This PDF file contains the front matter associated with SPIE Proceedings Volume 13831, including the Title Page, Copyright information, Table of Contents, and Conference Committee information.

Widefield handheld fundus photography systems are utilized in applications such as the screening of diabetic retinopathy and retinopathy of prematurity. However, these systems may cause patient stress and are prone to significant imaging artifacts. To address these issues, we present a widefield, handheld confocal scanning laser ophthalmoscope (WiSLO). Precision-machined opto-mechanics and a hybrid spiral scan are used to transform the WiSLO into a handheld form factor. The WiSLO can be used in a contact imaging mode, with a 133-degree field-of-view, and a noncontact imaging mode, with a 53-degree field-of-view. Representative images from adult imaging in both modes are shown.

We developed a handheld, non-mydriatic swept-source OCT system capable of ultra-widefield and high-resolution retinal imaging across both pediatric and adult patients without pharmacologic dilation. The proposed system employed a 400kHz VCSEL light source with an extended axial imaging range of 12mm in air and achieved a 140° visual angle with custom optics, enabling rapid, panretinal visualization with minimal beam wandering while maintaining diffraction-limited resolution across the field of view.

Retinal degenerations, such as retinitis pigmentosa (RP), are among the most challenging conditions in ophthalmology due to their rarity, genetic heterogeneity, and progressive vision loss caused by photoreceptor dysfunction and death. Despite advances in imaging technologies and clinical trial design, treatment options remain limited for most inherited retinal diseases (IRDs). Adaptive optics (AO) imaging offers a promising avenue for improving clinical trial outcomes by providing high-resolution, non-invasive visualization of photoreceptor structure and function. Techniques like adaptive optics scanning laser ophthalmoscopy (AOSLO) and optoretinography (ORG) allow researchers to study cone structure and function with cellular resolution, offering objective biomarkers for disease progression and treatment response. However, the widespread adoption of AO imaging in multicenter trials has been hindered by the lack of standardized protocols and systems. Natural history studies, such as the RUSH2A study of USH2A-related retinal degeneration, highlight the importance of identifying sensitive outcome measures for clinical trials. Conventional metrics like visual acuity and ellipsoid zone (EZ) area often fail to capture meaningful changes in disease progression, while visual function measures and AOSLO imaging show greater sensitivity. Innovative applications of AO imaging, such as AOMP and studies of enhanced S-cone syndrome (ESCS), demonstrate its ability to bridge structural and functional analysis. With ongoing efforts to standardize AO protocols and integrate them into trials like NAC Attack, AO imaging has the potential to revolutionize IRD research. By combining structural imaging, functional measures, and natural history data, AO imaging can unlock new opportunities for developing treatments for retinal degenerations.

Conventional fundus photography represents retinal vessels in two dimensions, precluding visualization of three-dimensional velocity distributions and quantification of local blood-flow mechanics at arteriovenous (AV) crossings. Using a Doppler OCT blood flowmeter, venous intravascular velocity profiles were extracted at AV crossings, and wall shear stress (WSS) was quantified. In this study, we report the physiological range of WSS at AV crossings in healthy eyes and systematically evaluate within-day repeatability and day-to-day reproducibility of these OCT-derived hemodynamic measurements.

Imaging the human retinal blood flow at the level of an individual capillary requires the use of adaptive optics (AO) to overcome ocular aberrations. Furthermore, to accurately capture the temporal dynamics, high speed imaging is also required. Here, a high-speed multi-line partially confocal scanning laser ophthalmoscope employing a digital micromirror device (DMD) is described. The DMD allows projection of multiple confocal lines at any direction and spacing and is coupled with the use of a high-speed two-dimensional complementary metal-oxide-semiconductor camera. The arrangement allows for the simultaneous acquisition of both the confocal and multiply scattered images. Blood vessel imaging results on human controls and those with primary open angle glaucoma are presented, including some initial results on vasodilation in response to flicker stimulation.

