Search Results (766)

Retinal vessel segmentation remains challenging for thin vessels and low-contrast bifurcations. We evaluate a Swin-UNet-style model family that conditions decoder features with a single-channel local fractal dimension prior refined by a short learnable anisotropic diffusion model and injected through Spatially-Adaptive Normalization (SPADE). On Fundus Image Vessel Segmentations (FIVES), the strongest no-test-time-augmentation result was obtained by OPT-I v2 at 200 epochs, reaching Dice 0.8899, clDice 0.8517, and Area Under the ROC Curve (AUC) 0.9904, compared with 0.8643, 0.8125, and 0.9856 for the matched 200-epoch baseline. In a matched Neural Cellular Automata (NCA)/no-NCA ablation using the same seed, data, 200-epoch budget, and evaluation pipeline, enabling NCA improved the test Dice from 0.8813 to 0.8907 and the test clDice from 0.8325 to 0.8518, with NCA winning on all 80 paired test images for both metrics. The results support PDE (partial differential equation)-SPADE fractal prior conditioning and NCA topology refinement as ablation-grounded improvements over the tested baseline family, while broader matched external validation requires future work. Full article

The correct classification of retinal lesions in retinal fundus images is important for supporting the analysis of diabetic retinopathy and age-related macular degeneration. State-of-the-art methods for this task are often based on black-box deep learning architectures that, despite their high performance, pose significant interpretability challenges, incur high computational costs, and lack computational interpretability at the feature-decision level. In this paper, a method based on surrogate-optimized features extracted by regular expressions is proposed for the classification of two retinal lesion categories (Drusen and Cotton Wool Spots). The method uses a compact and computationally interpretable row-by-row and column-by-column regular expression feature extractor together with a two-phase surrogate search over its discrete hyperparameters. Across 100 independent stratified executions under the repeated patch-level benchmark, the proposed method achieved a mean MCC of $0.7829\pm 0.0448$, a mean accuracy of $0.9008\pm 0.0217$, and a mean F1 score of $0.8529\pm 0.0294$. The best execution reached an MCC of $0.8433$, an accuracy of $0.9286$, and a macro F1 score of $0.8966$, which was the highest result among the evaluated baselines within that same benchmark. Additional source–image disjoint grouped analyses were carried out as leakage-aware robustness checks under stricter source–image separation and to examine validation overfitting. Together, these analyses support the usefulness of the compact run-based descriptor under the present experimental conditions, while indicating that the two-phase search should be interpreted as a practical hyperparameter selection heuristic rather than as a statistically superior search strategy. Full article

Background/Objectives: Granulomatous uveitis comprises a clinically heterogeneous group of inflammatory disorders, including ocular sarcoidosis, Vogt–Koyanagi–Harada disease, sympathetic ophthalmia, tuberculosis-associated uveitis, and syphilitic uveitis. Because these entities may share overlapping posterior segment findings, clinical examination alone is often insufficient for differential diagnosis, particularly when choroidal, retinal, or retinal vascular involvement predominates. Methods: This review provides a clinically oriented overview of multimodal imaging in granulomatous uveitis, including optical coherence tomography (OCT), enhanced-depth imaging OCT, swept-source OCT, OCT angiography, fundus autofluorescence, fluorescein angiography, indocyanine green angiography, and ultrawidefield imaging. Results: Emphasis is placed on imaging patterns that help localize the predominant anatomic compartment of inflammation, distinguish major etiologies, identify diagnostic pitfalls, and assess disease activity over time. By integrating current evidence with representative multimodal imaging findings, we propose an anatomic and decision-oriented framework for interpreting granulomatous posterior segment inflammation. Conclusions: Particular attention is given to the distinction between active inflammation and irreversible structural damage, as this distinction may influence treatment escalation or tapering, timing of elective surgery, local corticosteroid therapy, and the need for diagnostic sampling in infectious or masquerade-like presentations. Full article

