Assessing the efficacy of 2D and 3D CNN algorithms in OCT-based glaucoma detection.

Glaucoma is a progressive neurodegenerative disease characterized by the gradual degeneration of retinal ganglion cells, leading to irreversible blindness worldwide. Therefore, timely and accurate diagnosis of glaucoma is crucial, enabling early intervention and facilitating effective disease management to mitigate further vision deterioration. The advent of optical coherence tomography (OCT) has marked a transformative era in ophthalmology, offering detailed visualization of the macula and optic nerve head (ONH) regions. In recent years, both 2D and 3D convolutional neural network (CNN) algorithms have been applied to OCT image analysis. While 2D CNNs rely on post-prediction aggregation of all B-scans within OCT volumes, 3D CNNs allow for direct glaucoma prediction from the OCT data. However, in the absence of extensively pre-trained 3D models, the comparative efficacy of 2D and 3D-CNN algorithms in detecting glaucoma from volumetric OCT images remains unclear. Therefore, this study explores the efficacy of glaucoma detection through volumetric OCT images using select state-of-the-art (SOTA) 2D-CNN models, 3D adaptations of these 2D-CNN models with specific weight transfer techniques, and a custom 5-layer 3D-CNN-Encoder algorithm. The performance across two distinct datasets is evaluated, each focusing on the macula and the ONH, to provide a comprehensive understanding of the models' capabilities in identifying glaucoma. Our findings demonstrate that the 2D-CNN algorithm consistently provided robust results compared to their 3D counterparts tested in this study for glaucoma detection, achieving AUC values of 0.960 and 0.943 for the macular and ONH OCT test images, respectively. Given the scarcity of pre-trained 3D models trained on extensive datasets, this comparative analysis underscores the overall utility of 2D and 3D-CNN algorithms in advancing glaucoma diagnostic systems in ophthalmology and highlights the potential of 2D algorithms for volumetric OCT image-based glaucoma detection.

High-speed, phase contrast retinal and blood flow imaging using an adaptive optics partially confocal multi-line ophthalmoscope.

High-speed, phase contrast retinal and blood flow imaging using an adaptive optics partially confocal multi-line ophthalmosocope (AO-pcMLO) is described. It allows for simultaneous confocal and phase contrast imaging with various directional multi-line illumination by using a single 2D camera and a digital micromirror device (DMD). Both vertical and horizontal line illumination directions were tested, for photoreceptor and vascular imaging. The phase contrast imaging provided improved visualization of retinal structures such as cone inner segments, vessel walls and red blood cells with images being acquired at frame rates up to 500 Hz. Blood flow velocities of small vessels (<40 µm in diameter) were measured using kymographs for capillaries and cross-correlation between subsequent images for arterioles or venules. Cardiac-related pulsatile patterns were observed with normal resting heart-beat rate, and instantaneous blood flow velocities from 0.7 to 20 mm/s were measured.

Quantitative Analysis of the Vasculature and Cone Photoreceptors in Subjects With Diabetes Without Diabetic Retinopathy.

PURPOSE: To characterize any differences in the vasculature and cone photoreceptor packing geometry (CPG) between subjects with diabetes without/no diabetic retinopathy (NDR) and healthy controls. METHODS: Eight NDR and five controls were enrolled. Optical coherence tomography angiography (OCTA) taken at the macula was used to measure vessel density, vessel length density, and vessel density index (VDI) in three vascular plexuses, namely, the superficial vascular plexus, intermediate capillary plexus, and deep capillary plexus (DCP). The choriocapillaris (CC) flow deficit (FD) was also measured. OCTA images were binarized and processed to extrapolate the parafovea and parafoveal quadrants and the OCTA indices mentioned above. The CC was processed with six different radii to quantify FD. Adaptive optics - scanning laser ophthalmoscopy images were acquired and processed to extract CPG indices, i.e., cone density (CD), cone-to-cone spacing (CS), linear dispersion index, heterogeneity packing index and percent of cells with six neighbors at 3.6° in the temporal retina. RESULTS: In all eyes, statistically significant differences were found (i) in parafoveal FD across the six radii (p < 0.001) and (ii) in the correlation between the parafoveal temporal quadrant (PTQ) DCP VDI and CS (r = 0.606, p = 0.048). No other significant correlations were found. For OCTA or CPG indices, no significant differences were found between the cohorts in the parafovea or parafoveal quadrants. CONCLUSIONS: CS is the most sensitive CPG index for detecting alterations in the cone mosaic. The DCP and the cone photoreceptors are significantly correlated, indicating that alterations in the DCP can affect the cones. Future work elucidating the vascular alterations and neurodegeneration present in diabetic eyes should focus on the DCP and multiple CPG indices, not solely CD. Moreover, such alterations are highly localized, hence using larger regions e.g. parafovea versus smaller areas, such as the PTQ, will potentially mask significant correlations.