Visual simulators play a key role in the development of new visual corrections and in the prediction of the most suitable prescription for patients in the clinic. We present a novel binocular, open-view simulator incorporating automatic convergence control and simultaneous dynamic binocular aberrometry. In this configuration, all three accommodation cues (blur, proximity and convergence) are preserved, enabling realistic evaluation of different corrections. The system consists of a set of independent relay optics for each eye, allowing compensation of the patient’s refractive error and simulation of contact or intraocular lenses according to its spatial phase distribution. Using a single wavefront sensor with two regions of interest, ocular aberrations and accommodation are measured simultaneously in both eyes. Automatic convergence is achieved using one translational stage and two mirrors mounted on rotational motors for each eye, which maintain gaze alignment with the optical axis and the visual target. An additional translational stage positions the wavefront sensor in a conjugate pupil plane, ensuring accurate and stable measurements. This design allows patients to view stimuli at arbitrary distances while maintaining alignment with the system’s optical axis, with the simulated corrections (implemented via a spatial light modulator or a phase plate), centered or intentionally decentered relative to each pupil. Intended applications of the system include the evaluation of myopia control and presbyopia corrections, including multifocal lenses and monovision.

Retinal degenerative diseases of photoreceptors are a leading cause of blindness with no effective treatment. Retinal prostheses aim to restore sight by stimulating residual retinal cells. Typically, retinal prostheses are implantable devices requiring surgical insertion in the eye, and their commercialization is a lengthy, complex and costly process due to the need for long-term safety data. Prior work (Leong et al. 2025) demonstrated the potential of photoacoustic stimulation for visual restoration in blind patients using a subretinally implanted photoacoustic film. Here, we present a non-invasive photoacoustic contact lens for vision restoration. We characterized the photoacoustic properties of a commercial fully opaque “blindfold” contact lens with a 75μm needle hydrophone. The 5mm central part of the lens was then cut out and used as a modified contact lens to focus ultrasounds in vivo on the retinas of rats through the rat eye. PA stimulation via contact lens elicited retinal activity that propagated along the visual pathway to the brain, as measured by functional ultrasound imaging and electrocorticography. This confirmed that the lens’ focal gain adequately compensated for ocular ultrasound attenuation, even at high frequencies (> 20MHz). Control experiments confirmed that activation was not due to the residual 1064nm laser light transmitted through the contact lens. Mechanical Index, ultrasound intensity and Thermal Index were within the FDA’s safety thresholds for ophthalmic stimulation. These findings demonstrate the potential of photoacoustic stimulation as a non-invasive approach for visual restoration in blind patients.

We measured neurovascular coupling (NVC) in the living human retina using an Adaptive Optics Rolling Slit Ophthalmoscope, achieving sub-micrometer spatial precision and 10 ms temporal resolution. Flicker stimulation induced a significant arterial dilation (~5%) with a rapid, triphasic response, distinct from baseline vascular fluctuations. This approach enables high-resolution, stimulus-specific assessment of retinal NVC and may support early detection of neurovascular dysfunction in retinal diseases

The corneal-scleral limbus contains crucial aqueous channels and veins that are critical for aqueous outflow and glaucoma research. Conventional imaging of the limbal vasculature often requires multi-quadrant scans with complex projection-angle correction or suffers from defocus artifacts due to the pronounced curvature of the sclera. In this study, we present a circular-scan anterior segment optical coherence tomography (CircAS-OCT) system for rapid, depth-resolved imaging of the full limbal vasculature. The system employs a pericentric optical design to minimize the beam incidence angle and reduce defocus across the curved ocular surface. It enables single-shot acquisition of circumferential OCT angiography (OCTA) in just 2.67 seconds with an annular pattern. To integrate other imaging contrasts of the aqueous outflow pathway, we introduce a registration framework that aligns and stitches OCTA images from other systems or scanning protocols onto the CircAS-OCT angiogram. Experimental results demonstrate precise vessel alignment across stitched images, along with multi-contrast mapping of the limbal vascular network, collector channel locations, and the area and biomechanics of trabecular meshwork. This platform provides a fast, robust, and extensible solution for anterior segment examination, with promising applications in glaucoma diagnostics and multimodal data integration.

