Vision (Switzerland)最新文献

Deep learning for glaucoma screening often relies on high-resolution clinical images and convolutional neural networks (CNNs). However, these methods face significant performance drops when applied to noisy, low-resolution images from portable devices. To address this, our work investigates ensemble methods using multiple Transformer architectures for automated glaucoma detection in challenging scenarios. We use the Brazil Glaucoma (BrG) and private D-Eye datasets to assess model robustness. These datasets include images typical of smartphone-coupled ophthalmoscopes, which are often noisy and variable in quality. Four Transformer models-Swin-Tiny, ViT-Base, MobileViT-Small, and DeiT-Base-were trained and evaluated both individually and in ensembles. We evaluated the results at both image and patient levels to reflect clinical practice. The results show that, although performance drops on lower-quality images, ensemble combinations and patient-level aggregation significantly improve accuracy and sensitivity. We achieved up to 85% accuracy and an 84.2% F1-score on the D-Eye dataset, with a notable reduction in false negatives. Grad-CAM attention maps confirmed that Transformers identify anatomical regions relevant to diagnosis. These findings reinforce the potential of Transformer ensembles as an accessible solution for early glaucoma detection in populations with limited access to specialized equipment.

Models suggest social anxiety is characterized by negative processing biases. Negative biases also arise from negative mood, i.e., state affect. We examined how social anxiety influences emotional processing and whether state affect, or mood, modified the relationship between social anxiety and perceptual bias. We quantified bias by determining the point of subjective equality, PSE, the face judged equally often as happy and as angry. We found perceptual bias depended on social anxiety and state affect. PSE was greater in individuals high (mean PSE: 8.69) versus low (mean PSE: 3.04) in social anxiety. The higher PSE indicated a stronger negative bias in high social anxiety. State affect modified this relationship, with high social anxiety associated with stronger negative biases, but only for individuals with greater negative affect. State affect and trait anxiety interacted such that social anxiety status alone was insufficient to fully characterize perceptual biases. This raises several issues such as the need to consider what constitutes an appropriate control group and the need to consider state affect in social anxiety. Importantly, our results suggest compensatory effects may counteract the influences of negative mood in individuals low in social anxiety.

We provide an automated characterization of human retinal cells, i.e., RPE's based on the non-invasive AO-TFI retinal imaging and PR's based on the non-invasive AO-FI retinal imaging on a large-scale study involving 171 confirmed healthy eyes from 104 participants of 23 to 80 years old. Comprehensive standard checkups based on SD-OCT and Fondus imaging modalities were carried out by Ophthalmologists from the Luzerner Kantonsspital (LUKS) to confirm the absence of retinal pathologies. AO imaging imaging was performed using the Cellularis^® device and each eye was imaged at various retinal eccentricities. The images were automatically segmented using a dedicated software and RPE and PR cells were identified and morphometric characterizations, such as cell density and area were computed. The results were stratified based on various criteria, such as age, retinal eccentricity, visual acuity, etc. The automatic segmentation was validated independently on a held-out set by five trained medical students not involved in this study. We plotted cell density variations as a function of eccentricity from the fovea along both nasal and temporal directions. For RPE cells, no consistent trend in density was observed between 0° to 9° eccentricity, contrasting with established histological literature demonstrating foveal density peaks. In contrast, PR cell density showed a clear decrease from 2.5° to 9°. RPE cell density declined linearly with age, whereas no age-related pattern was detected for PR cell density. On average, RPE cell density was found to be ≈6313 cells/mm² (±σ=757), while the average PR cell density was calculated as ≈10,207 cells/mm² (±σ=1273).

Background: This paper aims to provide an overview of corneal birefringence (CB), systematize the knowledge and current understanding of CB, and identify difficulties associated with introducing CB into mainstream clinical practice.

Methods: Literature reviews were conducted, seeking articles focused on CB published between the early 19th century and the present time. Secondary-level searches were made examining relevant publications referred to in primary-level publications, ranging back to the early 17th century. The key search words were "corneal birefringence" and "non-invasive measurements".

Results: CB was first recorded by Brewster in 1815. Orthogonally polarized rays travel at different speeds through the cornea, creating a slow axis and a fast axis. The slow axis aligns with the pattern of most corneal stromal collagen fibrils. In vivo, it is oriented along the superior temporal-inferior nasal direction at an angle of about 25° (with an approximate range of -54° to 90°) from the horizontal. CB has been reported to (i) influence the estimation of retinal nerve fiber layer thickness; (ii) be affected by corneal interventions; (iii) be altered in keratoconus; (iv) vary along the depth of the cornea; and (v) be affected by intra-ocular pressure.