Diabetic retinopathy (DR) and diabetic macular oedema (DME) are two of the most significant preventable contributors to blindness in the adult population worldwide, yet current automated screening systems typically address each condition in isolation and rely on a single imaging modality. In this study, we propose a deep learning model that simultaneously grades DR severity and detects DME by fusing paired colour fundus and optical coherence tomography (OCT) images acquired from the same eye during the same clinical visit. Our architecture employs two parallel EfficientNet-B0 backbones pre-trained on ImageNet, one for each modality, whose 1280-dimensional feature vectors are concatenated into a 2560-dimensional joint representation. This fused representation passes through a shared fully connected block before branching into a three-class DR classification head and a binary DME detection head. We train and evaluate the model on a private dataset of 425 paired fundus and OCT eye images (850 images). The proposed architecture adopts feature-level fusion, in which modality-specific deep features are independently extracted from fundus and OCT images using separate convolutional backbones and subsequently concatenated to form a joint representation for multi-task learning. On the held-out test set (n= 85), the fusion model achieves 82.4% DR accuracy (area under the receiver operating characteristic curve [AUC] = 0.929, macro sensitivity = 0.81, macro specificity = 0.905) and 97.6% DME accuracy (AUC = 0.999, sensitivity = 0.833, specificity = 1.000). The fusion model detects 10 of 12 DME-positive eyes compared with only 7 of 12 for either the fundus-only or OCT-only baselines, representing a 43% relative improvement in DME sensitivity. Stratified five-fold cross-validation (n = 425 aggregated predictions) corroborates these findings, with the fusion model reaching 87.1% DR accuracy (AUC = 0.978) and 99.1% DME accuracy (AUC = 1.000). Gradient-weighted class activation mapping visualisations confirm that the fundus branch attends to clinically relevant macular lesions, whereas the OCT branch highlights retinal layer disruptions and subretinal fluid, providing interpretability. To the best of our knowledge, the proposed MultiRetNet is the first lightweight, task-specific multimodal architecture to jointly grade DR severity and detect DME from paired same-eye, same-visit fundus and OCT images through explicit feature-level fusion within a single end-to-end multi-task framework, distinct from recent generalist ophthalmic foundation models, supporting the value of multimodal fusion for comprehensive diabetic eye screening pending external validation. Full article

Primary open-angle glaucoma (POAG) is rarely recognized in large felids, yet it may cause severe vision loss and chronic pain. This case report presents the ophthalmological evaluation of a four-year-old Siberian tiger (Panthera tigris altaica) maintained under human care that exhibited signs of visual impairment and altered sensorium. Under neuroleptic analgesia, clinical findings included bilateral convergent strabismus, persistent mydriasis, pathologically elevated intraocular pressure (48–52 mmHg), extensive map-like retinopathy, and focal retinal degeneration, supporting a diagnosis of advanced-stage bilateral POAG with irreversible vision loss. Pupillary light reflexes were absent, while fluctuations in pupil size were attributed to ketamine. Although medical and surgical options used in domestic cats may be applicable to non-domestic felids, frequent topical treatment was impracticable and unsafe, vision-preserving surgery was no longer indicated, and advanced diagnostics and specialist ophthalmic surgery were unavailable on site. The therapeutic goal was therefore shifted from vision preservation to long-term pain surveillance and welfare maintenance, including housing adaption and a defined threshold for humane euthanasia. Two types of neurological episodes were also documented as secondary findings; however, their causal relationship with the ocular disease could not be established without advanced neuroimaging. This case highlights the need for accessible on-site or referral-based ophthalmic diagnostics and surgical capacity in zoo settings. Full article