Recently, we introduced a novel method—Transmission Interferometric Microscopy (TIM)—for in vivo imaging of the anterior eye. The technique is inspired by retroillumination approaches that exploit the strong back-scattering of the sclera to create a transmission-illumination geometry for the anterior segment. In our implementation, we showed that the spatial coherence of light back-scattered from the sclera can be tuned to boost the interferometric phase contrast of corneal and crystalline-lens cells by several times. To date, TIM has been demonstrated in healthy subjects and in a small number of patients. Here, we report results from the first larger-scale clinical study of this technology, involving over 100 patients with a wide range of corneal pathologies. We demonstrate a direct field-to-field comparison with clinical specular microscopy, showing that this method provides new insights into cellular structure and the inflammatory burden of ocular diseases. Furthermore, we establish the first strategies for obtaining quantitative biomarkers in transmission. Finally, we present the optimized configuration of the transmission device that enabled successful imaging in a clinical research setting.

We present a GPU-accelerated sub-B-scan registration pipeline for adaptive-optics optical coherence tomography (AO-OCT) that achieves sub-pixel correction of non-rigid motion and significant speckle reduction. Based on the pioneering work by Kurokawa et al. (JBO 2021), we further subdivide B-scans and apply sub-pixel reference alignment via normalized cross-correlation, significantly improving registration precision and partially correcting shearing artifacts unresolved by standard coarse-to-fine alignment approaches. We demonstrate volume registration in a wavefront sensorless AO-OCT system, enhancing contrast-to-noise ratio enough to visualize displaced ganglion cells. Combining sensorless AO-OCT with less complex volume registration could unlock practical high-resolution retinal imaging in clinical studies.

In ophthalmology, subretinal injection is essential for treating retinal diseases like AMD and inherited disorders, but current pressure modulation relies on surgeons' manual control and visual interpretation, which is highly variable. To address this, we developed a real-time closed-loop control system using OCT speckle dynamics to dynamically adjust injection pressure. The system detects retinal-RPE separation through speckle decorrelation and displacement, switching from high-pressure injection to low-pressure expansion based on tissue-specific feedback. Experiments on ex vivo porcine eyes revealed a three-phase speckle pattern, confirming OCT speckle dynamics as a reliable real-time indicator of tissue separation. This closed-loop system enhances subretinal therapy precision and safety by preventing overexpansion and integrating seamlessly with surgical workflows.

Fluorescent ophthalmoscopy provides noninvasive imaging of living retina, offering access to valuable structural and functional information for studying the visual and central nervous systems at a single cell level. However, volumetric fluorescent imaging of retinal cells remains challenging due to ocular aberrations, slow z-stacking, and phototoxicity. Here, we present a plenoptic illuminated scanning laser ophthalmoscopy (PI-SLO), a novel 3D fluorescent retinal imaging strategy that achieves rapid single-cell resolution volumetric imaging with low phototoxicity. We demonstrate structural and functional imaging in the living mouse eye by 1) imaging the dynamics of single microglia, 2) 3D fluorescein angiography, and 3) capturing calcium signaling of inner retinal neurons.

The precise spatial control of photoreceptor stimulation is essential for understanding changes in retinal ganglion cell (RGC) function in neurodegenerative diseases, such as glaucoma. We present a fundus camera-guided system for spatially targeted retinal stimulation featuring closed-loop focus adjustment with real-time imaging feedback. This system uses a 415 nm stimulation pathway that effectively drives both S- and M-cones in the mouse retina, given their overlapping short-wavelength sensitivity. To project a spatially-modulated pattern, a digital micromirror device (DMD) is integrated into a custom fundus imaging pathway, enabling micrometer-scale alignment between the stimulation pattern and retinal features. Focus correction is achieved by measuring a segment perpendicular to a retinal blood vessel and iteratively adjusting an electronically tunable lens (ETL) to minimize its full width at half maximum (FWHM). System performance was validated by positioning a CMOS sensor at the eye plane, confirming a projected pattern resolution of 1.6μm. Moreover, the modulation transfer function (MTF) was derived from a slanted-edge projection, achieving a 50% MTF of 35-40 cycles/mm. These results demonstrate the feasibility of cellular-resolution stimulation in vivo and establish a foundation for future investigations of RGC activity and functional alterations in retinal disease models.