Conclusions: Under precisely controlled conditions, capturing the CB pattern is the first step in a non-destructive process used to model the ultra-fine structure of the individual cornea, and changes thereof, in vivo.

Purpose: This study aimed to determine the prevalence and associated risk factors of keratoconus (KC) among high school students in Couva, Trinidad and Tobago. Method: A cross-sectional, school-based approach was used, involving a simple random sampling technique to select schools and students. A structured questionnaire assessed KC risk factors, while clinical assessments, including visual acuity, refraction, slit lamp biomicroscopy, and topography, were performed. Data were analyzed using R. Exact tests were used for KC (n = 2 cases) and robust Poisson regression estimated adjusted prevalence ratios for the 'at-risk' screening endpoint. Results: A total of 432 students aged 12-17 years participated, with a response rate of 97.5%. Most participants were of East Indian descent (48.1%), female (52.1%), and 14 years old (23.1%). Approximately 47.7% (95% CI 43.0-52.5%) were at risk of KC, with 0.5% (2/432; exact 95% CI 0.06-1.67%) diagnosed with the condition. The most common risk factors were eye rubbing (87.4%), over eight hours of sun exposure weekly (71.8%), and atopy (68.4%). KC was observed to be significantly higher among people with a family history (p = 0.018). Conclusions: The study highlights a low prevalence and a high risk of KC among high school students, with a strong link to family history and common risk factors such as eye rubbing and sun exposure. These findings emphasize the urgent need for regular KC screening in schools to ensure early diagnosis and effective management.

Face recognition is an important skill that helps people make social judgments by identifying both who a person is and other characteristics such as their expression, age, and ethnicity. Previous models of face processing, such as those proposed by Bruce and Young and by Haxby and colleagues, suggest that identity and other facial features are processed through partly independent systems. This study aimed to compare the efficiency with which different facial characteristics are processed in a visual search task. Participants viewed arrays of two, four, or six faces and judged whether one face differed from the others. Four tasks were created, focusing separately on identity, expression, ethnicity, and gender. We found that search times were significantly longer when looking for identity and shorter when looking for ethnicity. Significant correlations were found among almost all tests in all outcome variables. Comparison of target-present and target-absent trials suggested that performance in none of the tests seems to follow a serial-search-terminating model. These results suggest that different facial characteristics share early processing but differentiate into independent recognition mechanisms at a later stage.

Glaucoma management relies on lowering intraocular pressure (IOP), but determining the target reduction at presentation is challenging, particularly in normal-tension glaucoma (NTG). We developed and internally validated a neural network regression model using retrospective clinical data from Qiu et al. (2015), including 270 patients (118 with NTG). A single-layer artificial neural network with five nodes was trained in MATLAB R2024b using the Levenberg-Marquardt algorithm. Inputs included demographic, refractive, structural, and functional parameters, with IOP reduction as the output. Data were split into 65% training, 15% validation, and 20% testing, with training repeated 10 times. Model performance was strong and consistent (average RMSE: 1.90 ± 0.29 training, 2.18 ± 0.34 validation, 2.11 ± 0.30 testing; Pearson's r: 0.92 ± 0.02, 0.88 ± 0.02, 0.88 ± 0.04). The best-performing model achieved RMSEs of 1.57, 2.90, and 1.77 with r values of 0.93, 0.91, and 0.93, respectively. Feature ablation revealed significant contributions from IOP, axial length, CCT, diagnosis, VCDR, spherical equivalent, mean deviation, and laterality. This study demonstrates that a simple neural network can reliably predict individualized IOP reduction targets, supporting personalized glaucoma management and improved outcomes.