Machine learning models for medical image analysis are vulnerable to hidden confounders, which can compromise generalization and clinical reliability. Existing detection strategies typically require explicit knowledge or labels of the confounder, which are often unavailable. In this work, we propose an ensemble-based framework to detect potential confounder-driven learning without explicitly defining the confounders, but only which samples might be affected. Our approach leverages the variability of model performance across ensembles to identify signatures of shortcut learning. Shortcut learning occurs when a model uses non-robust features or correlations rather than learning the true underlying task, and it is often observed when confounders are present. We generate controlled dataset variants with increasing confounding levels and analyze distributions of AUC (area under the ROC curve) scores across training, validation, and test splits, revealing converging performance and reduced variance as confounding intensifies. We validate our method on two clinically relevant tasks, diabetic retinopathy detection from retinal fundus images and tumor detection from brain MRI slices. Then, we further demonstrate its practical utility on another dataset and image modality with a stroke reperfusion prediction task with suspected hidden confounders. This work provides a practical, data-driven diagnostic tool to flag potential confounding and support the reliability assessment of machine learning models in medical imaging. Full article

Background/Objectives: This retrospective cohort study compared hospitalization and follow-up rates in patients with newly diagnosed proliferative diabetic retinopathy (PDR) versus those without prior diabetic retinopathy (DR) screening. Methods: Using TriNetX, a global electronic health record database, 57,964 patients aged ≥ 40 years with type 2 diabetes and newly diagnosed PDR without diabetic macular edema (DME) requiring panretinal photocoagulation or intravitreal injection were included. Patients were stratified based on the presence or absence of prior DR screening in the last 5 years and balanced using propensity score matching (PSM). Primary outcomes included 30-, 60-, and 90-day hospitalization rates and repeat ophthalmic follow-up as estimated using repeat PDR diagnosis codes and repeat retinal imaging codes, including OCT, fundus photography, and fluorescein angiography. Results: Of 57,964 patients, 25,003 had no prior DR screening and 32,961 had prior DR screening. After matching, 19,316 patients were included per cohort. Patients without known DR screening had significantly higher hospitalization rates at 30 days (RR = 1.78, 95% CI 1.67–1.89), 60 days (RR = 1.59, 95% CI 1.51–1.67), and 90 days (RR = 1.51, 95% CI 1.44–1.58), and lower repeat ophthalmic visits by PDR codes at 30 days (RR = 0.458, 95% CI 0.440–0.476), 60 days (RR = 0.450, 95% CI 0.437–0.463) and 90 days (RR = 0.420, 95% CI 0.408–0.432) or by repeat retinal imaging codes at 30 days (RR = 0.450, 95% CI 0.423–0.478), 60 days (RR = 0.394, 95% CI 0.377–0.411), and 90 days (RR = 0.381, 95% CI 0.366–0.396) (all p < 0.0001). Conclusions: Absence of known prior DR screening in PDR patients is associated with higher hospitalization risk and reduced ophthalmic follow-up, suggesting that a lack of screening indicates broader gaps in healthcare engagement and disease control. Tailored strategies are needed to prevent vision loss as well as systemic complications. Full article

Background: Retinal and optic nerve disorders remain major causes of visual morbidity worldwide. Ocular fundus flavoprotein fluorescence (FPF) imaging has emerged as a potential noninvasive biomarker of mitochondrial dysfunction for earlier detection and evaluation of disease severity. Methods: We conducted a Systematic Scoping Review of the diagnostic and correlational utility of quantitative FPF parameters in retinal and optic nerve diseases compared with healthy controls. Following PRISMA-ScR guidelines, we searched MEDLINE, Web of Science, Scopus, and CENTRAL for peer-reviewed human studies available online before 31 December 2025. Results: Seventeen studies were included, encompassing 1914 eyes and 1339 participants, and were predominantly cross-sectional. In healthy eyes, mean macular and optic nerve head FPF intensity were reported as 24.1 ± 12.2 gsu and 30.6 ± 14.6 gsu, respectively. Higher signals were reported in several disorders, including diabetes mellitus (76.0 [67.0–92.0] gsu), neovascular age-related macular degeneration (67.47 ± 17.77 gsu), and retinitis pigmentosa (50.5 ± 12.2 gsu). However, lower, unchanged, or stage-dependent signals were also observed within the same disease categories. Interpretation across studies was limited by substantial heterogeneity in patient selection, disease definitions, imaging protocols, control groups, and FPF outcome metrics. The precise cellular and sublayer origin of the detected signal also remains challenging to determine. Conclusions: Ocular fundus FPF imaging provides promising metabolic insight into retinal and optic nerve diseases. However, current evidence remains heterogeneous and largely cross-sectional, limiting clinical interpretability and generalizability. Longitudinal studies, technical standardization, and multimodal integration are needed to define reproducible disease-specific FPF profiles and improve translational applicability. Full article