Vision begins with photoisomerization, triggering conformational changes in opsins. Microelectrode recordings are invasive and insensitive to early receptor potential in rods. Phase-sensitive OCT (optoretinography, ORG) enables noninvasive label-free detection of nanoscale cellular deformations caused by physiological processes. Green flashes induce rapid contraction of the rod outer segments, while UV flashes elicit an opposite response. These deformations are explained as electromechanical manifestations of charge transfer across rod disc membranes during isomerization into Meta II (green) and Meta III (UV). ORG reveals rhodopsin dynamics with greater sensitivity than electrical methods, offering insights into visual transduction and potential for early diagnosis of photoreceptor dysfunction.

We present a wide-field, cellular-resolution optoretinography (ORG) method based on time-domain full-field optical coherence tomography (TDFFOCT). The approach combines spatially incoherent with spectrally shaped illumination and temporal gating to achieve depth-resolved amplitude-based ORG of individual cone photoreceptors across a 5° × 5° field of view, without requiring the phase stability or classical adaptive optics correction typical of current cellular-resolution ORG approaches. In healthy subjects, the system reveals layer-dependent ORG responses to visible-light stimulation and captures eccentricity-dependent variations in cone ORG amplitudes. The reduced data burden of en face acquisition, the wide field-of-view and compatibility with low-order sensorless adaptive optics suggest that time-domain FFOCT-based ORG can provide a practical foundation for functional retinal imaging in research and future clinical applications.

Accurate measurement of choroidal tumor thickness is essential for diagnosis, treatment planning, and prognostication. Ultrasonography remains the gold standard, but it requires ocular contact, specialized training, and patient cooperation, which limits accessibility in many settings. This study introduces a stereo-vision-inspired deep learning framework capable of estimating choroidal tumor thickness directly from two-dimensional ultra-widefield fundus photographs. The model leverages spectral disparity between the color channels because red and green wavelengths penetrate to different retinal depths. It therefore infers relative elevation by learning spatial relationships encoded across these spectral cues. A dataset of 265 fundus images with corresponding ultrasound measurements was used to train and evaluate models based on individual color channels, full-color images, and the fusion approach. The stereo-fusion approach achieved the best overall agreement with ultrasound, with a root mean squared error of 1.35 ± 0.26 mm, a mean absolute error of 0.95 ± 0.16 mm, and an R² of 0.79 ± 0.08. Gradient-weighted class activation mapping revealed that the model focused on both the tumor and the surrounding flat retina, indicating that depth inference relied on contextual spatial relationships rather than isolated intensity features. These findings demonstrate that deep learning can extract latent depth information from fundus images, providing a practical and non-contact alternative for estimating tumor thickness. The approach has the potential to enhance screening, teleophthalmology, and longitudinal monitoring, particularly in regions where access to ocular ultrasound is limited.

Gapped spectra pose challenges for OCT reconstruction. In this work, a neural network is presented that takes individually reconstructed B-scans as inputs and outputs a single, high-resolution scan, successfully recovering anatomical details. As a side-effect, the resulting network output exhibits strong noise-reduction. By artificially introducing different gaps in high-bandwidth scans, it is demonstrated that this denoising can be controlled by varying the spectral gap.