Foveal cone loss in tamoxifen maculopathy: a case report.

BACKGROUND: Tamoxifen is used in low dose concentrations (20-40 mg per day) as a therapy for breast cancer but is known to have ocular side effects. In this case report, the foveal cone integrity in a tamoxifen-treated patient who complained of a small central scotoma in the left eye while reading was examined using high resolution adaptive optics imaging. CASE PRESENTATION: Both eyes of a 54-year-old Caucasian, non-hispanic female who had been treated with tamoxifen for 1.5 years were examined using various imaging modalities including fundus photography, fundus autofluorescence, fluorescein angiography, spectral-domain optical coherence tomography, and adaptive optics scanning laser ophthalmoscopy. Clinical spectral-domain optical coherence tomography showed a very small disruption to the photoreceptor layer at the fovea in the left eye only. However, adaptive optics scanning laser ophthalmoscopy imaging revealed foveal cone loss in both eyes, but to a lesser extent in the right eye. Inner retinal changes were not observed in either eye. CONCLUSION: The area of cone loss was similar in size to a single newsprint letter when projected onto the retina, matching the patient's description of a scotoma in the left eye. Given the isolated loss of foveal cone photoreceptors with the absence of previously reported inner retinal and vascular changes, our results may indicate the earliest retinal changes associated with tamoxifen retinopathy.

High Resolution Imaging of the Outer Retina in Type 2 Acute Macular Neuroretinopathy.

PURPOSE: To investigate the outer retinal changes in a patient with type 2 acute macular neuroretinopathy (AMN). METHODS: A 35-year-old Caucasian female complaining of a unilateral blind spot was imaged using various retinal imaging modalities including clinical optical coherence tomography (OCT), OCT-angiography, fundus fluorescein angiography and adaptive optics (AO). RESULTS: Fundus examination revealed multiple paracentral reddish-brown petaloid lesions in the symptomatic left eye, while the other eye was unremarkable. Clinical OCT showed areas of hyper-reflectance at the outer plexiform layer / outer nuclear layer complex with a disrupted inner /outer segment junction, which are characteristic features of type 2 AMN. AO imaging further revealed either shortening or absence of cone outer segments within the AMN lesions attributing to the darker features observed in the enface images from fundus photography and scanning laser ophthalmoscopy. CONCLUSION: The AO findings indicate that the petaloid lesions in type 2 AMN are caused by a combination of the shortening and absence of the outer segment in individual cone photoreceptors.

Programmable, high-speed, adaptive optics partially confocal multi-spot ophthalmoscope using a digital micromirror device.

A high-speed, adaptive optics partially confocal multi-spot ophthalmoscope (AO-pcMSO) using a digital micromirror device (DMD) in the illumination channel and a fast 2D CMOS camera is described. The camera is synchronized with the DMD allowing projection of multiple, simultaneous AO-corrected spots onto the human retina. Spatial filtering on each raw retinal image before reconstruction works as an array virtual pinholes. A frame acquisition rate of 250 fps is achieved by applying this parallel projection scheme. The contrast improves by 2-3 fold when compared to a standard flood illumination architecture. Partially confocal images of the human retina show cone and rod photoreceptors over a range of retinal eccentricities.

Using deep learning for the automated identification of cone and rod photoreceptors from adaptive optics imaging of the human retina.

Adaptive optics imaging has enabled the enhanced in vivo retinal visualization of individual cone and rod photoreceptors. Effective analysis of such high-resolution, feature rich images requires automated, robust algorithms. This paper describes RC-UPerNet, a novel deep learning algorithm, for identifying both types of photoreceptors, and was evaluated on images from central and peripheral retina extending out to 30° from the fovea in the nasal and temporal directions. Precision, recall and Dice scores were 0.928, 0.917 and 0.922 respectively for cones, and 0.876, 0.867 and 0.870 for rods. Scores agree well with human graders and are better than previously reported AI-based approaches.

CONE PHOTORECEPTOR INTEGRITY ASSESSED WITH ADAPTIVE OPTICS IMAGING AFTER LASER POINTER-INDUCED RETINAL INJURY.

