TCA correction.

In this study we present a desktop swept-source optical coherence tomographic angiography (OCTA) system for a diverse range of ophthalmic applications. The system can achieve 500 kHz ultrafast imaging with a 75-degree field of view, assisted by a GPU-accelerated real-time cross-sectional and en face OCT/OCTA acquisition interface, a self-tracking method for motion correction, and an AI-based retinal layer segmentation algorithm. High-resolution and high-sensitivity OCTA images from different retinal disease types are demonstrated.

Monitoring the tissue temperature is highly desirable for predictable and reproducible retinal laser therapy. We demonstrate that the temperature rise can be determined using pOCT imaging of the optical path length changes within the laser spot relative to the non-heated region. A temperature-sensitive fluorescent dye was used for initial calibration. By matching the thermal expansion across the beam width and along the full course of heating and cooling, the temperature distribution can be determined with a precision of about 10% (under 2℃ with a peak heating of 17℃) following a single laser pulse of 20ms in duration.

Alzheimer’s Disease (AD) is a degenerative neurological condition. Here, we use polarization-sensitive optical coherence tomography to characterize APPSL and 5xFAD mouse models for AD research. It was observed at approximately one year of age, both control and transgenic mice had developed similar retinal features, but AD models demonstrated significantly higher counts of hyperscattering foci compared to controls and some 5xFAD mice demonstrated a thickening of the photoreceptor layers. This study identifies potential AD biomarkers and reemphasizes the importance of using proper age-matched controls for animal studies, particularly when studying age-dependent diseases such as AD.

Our research investigates retinal pigmentary abnormalities in retinitis pigmentosa (RP) patients using a wide-field high-speed polarization diversity optical coherence tomography (PD-OCT) in a clinical setting. To account for the retinal curvature in the wide field-of-view, adaptive kernel-based spatial averaging is employed for degree-of-polarization-uniformity (DOPU) contrast formation with two complex OCT signals from two orthogonal polarization channels. In 7 patients diagnosed with RP, retinal pigment epithelium (RPE) melanin loss centered at the macula is compared to standard multimodal imaging techniques, including intensity-based OCT, fundus photography, and short-wavelength fundus autofluorescence images.

Polarization-sensitive optical coherence tomography (PS-OCT) is used to investigate changes in fiber structures in the eyes of myopia, wet age-related macular degeneration (wAMD), and glaucoma patients. A depth-multiplexed fiber-based PS-OCT system is used to extract local optic axis orientations of birefringent structures in the eye locally in 3D in vivo. For healthy volunteers, birefringent structures include the retinal nerves, Henle’s fiber layer, and sclera. Scleral collagen architecture is observed to be altered in myopia patients. In wAMD patients fibrosis orientation and volume can be extracted using this technique.

This conference presentation was prepared for the Ophthalmic Technologies XXXIII conference at SPIE BiOS, SPIE Photonics West 2023.

This conference presentation was prepared for the Ophthalmic Technologies XXXIII conference at SPIE BiOS, SPIE Photonics West 2023.

This conference presentation was prepared for the Ophthalmic Technologies XXXIII conference at SPIE BiOS, SPIE Photonics West 2023.

This conference presentation was prepared for the Ophthalmic Technologies XXXIII conference at SPIE BiOS, SPIE Photonics West 2023.

This conference presentation was prepared for the Ophthalmic Technologies XXXIII conference at SPIE BiOS, SPIE Photonics West 2023.

This conference presentation was prepared for the Ophthalmic Technologies XXXIII conference at SPIE BiOS, SPIE Photonics West 2023.

Current publications show promising results in the in-vivo detection of amyloid deposits in the retina of Alzheimer’s disease (AD) patients as well in post-mortem flat mounted retinal tissue. The results are promising for the detection of early alterations associated with AD. The aim of our study was to confirm recently published findings using almost identical methodology, blue (ex: λ = 486 nm) fluorescence retinal imaging, curcumin as labelling fluorophore, and a similar data analyzing process.

This conference presentation was prepared for the Ophthalmic Technologies XXXIII conference at SPIE BiOS, SPIE Photonics West 2023.

This conference presentation was prepared for the Ophthalmic Technologies XXXIII conference at SPIE BiOS, SPIE Photonics West 2023.