Accurate diagnosis of major eye related diseases such as glaucoma, diabetic retinopathy, diabetic macular edema, pathological myopia, and age related macular degeneration is essential for timely intervention, yet remains challenging in diverse clinical and resource settings. We develop a highly accurate multi-task disease diagnosis model that combines patient metadata with fundus images to diagnose five eye diseases simultaneously. We applied Metafusion, a fusion technique to effectively combine patient metadata (age, gender, history of diabetes and hypertension) with fundus images for diagnosis. In addition, through the development of a novel pre-training method combining supervised and self-supervised losses, we leverage multiple fundus image datasets for diabetic retinopathy and glaucoma to train a multi-task image encoder. The disease diagnosis model demonstrates strong generalization across both lab-captured and smartphone-captured images (Brazilian Multilabel Ophthalmological Dataset, or BRSET, and Mobile BRSET, or mBRSET, datasets respectively). On the BRSET dataset, the disease diagnosis model shows an improvement of 6% in balanced accuracy across all five diseases compared to a model which relies only on images without pre-training enhancements. On the mBRSET dataset there is a 4% improvement in balanced accuracy for diabetic retinopathy and diabetic macular edema. Furthermore, training on a joint (mBRSET+BRSET) dataset preserves performance across both domains, demonstrating model robustness across a range of imaging conditions. This work demonstrates how multimodal data fusion and self-supervised pre-training can improve disease detection accuracy while maintaining high performance across different imaging conditions, which is an important requirement for future ophthalmic diagnostic systems.

Our study investigates diffusion-model based translation of optical coherence tomography (OCT) into OCT Angiography (OCTA), with a focused evaluation on the publicly available OCT500 dataset. A Brownian Bridge Diffusion Model (BBDM) was trained and tested exclusively on paired OCT–OCTA volumes from OCT500, encompassing 3mm and 6mm fields of view (FoV). Performance was assessed using standard quality metrics: structural similarity index (SSIM), Fréchet inception distance (FID), and perceptual contrast quality index (PCQI) alongside clinically relevant vascular biomarkers including blood vessel density (BVD), blood vessel caliber (BVC), blood vessel tortuosity (BVT) and vessel perimeter index (VPI). The diffusion model demonstrated strong structural fidelity, particularly for larger FoV scans. While SSIM for 3mm scans was comparable to GAN-based baselines, BBDM achieved higher SSIM and PCQI scores in the 6 mm subset, indicating improved preservation of global vascular structure and perceptual contrast. Quantitative feature analysis revealed close agreement between translated OCTA (TR-OCTA) and ground-truth OCTA for BVD in 3mm scans, with moderate deviations observed in BVC, BVT and VPI across both FoVs. Importantly, the diffusion model preserved anatomically meaningful trends in vascular features despite absolute differences, supporting its clinical interpretability. Although FID values remained higher than GAN counterparts, qualitative assessment showed reduced hallucinated vascular artifacts, highlighting improved reliability of generated vasculature. Overall, this study demonstrated that diffusion-based OCT-to-OCTA translation can generate structurally consistent and clinically meaningful vascular representations, particularly in wider FoV, setting BBDM as a promising alternative to adversarial methods for OCTA synthesis from standard OCT imaging.

Optical coherence tomography (OCT) imaging is fundamental for diagnosing retinal diseases. However, existing AI models for OCT lack interpretability, operating as “black boxes” that limit clinical adoption. To address this, we introduce OCT-BLIP, a compact vision–language model (VLM) that enhances diagnostic performance and provides anatomically grounded, layer-specific explanations. We assembled a multimodal dataset of 40,000 OCT image–text pairs. Images were sourced from both private institutional collections and publicly available repositories. For each image, preliminary layer-specific pathological descriptions spanning five retinal layers were generated using GPT-4o. All generated descriptions underwent subsequent expert review and manual refinement to ensure clinical accuracy and anatomical precision, and consistency. The dataset covers six diagnostic categories: CNV, drusen, DR, GA, DME, and healthy cases. OCT-BLIP, with roughly 247 million parameters based on the BLIP architecture, combines a Vision Transformer encoder and a BERT-style text decoder fine-tuned for classification and captioning. Trained for 50 epochs on 39,000 paired samples using AdamW with a cosine learning-rate schedule, OCT-BLIP achieves 96% classification accuracy, outperforming an 83% standalone ViT baseline and RetinaVLM (< 15%). The model produces precise captions, attaining a mean SBERT similarity of 80.3% and a BERTScore-F1 of 71.5%, substantially surpassing RetinaVLM (71.4% and 42.7%, respectively). The interpretability study assessed structured output clarity, diagnostic usefulness, and anatomical accuracy.