Ultra-wide field (UWF) fundus image classification is an important part of the entire process of medical screening and decision support. However, the discrimination of various retinal disease classes is difficult due to the similarity between classes, class imbalance, and the indeterminacy of visual patterns. In our research, an explainable neutrosophic knowledge distillation (NKD) model for UWF fundus image classification is proposed. In the proposed model, the teacher model is a ResNet50 architecture that provides the student model with supervisory information that is aware of the indeterminacy of predictions. The proposed model combines the CLAHE-based preprocessing method with the neutrosophic distillation method to enable the student model to learn from the hard labels as well as the teacher model. The experimental results were evaluated using the 5-fold cross-validation method with an additional hold-out evaluation. The experimental results show that the proposed NKD model has a mean accuracy of 84.00%, specificity of 97.33%, precision of 84.99%, recall of 84.00%, and F1-score of 84.02%. The proposed model also has an accuracy of 87.86% with specificity of 97.48% and AUC of 97.48% in the ablation-based full model evaluation. It outperformed classical machine learning baselines based on Local Binary Patterns (LBP), Histogram of Oriented Gradients (HOG), and LBP + HOG features with Support Vector Machines (SVM) classifiers, as well as the baseline student, fuzzy student, and teacher Convolutional Neural Network (CNN) models. For improved interpretability, the Grad-CAM++ technique was used to analyze the proposed NKD model. This analysis showed that the network attended to relevant retinal regions during classification. These results suggest that the proposed model can be an effective tool for UWF fundus image classification. Full article

Objectives: This study investigated whether visual field abnormalities are present in eyes with suspected nasal optic disc hypoplasia (NOH) by using fundus photography and optical coherence tomography (OCT). Methods: NOH was diagnosed using the following criteria: (1) small optic disc, (2) nasal optic disc pallor or optic disc margin irregularity, (3) wedge-shaped temporal visual field defects extending from Mariotte’s blind spot, and (4) reduced nasal circumpapillary retinal nerve fiber layer (cpRNFL) thickness. Eyes fulfilling criteria 1, 2, and 4 without visual field abnormalities were classified as pseudo-NOH (pNOH), whereas eyes without visual field or cpRNFL abnormalities were considered normal. Nasal cpRNFL thickness was measured using OCT, radial peripapillary capillary (RPC) density was assessed using OCT angiography (OCTA), visual field testing was performed, and optic disc blood flow velocity was evaluated using the mean blur rate (MBR) and laser speckle flowgraphy (LSFG). Results: Seven eyes with NOH, 13 eyes with pNOH, and 24 normal right eyes were included. Nasal cpRNFL thickness and MBR were significantly reduced in both the NOH and pNOH groups compared with the normal group, with no significant difference between the NOH and pNOH groups. Nasal RPC density was significantly lower in the NOH group than in both the pNOH and normal groups, and no significant difference was observed between the pNOH and normal groups. Conclusions: Even when NOH was suspected from fundus, LSFG, and OCT C-scan findings, visual field abnormalities were not consistently present. Differences in RPC density measured using OCTA may have contributed to this variability. This study examined whether suspected nasal optic disc hypoplasia (NOH) is always associated with visual field defects. Using fundus imaging, OCT, OCT angiography, and laser speckle flowgraphy, we compared eyes with NOH, pseudo-NOH, and normal eyes. Although structural changes such as reduced nasal nerve fiber layer thickness and decreased blood flow were observed in both NOH and pseudo-NOH, visual field abnormalities were not consistently present. Notably, reduced radial peripapillary capillary density was specific to NOH, suggesting that vascular differences may explain variability in visual function. These findings highlight the importance of multimodal imaging in NOH evaluation. Full article