PURPOSE: To examine the three-dimensional foveal cone photoreceptor structure in a patient who had suffered laser pointer-induced retinal injury. METHODS: Patient underwent standard fundus photography and clinical spectral domain optical coherence tomography imaging. High-resolution imaging was performed using an adaptive optics-optical coherence tomography-scanning laser ophthalmoscope. RESULTS: Adaptive optics imaging revealed loss of inner and outer segments of cone photoreceptors whereas the anterior retinal layers appeared healthy. Analysis of cone topology showed an increase in Voronoi domain area and a less regular hexagonal packing structure closer to the lesion site. CONCLUSION: Exposure to laser pointer radiation, however brief, can result in damage to the retina. Here, repeated imaging nine months later showed a decrease in the size of the lesions (ranging from 3.7 to 23.9%) compared with the first time point. However, the longer-term prognosis is likely permanent scarring.

Structural Integrity of Individual Cone Photoreceptors After Short-Wavelength Subthreshold Micropulse Laser Therapy for Diabetic Macular Edema.

BACKGROUND AND OBJECTIVE: Subthreshold micropulse laser (SML) treatment at 577 nm has been proposed as a safe and efficacious therapy for diabetic macular edema (DME). The study objective was to evaluate the integrity of individual cone photoreceptors after SML treatment using high-resolution retinal imaging. PATIENTS AND METHODS: An observational cohort study of four subjects with DME treated using SML was followed over time. Cone inner and outer segment lengths and total retinal thicknesses (TRT) were measured as the edema resolved. The primary outcome was the detection of any laser-induced photoreceptor damage / change following the SML treatment using adaptive optics imaging. RESULTS: Individual cones observed pre-treatment remained visible, whereas cones that were initially obscured by the DME became more discernable after the treatment. TRT showed statistically significant thinning in half of the subjects. One subject showed no significant change, whereas another showed a statistically significant increase in TRT despite the treatment. No subject was found to have photoreceptor damage following treatment. CONCLUSIONS: SML at 577 nm did not result in measurable structural damage to the underlying photoreceptor layer, supporting previous work that SML is a safe alternative for treating DME. [Ophthalmic Surg Lasers Imaging Retina. 2018;49:946-954.].

Inner Retinal Changes in Primary Open-Angle Glaucoma Revealed Through Adaptive Optics-Optical Coherence Tomography.

PURPOSE: To examine the microstructural changes in the inner nuclear layer (INL) and ganglion cell layer (GCL) in a primary open-angle glaucoma (POAG) subject at 2 timepoints, 4 months apart. PATIENTS AND METHODS: This case-control study (1 POAG subject and 1 normal control) used the single cell, 3-dimensional volumetric imaging capability of an adaptive optics-optical coherence tomography-scanning laser ophthalmoscopy system to examine the inner retina. RESULTS: At the area of greatest glaucomatous change in the POAG subject [3-degrees temporal (T), 3-degrees inferior (I), right eye], the GCL was greatly thinned at both timepoints, yet retinal ganglion cell soma remained visible amid a meshwork of capillaries. Microcystic lesions in the INL were visible at both timepoints, ranging in diameter from 8 to 43 µm on day 1 to 11 to 64 µm at 4 months, with an average diameter increase of ∼124%. Small hyperreflective features (not seen in the contralateral eye or control subject) at a depth midway through the INL seemed correlated to the development of microcysts. CONCLUSIONS: We demonstrate the ability to image microcystic lesions early in their development and have quantified longitudinal changes. The presence of small hyperreflective structures at a layer midway through the INL seems to be a precursor to their formation and is a potential biomarker for assessing POAG severity and progression. The adaptive optics imaging system is also able to visualize retinal ganglion cells in this subject, despite severe thinning of the GCL.

Volumetric imaging of rod and cone photoreceptor structure with a combined adaptive optics-optical coherence tomography-scanning laser ophthalmoscope.

We have designed and implemented a dual-mode adaptive optics (AO) imaging system that combines spectral domain optical coherence tomography (OCT) and scanning laser ophthalmoscopy (SLO) for in vivo imaging of the human retina. The system simultaneously acquires SLO frames and OCT B-scans at 60 Hz with an OCT volume acquisition time of 4.2 s. Transverse eye motion measured from the SLO is used to register the OCT B-scans to generate three-dimensional (3-D) volumes. Key optical design considerations include: minimizing system aberrations through the use of off-axis relay telescopes, conjugate pupil plane requirements, and the use of dichroic beam splitters to separate and recombine the OCT and SLO beams around the nonshared horizontal scanning mirrors. To demonstrate system performance, AO-OCT-SLO images and measurements are taken from three normal human subjects ranging in retinal eccentricity from the fovea out to 15-deg temporal and 20-deg superior. Also presented are en face OCT projections generated from the registered 3-D volumes. The ability to acquire high-resolution 3-D images of the human retina in the midperiphery and beyond has clinical importance in diseases, such as retinitis pigmentosa and cone-rod dystrophy.