Geographic atrophy, an advanced form of age-related macular degeneration causing irreversible vision loss, requires accurate progression prediction to optimize emerging therapies. This study investigated whether temporal deep learning models can predict geographic atrophy progression directly from longitudinal optical coherence tomography imaging without manual lesion segmentation. Using retrospective imaging data from 2013-2023, we encoded individual volumes into visit-level features and modeled them as temporal sequences to predict two clinically critical transitions within a two-year window: progression from no atrophy to disease onset, and progression from non-central to central foveal involvement. We systematically compared recurrent neural networks, long short-term memory networks, and transformer architectures using patient-level data splits. Recurrent models achieved robust performance for predicting disease onset, while transformer models performed best for the more challenging central progression task. This annotation-minimal framework demonstrates the clinical feasibility of predicting geographic atrophy progression from routinely collected imaging data, enabling earlier risk stratification and enhanced therapeutic decision-making.

Glaucoma is a progressive optic neuropathy that accompanied by the development of visual functional defects induced by elevated intraocular pressure (IOP). In clinic, glaucoma is generally treated by pharmacological therapy and surgical intervention. Among the therapeutic techniques, cyclocryotherapy and cyclophotocoagulation can suppress aqueous humor secretion by destroying the ciliary body to lower IOP, while lack of precision in clinical application. In this study, a precise therapeutic ciliary body destruction technique, ciliary body radiofrequency ablation (RFA), was used utilizing high-frequency alternating current to generate localized thermal energy to destroy ciliary body. The effectiveness and feasibility of glaucoma treatment were evaluated and verified using multimodal methods including optical coherence tomography (OCT), OCT angiography (OCTA), and electroretinogram (ERG). Using an ocular hypertension (OHT) model in C57BL/6 mice via anterior chamber microbead injection, RFA of the ciliary body was performed and evaluated during 4 weeks. IOP shows a steady and continuous decrease after treatment. OCT images reveals that retinal thickness exhibits a markedly slower thinning trend and is markedly thicker than those in the glaucoma model group. OCTA analysis demonstrates significantly higher vascular densities in superficial and deep retinal layers in the treatment group. PhNR assessments indicates that RFA therapy preserved RGC function and reduced RGC damage. Experimental results prove that RFA of the ciliary body could be used as a strategy to control IOP and as a treatment for glaucoma.

Traditional eye refraction methods rely on patient feedback and are inherently subjective, typically correcting only defocus and astigmatism. While modern lens fabrication can correct higher-order aberrations, current wavefront sensors like Shack-Hartmann systems are expensive and mostly confined to clinical or hospital settings. This project leverages AI4Wave, an AI-based image-space wavefront sensing technology originally developed for astronomy, metrology, and adaptive optics [1]. AI4Wave computes the eye’s wavefront directly from a n image projected on the retina using a compact, low-cost system. It requires minimal hardware and can be easily integrated with standard optometric tools. This enables objective, full-aberration refraction at a fraction of the cost, paving the way for broader access to personalized, high-precision vision correction. The proposed approach is entirely deterministic. AI4Wave phase retrieval leverages feedforward neural network computation, eliminating the need for iterations. This approach ensures consistent results and computation times, free from real-time optimization challenges or issues related to local minima.

Building upon the positive defocus strategy for myopia control, this study proposes an innovative optical design that integrates a microlens array with a metasurface to effectively regulate focal positioning and adapt to multi-wavelength environments within a compact form factor. By introducing +3.5D positive defocus and +2.5D longitudinal chromatic aberration, the design aims to modulate axial eye growth through combined multi-point defocus and chromatic cues. This system is expected to provide both retinal feedback across different wavelengths and therapeutic effects through spatially distributed defocus to suppress axial elongation. By combining dispersion modulation with multifocal functionality, the design simulates the complex lighting conditions of outdoor environments, addressing the limitations of current single-condition myopia control lenses. It offers an effective approach to suppress early myopia progression and is expected to further reduce axial elongation by approximately 20%.