Background/Objectives: General-purpose and domain-specific multimodal foundation models show considerable promise in medical image analysis. In this study, we evaluated the classification accuracy of diabetic retinopathy vs. normal fundus images using general-purpose conversational models (Gemini 3 Flash, GPT-5.2, and Pixtral-Large), a medical conversational model (MedGemma-1.5), and its image-encoder (MedSigLIP), as well as ophthalmology-specific models (RETFound and EyeCLIP). Methods: We applied zero-/few-shot to general-purpose conversational models, linear probing, and fine-tuning approaches to domain-specific models for evaluation purposes. Results: We found that the zero-shot accuracies for Pixtral-Large (70.7%) and fine-tuned RETFound (77.1%) were comparable but lower than those of GPT-5.2 (77.9%), MedGemma-1.5 (88.2%), and Gemini 3 (88.5%) as well as the fine-tuned EyeCLIP (85.8%) and MedSigLIP (94.8%). The accuracy gains from few-shot prompting were substantial for Pixtral-Large (+7.4%) but were limited for GPT-5.2 (+3.6%), Gemini 3 (−3.4%), and MedGemma-1.5 (−1.1%). Embedding-based linear probing further improved accuracy over fine-tuning for RETFound (+9.7%) and yielded only marginal gains for EyeCLIP (+2.3%) but did not benefit MedSigLIP (−0.8%). Overall, with minimal prompting enhancement, general-purpose conversational models such as Gemini 3 and GPT-5.2 achieved performance comparable to ophthalmology-specific models that were either fine-tuned or enhanced via embedding-based linear probing, but remained inferior to MedSigLIP and its conversational counterpart, MedGemma-1.5. Conclusions: The findings highlight a trade-off between specialization and flexibility, where domain-specific models provide higher accuracy and stability, while general-purpose multimodal models offer greater accessibility, adaptability, and interactive reasoning, serving as complementary tools for retinal disease screening and clinical decision support. Full article

Objectives: The aim of this study was to compare a widely applied table-top digital non-mydriatic camera (Topcon TRC-NW-8) with a handheld digital non-mydriatic camera (Optomed Aurora IQ) regarding the quantitative assessment of the retinal microcirculation using established biomarkers: central retinal arteriolar equivalent (CRAE), central retinal venular equivalent (CRVE) and arterio-venous ratio (AVR). Methods: The present cross-sectional study included 26 randomly selected participants (51 eyes) who underwent retinal imaging of both eyes with the two devices and were analyzed using a static retinal vessel analyzer. Results: The mean differences in CRAE, CRVE and AVR between the two devices (Topcon/Aurora) were 24.96 ± 11.7, 22.7 ± 11.7 and 0.026 ± 0.045, respectively. Strong correlations were observed between devices (r = 0.84 for CRAE, 0.75 for CRVE and 0.83 for AVR; all p < 0.001), with high agreement as indicated by ICC values (0.91, 0.85, and 0.90, respectively). Bland–Altman plots indicated evidence of systemic bias (95% within 2 SD) with no proportional bias, as the differences were consistently distributed across the range of average values. Regression-based equations were derived to approximate the transformation of measurements between devices. Conclusions: The handheld fundus camera demonstrates strong correlation and good relative agreement with the table-top device; however, a consistent device-dependent bias limits the direct interchangeability of absolute measurements. The derived transformation equations may facilitate approximate cross-device comparison, although external validation is required. These findings support the complementary use of handheld devices and highlight the need for calibration strategies when integrating measurements across platforms. Full article