Vision science and adaptive optics, the state of the field.

Adaptive optics is a relatively new field, yet it is spreading rapidly and allows new questions to be asked about how the visual system is organized. The editors of this feature issue have posed a series of question to scientists involved in using adaptive optics in vision science. The questions are focused on three main areas. In the first we investigate the use of adaptive optics for psychophysical measurements of visual system function and for improving the optics of the eye. In the second, we look at the applications and impact of adaptive optics on retinal imaging and its promise for basic and applied research. In the third, we explore how adaptive optics is being used to improve our understanding of the neurophysiology of the visual system.

Directionality of individual cone photoreceptors in the parafoveal region.

The pointing direction of cone photoreceptors can be inferred from the Stiles-Crawford Effect of the First Kind (SCE-I) measurement. Healthy retinas have tightly packed cones with a SCE-I function peak either centered in the pupil or with a slight nasal bias. Various retinal pathologies can change the profile of the SCE-I function implying that the arrangement or the light capturing properties of the cone photoreceptors are affected. Measuring the SCE-I may reveal early signs of photoreceptor change before actual cell apoptosis occurs. In vivo retinal imaging with adaptive optics (AO) was used to measure the pointing direction of individual cones at eight retinal locations in four control human subjects. Retinal images were acquired by translating an aperture in the light delivery arm through 19 different locations across a subject's entrance pupil. Angular tuning properties of individual cones were calculated by fitting a Gaussian to the reflected intensity profile of each cone projected onto the pupil. Results were compared to those from an accepted psychophysical SCE-I measurement technique. The maximal difference in cone directionality of an ensemble of cones, ρ¯, between the major and minor axes of the Gaussian fit was 0.05 versus 0.29mm(-2) in one subject. All four subjects were found to have a mean nasal bias of 0.81mm with a standard deviation of ±0.30mm in the peak position at all retinal locations with mean ρ¯ value decreasing by 23% with increasing retinal eccentricity. Results show that cones in the parafoveal region converge towards the center of the pupillary aperture, confirming the anterior pointing alignment hypothesis.

Measurement of the photoreceptor pointing in the living chick eye.

The chick eye is used in the study of ocular growth and emmetropization; however optical aberrations in the lens and cornea limit the ability to visualize fine retinal structure in living eyes. These aberrations can be corrected using adaptive optics (AO) allowing for cellular level imaging in vivo. Here, this capability is extended to measure the angular tuning properties of individual photoreceptors. The left eyes from two White Leghorn chicks (Gallus gallus domesticus) labeled chick A and chick B, were imaged using an AO flood illuminated fundus camera. By translating the entrance pupil position, the same retinal location was illuminated with light of varying angles allowing for the measurement of individual photoreceptor pointing. At 30° nasal from the pecten tip, the pointing direction for both chicks was towards the pupil center with a narrow distribution. These particular chicks were found to have a temporal (T) and inferior (I) bias in the alignment with peak positions of (0.81 T, 0.23 I) and (0.57 T, 0.18 I) mm from the pupil center for chicks A and B respectively. The rho, ρ, values for the major, ρL, and minor, ρs, axes were 0.14 and 0.17mm(-2) for chick A and 0.09 and 0.20mm(-2) for chick B. The small disarray in the alignment of the chick photoreceptors implies that the photoreceptors are aligned to optimize the light entering the eye through the central portion of the pupil aperture. The ability to measure pointing properties of individual photoreceptors will have application in the study of eye growth and various retinal disorders.

Single cell imaging of the chick retina with adaptive optics.