Mutations in the USH2A gene are the leading cause of Usher syndrome type 2A, accounting for over half of all Usher syndrome patients, and closely linked to non-syndromic autosomal recessive retinitis pigmentosa (RP). Despite the clinical significance of RP in USH2A, the characterization of retinal degeneration in these models poses significant challenges due to the late onset and gradual progression of RP. This study leverages the non-invasive imaging capabilities of optical coherence tomography (OCT) to characterize retinal degeneration in a USH2A KO rabbit model, aiming to establish OCT-derived biomarkers for both model characterization and therapeutic evaluation.

Early and accurate identification of ophthalmic diseases is critical for timely intervention and prevention of irreversible vision loss. Optical coherence tomography (OCT) has become a central imaging modality for retinal disease diagnosis; however, the development of robust deep learning models is often hindered by data silos, privacy constraints, and substantial heterogeneity across institutions. Federated learning (FL) offers a promising solution by enabling collaborative model training without centralized data sharing, yet existing federated optimization methods struggle under non-IID data distributions, device variability, and inconsistent disease taxonomies commonly encountered in ophthalmology. In this work, we propose FedSim, a federated multi-task learning framework that addresses these challenges through geometry-aware relevance aggregation. FedSim decomposes the model into a shared feature encoder and site-specific classification heads, allowing each institution to optimize its local diagnostic task while contributing to a common representation space. Rather than indiscriminately averaging model parameters, FedSim estimates cross-site relevance using eigen-decomposition of local feature covariance structures and selectively aggregates updates from peers with compatible feature geometry, thereby mitigating negative transfer. We evaluate FedSim across eight OCT datasets, including six public benchmarks and two private clinical cohorts, encompassing multiple imaging devices, acquisition protocols, and retinal disease categories. Experimental results demonstrate that FedSim consistently outperforms strong personalized federated baselines, including FedAvg and FedProx, in terms of AUC-ROC and macro F1 score. Notably, FedSim maintains stable performance on external and private datasets collected under real-world clinical conditions, highlighting its robustness to domain shift. These findings suggest that relevance-aware aggregation provides an effective and privacy-preserving mechanism for federated ophthalmic learning, enabling scalable collaboration across heterogeneous institutions while maintaining strong diagnostic performance.

Glaucoma is a leading cause of irreversible blindness worldwide, where early detection and longitudinal monitoring rely on subtle structural cues in retinal fundus images and expert-level clinical interpretation. While recent vision–language models (VLMs) have demonstrated strong general-purpose multimodal reasoning, their performance remains limited in ophthalmology due to the absence of glaucoma-specific visual priors, specialized medical vocabulary, and complexity-aware clinical language generation. In this work, we introduce GlaucomaVLM, a domain-specific vision–language model designed to jointly perform glaucoma classification, medical image–text retrieval, and clinical caption generation. Our approach employs a progressive multi-task learning framework consisting of three stages: (i) a discriminative foundation phase that establishes robust visual representations for glaucoma classification and retrieval, (ii) a medical domain adaptation phase that incorporates 241 glaucoma-specific terms stratified across four levels of clinical complexity, and (iii) a progressive vision–language integration phase using curriculum learning to align visual features with expert-level clinical language. To support this training paradigm, we construct a curated dataset of 6,683 high-quality glaucoma image–caption pairs mined from PubMed Central literature, with systematic filtering and complexity stratification. Extensive experiments demonstrate that GlaucomaVLM substantially outperforms general and biomedical VLM baselines, achieving 81.16% classification accuracy (vs. 59.93% for BiomedCLIP), 42.84% Recall@1 in image–text retrieval (vs. 7.58% for general CLIP), and improved clinical caption quality as measured by BLEU scores. These results highlight the importance of domain-specific multimodal adaptation for ophthalmic AI and position GlaucomaVLM as a scalable framework for automated glaucoma screening, clinical decision support, and medical education.