Background/Objectives: Varicosities of the vortex vein ampulla are transient dilations of vortex vein ampullae that appear as red-brown choroidal masses. The purpose of this manuscript is to describe a retrospective case series of 53 patients with varicosities of the vortex vein ampulla and perform a literature review on this topic. Methods: Our case series demonstrates the clinical features of a large cohort of varicosities of the vortex vein ampulla, including their locations in the ocular fundus, sizes when congested, direction of gaze that resulted in detection, frequency of multiple lesions in a single eye, and frequency of bilateral cases. The literature review utilized PubMed and Embase libraries and included all studies published through December 2025. Results: The literature review yielded 44 articles, of which 36 were deemed relevant. Several studies described the appearance of these lesions using imaging modalities, including B-scan ultrasonography, optical coherence tomography, and indocyanine green angiography. Others underscored the potential for these lesions to be mistaken for other types of choroidal masses, such as choroidal melanomas. Conclusions: This extensive series demonstrates that these lesions are most often located nasally, sometimes multiple or bilateral, and often mistaken for choroidal nevi or melanomas, highlighting the importance of understanding clinical characteristics for appropriate diagnosis. In addition, some studies described possible associations with conditions such as nodular scleritis and Donnai–Barrow syndrome. Full article

Traditional deep learning-based models have achieved promising results in medical image segmentation. However, their performance degrades severely when applied to unseen domains due to variations in imaging protocols, acquisition devices, and patient populations across medical centers, which lead to significant distribution shifts. With the emergence of the Segment Anything Model (SAM), a single model now exhibits significantly improved generalization and adaptability to various image types. Nevertheless, while SAM has learned structure representations from large-scale natural images, it lacks fine-grained structural knowledge specific to the medical imaging domain, remaining relatively invariant across imaging domains. In addition, its structural enhancement is vulnerable to unreliable prompts, and patch-wise inference disrupts structural continuity, leading to suboptimal performance in capturing anatomical details. To address this, we propose a novel Medical Fine-grained Segment Anything Model (termed MedFineSAM), which integrates three key modules: shared fine-grained structural enhancement, which extracts and selectively enhances fine-grained structural features shared between prompts and image embeddings via a structural dictionary; a prompt gating mechanism, which estimates prompt confidence and dynamically adjusts prompt weights to avoid erroneous enhancement; and a structural continuity diffusion in frequency domain (SCFD), which performs frequency-domain smoothing during decoding to alleviate structural discontinuity caused by patch aggregation. Experiments on the fundus benchmark and prostate MRI benchmark demonstrate superior generalization performance, offering new insights into leveraging SAM for single-source domain generalization in medical image segmentation. Full article

Background and Clinical Significance: Blunt ocular trauma is a significant but often underestimated cause of visual impairment, particularly among adolescents involved in high-risk activities such as horseback riding. While most equestrian injuries affect the head and extremities, ocular trauma, especially commotio retinae, can result in severe visual complications. Case Presentation: We report the case of a 15-year-old girl who sustained blunt ocular trauma to the left eye following a fall from a horse and presented with decreased visual acuity. Multimodal imaging revealed outer retinal abnormalities on spectral-domain optical coherence tomography (OCT), including ellipsoid zone irregularities. Early-phase fluorescein angiography showed central hypofluorescence in the foveal region with surrounding mild mottled hyperfluorescence, without clear vascular abnormalities. Fundus photography demonstrated subtle macular changes. Visual acuity improved significantly following treatment, with partial resolution of macular changes, although mild outer retinal irregularities persisted on follow-up imaging. Conclusions: These findings underscore the importance of early ophthalmic evaluation and advanced retinal imaging in blunt ocular trauma. Given the high risk of visual injury during equestrian activities, especially in pediatric and adolescent populations, preventive strategies such as mandatory helmet use and rider education are essential. Implementation of standardized follow-up protocols is also recommended to monitor long-term retinal changes in patients with traumatic maculopathy. Full article