PURPOSE: The chick eye is extensively used as a model in the study of myopia and its progression; however, analysis of the photoreceptor mosaic has required the use of excised retina due to the uncorrected optical aberrations in the lens and cornea. This study implemented high resolution adaptive optics (AO) retinal imaging to visualize the chick cone mosaic in vivo. METHODS: The New England College of Optometry (NECO) AO fundus camera was modified to allow high resolution in vivo imaging on two 6-week-old White Leghorn chicks (Gallus gallus domesticus)-labeled chick A and chick B. Multiple, adjacent images, each with a 2.5(o) field of view, were taken and subsequently montaged together. This process was repeated at varying retinal locations measured from the tip of the pecten. Automated software was used to determine the cone spacing and density at each location. Voronoi analysis was applied to determine the packing arrangement of the cones. RESULTS: In both chicks, cone photoreceptors were clearly visible at all retinal locations imaged. Cone densities measured at 36(o) nasal-12(o) superior retina from the pecten tip for chick A and 40(o) nasal-12(o) superior retina for chick B were 21,714 ± 543 and 26,105 ± 653 cones/mm(2) respectively. For chick B, a further 11 locations immediately surrounding the pecten were imaged, with cone densities ranging from 20,980 ± 524 to 25,148 ± 629 cones/mm(2). CONCLUSION: In vivo analysis of the cone density and its packing characteristics are now possible in the chick eye through AO imaging, which has important implications for future studies of myopia and ocular disease research.

In vivo imaging of the human rod photoreceptor mosaic.

Although single cone receptors have been imaged in vivo, to our knowledge there has been no observation of rods in the living normal eye. Using an adaptive optics ophthalmoscope and post processing, evidence of a rod mosaic was observed at 5° and 10° eccentricities in the horizontal temporal retina. For four normal human subjects, small structures were observed between the larger cones and were observed repeatedly at the same locations on different days, and with varying wavelengths. Image analysis gave spacings that agree well with rod measurements from histological data.

Evidence of outer retinal changes in glaucoma patients as revealed by ultrahigh-resolution in vivo retinal imaging.

UNLABELLED: AIMS; It is well established that glaucoma results in a thinning of the inner retina. To investigate whether the outer retina is also involved, ultrahigh-resolution retinal imaging techniques were utilised. METHODS: Eyes from 10 glaucoma patients (25-78 years old), were imaged using three research-grade instruments: (1) ultrahigh-resolution Fourier-domain optical coherence tomography (UHR-FD-OCT), (2) adaptive optics (AO) UHR-FD-OCT and (3) AO-flood illuminated fundus camera (AO-FC). UHR-FD-OCT and AO-UHR-FD-OCT B-scans were examined for any abnormalities in the retinal layers. On some patients, cone density measurements were made from the AO-FC en face images. Correlations between retinal structure and visual sensitivity were measured by Humphrey visual-field (VF) testing made at the corresponding retinal locations. RESULTS: All three in vivo imaging modalities revealed evidence of outer retinal changes along with the expected thinning of the inner retina in glaucomatous eyes with VF loss. AO-UHR-FD-OCT images identified the exact location of structural changes within the cone photoreceptor layer with the AO-FC en face images showing dark areas in the cone mosaic at the same retinal locations with reduced visual sensitivity. CONCLUSION: Losses in cone density along with expected inner retinal changes were demonstrated in well-characterised glaucoma patients with VF loss.

High-resolution Fourier-domain optical coherence tomography of choroidal neovascular membranes associated with age-related macular degeneration.

PURPOSE: To investigate the use of high-resolution Fourier-domain optical coherence tomography (Fd-OCT) to image choroidal neovascular membranes (CNVMs) associated with exudative age-related macular degeneration (eAMD). METHODS: An Fd-OCT instrument with axial resolution of 4 to 4.5 microm and transverse resolution of 10 to 15 microm was used to image 21 eyes (19 subjects) with newly diagnosed eAMD. A raster series of 100 B-scans separated by 60 microm was used to study the growth pattern of CNVM and associated morphologic changes. CNVM size was determined using 250 to 300 serial virtual C-scans of reconstructed three-dimensional macular volume. RESULTS: A highly reflective subretinal and/or subretinal pigment epithelial (RPE) lesion that co-localized with the CNVM seen on fluorescein angiography was detected in all eyes by Fd-OCT. Although a combined subretinal and sub-RPE growth pattern of various degrees was noted in 15 (71%) eyes, a statistically significant difference in the distribution of growth pattern was noted when classic CNVM was compared with occult CNVM (chi(2) = 10.4, df = 2, P < 0.005). Classic lesions had >90% subretinal growth pattern, whereas occult lesions had a more variable growth pattern. Angiographic CNVM size correlated with size on Fd-OCT but correlation was better for classic CNVM (classic, r = 0.99, P < 0.0001; nonclassic, r = 0.78, P < 0.001). CONCLUSIONS: Fd-OCT is a promising potential alternative modality to visualize CNVM with AMD. Angiographic lesion size and type correlated with growth pattern and size of CNVM on Fd-OCT, with correlation being stronger for classic lesions.