Divergence between the Nd:YAG treatment beam focal-point and the aiming beams’ intersection, is modeled combining analytical chromatic-aberration theory and ray-tracing. Divergence reverses sign at 8mm depth, reaching 305μm posteriorly. The resulting calibration curve enables safe posterior-segment laser use.

Maintaining sufficient surgical visibility under low dye staining remains a critical challenge in ophthalmic surgery due to the trade-off between contrast enhancement and dye-related tissue toxicity. This study investigates surgical visibility under extremely low dye concentration conditions through a quantitative evaluation framework, with spectrally optimized illumination employed as a representative application. Illumination spectra, referred to as Spectral Characteristics Optimized Imaging (SCOI), are designed based on the absorption characteristics of clinically used ophthalmic surgical dyes and evaluated using gelatin-based thin membrane models representing transparent surgical membranes encountered in ophthalmic surgery. Images are acquired under conventional xenon illumination and SCOI illumination using a controlled camera-based imaging setup. Surgical visibility is quantitatively assessed using complementary objective metrics, including the CIEDE2000 color difference (ΔE₀₀) and the contrast-to-noise ratio (CNR), reflecting perceptual separation and signal robustness, respectively. Under an extremely low dye concentration conditions (0.005% trypan blue), SCOI illumination demonstrates enhanced visibility differentiation relative to conventional xenon lighting, while enabling systematic quantitative interpretation under low-contrast conditions. Taken together, these results highlight surgical visibility as a multidimensional observable outcome. Furthermore, they demonstrate that a clinician-informed, multi-metric evaluation framework enables systematic and fair comparison of illumination and imaging conditions under low dye staining, independent of specific optical implementations.

Optical transparency is a critical quality attribute of corneal lenticules intended for refractive correction, yet current evaluation methods rely primarily on qualitative visual inspection. In this study, we present a benchtop transparency measurement device that provides quantitative, spatially resolved transparency maps of corneal tissue samples, including lenticules measured directly within their clinical packaging. The system operates in transmission mode and quantifies transparency by analyzing contrast degradation of a high-precision grid pattern imaged through the sample. Device performance was validated using water–milk liquid phantoms with graded turbidity, enabling definition of quantitative thresholds corresponding to low, medium, and high transparency levels. These thresholds showed strong agreement with independent visual assessments and prior literature. The device demonstrated high repeatability, with a coefficient of variation below 1.1%. The system was applied to evaluate human corneal lenticules generated for presbyopia correction. Transparency measurements showed no dependence on stromal depth of origin. Additionally, paired measurements acquired before and after secondary electron-beam (e-beam) sterilization demonstrated that lenticules maintained high optical transparency following sterilization. Control measurements confirmed that minor transparency changes observed post-sterilization were attributable primarily to the storage medium rather than the tissue itself. This transparency-mapping approach enables objective, quantitative quality control of corneal lenticules and represents a significant improvement over traditional visual inspection methods, supporting reliable assessment of implant optical quality prior to clinical use.

We evaluated the long-term stability of an OCT angiography (OCT A) model eye that employs Brownian-motion induced particle displacement instead of liquid flow. The model eye contains a retinal phantom, which uses polystyrene nanoparticles (200 ± 20nm) distributed in liquids of water and glycerol with different mixture ratios to generate desired OCT-A signal intensity. We introduced a sealing structure with glass plates and a low permeability epoxy. We assessed three-month stability in terms of OCT-A signal intensity. The particle-distributed liquids of 10 and 20% water mixture successfully generated almost constantly relatively weak OCT-A signals over 91 days.

This presentation, originally published on 5 March 2026, was replaced with a corrected version on 23 March 2026. Per author request, the title was updated with a spelling correction, and an author was added. If you downloaded the original PDF but are unable to access the revision, please contact SPIE Digital Library Customer Service for assistance. Download the erratum for additional details.