Background: This study compared the performance of a Head-mounted Virtual Reality Perimeter (HVRP) with the Humphrey Field Analyzer (HFA), the standard in automated perimetry. The HFA is the established standard for automated perimetry but is constrained by lengthy testing, bulky equipment, and limited patient comfort. Comparative data on newer head-mounted virtual reality perimeters are limited, leaving uncertainty about their clinical reliability and potential advantages. Aim: The aim was to evaluate parameters such as visual field outcomes, portability, patient comfort, eye tracking, and usability. Methods: Participants underwent testing with both devices, assessing metrics like mean deviation (MD), pattern standard deviation (PSD), and duration. Results: The HVRP demonstrated small but statistically significant differences in MD and PSD compared to the HFA, while maintaining a consistent trend across participants. MD values were slightly more negative for HFA than HVRP (average difference -0.60 dB, p = 0.0006), while pattern standard deviation was marginally higher with HFA (average difference 0.38 dB, p = 0.00018). Although statistically significant, these differences were small in magnitude and do not undermine the clinical utility or reproducibility of the device. Notably, HVRP showed markedly shorter testing times with HVRP (7.15 vs. 18.11 min, mean difference 10.96 min, p < 0.0001). Its lightweight, portable design allowed for bedside and home testing, enhancing accessibility for pediatric, geriatric, and mobility-impaired patients. Participants reported greater comfort due to the headset design, which eliminated the need for chin rests. The device also offers potential for AI integration and remote data analysis. Conclusions: The HVRP proved to be a reliable, user-friendly alternative to traditional perimetry. Its advantages in comfort, portability, and test efficiency support its use in both clinical settings and remote screening programs for visual field assessment. Its portability and user-friendly design support broader use in clinical practice and expand possibilities for bedside assessment, home monitoring, and remote screening, particularly in populations with limited access to conventional perimetry.

This study aimed to evaluate the performance of seven advanced AI Large Language Models (LLMs)-ChatGPT 4o, ChatGPT O3 Mini, ChatGPT O1, DeepSeek V3, DeepSeek R1, Gemini 2.0 Flash, and Grok-3-in answering multiple-choice questions (MCQs) in optics and refractive surgery, to assess their role in medical education for residents. The AI models were tested using 134 publicly available MCQs from national ophthalmology certification exams, categorized by the need to perform calculations, the relevant subspecialty, and the use of images. Accuracy was analyzed and compared statistically. ChatGPT O1 achieved the highest overall accuracy (83.5%), excelling in complex optical calculations (84.1%) and optics questions (82.4%). DeepSeek V3 displayed superior accuracy in refractive surgery-related questions (89.7%), followed by ChatGPT O3 Mini (88.4%). ChatGPT O3 Mini significantly outperformed others in image analysis, with 88.2% accuracy. Moreover, ChatGPT O1 demonstrated comparable accuracy rates for both calculated and non-calculated questions (84.1% vs. 83.3%). This is in stark contrast to other models, which exhibited significant discrepancies in accuracy for calculated and non-calculated questions. The findings highlight the ability of LLMs to achieve high accuracy in ophthalmology MCQs, particularly in complex optical calculations and visual items. These results suggest potential applications in exam preparation and medical training contexts, while underscoring the need for future studies designed to directly evaluate their role and impact in medical education. The findings highlight the significant potential of AI models in ophthalmology education, particularly in performing complex optical calculations and visual stem questions. Future studies should utilize larger, multilingual datasets to confirm and extend these preliminary findings.

We previously developed machine learning (ML) models for predicting cycloplegic spherical equivalent refraction (SER) and myopia using non-cycloplegic data and following a standardized protocol (cycloplegia with 0.5% tropicamide and biometry using NIDEK A-scan), but the models' performance may not be generalizable to other settings. This study evaluated the performance of ML models in an independent cohort using a different cycloplegic agent and biometer. Chinese students (N = 614) aged 8-13 years underwent autorefraction before and after cycloplegia with 0.5% tropicamide (n = 505) or 1% cyclopentolate (n = 109). Biometric measures were obtained using an IOLMaster 700 (n = 207) or Optical Biometer SW-9000 (n = 407). ML models were evaluated using R², mean absolute error (MAE), sensitivity, specificity, and area under the ROC curve (AUC). The XGBoost model predicted cycloplegic SER very well (R² = 0.95, MAE (SD) = 0.32 (0.30) D). Both ML models predicted myopia well (random forest: AUC 0.99, sensitivity 93.7%, specificity 96.4%; XGBoost: sensitivity 90.1%, specificity 96.8%) and accurately predicted the myopia rate (observed 62.9%; random forest: 60.6%; XGBoost: 58.8%) despite heterogeneous cycloplegia and biometry factors. In this independent cohort of students, XGBoost and random forest performed very well for predicting cycloplegic SER and myopia status using non-cycloplegic data. This external validation study demonstrated that ML may provide a useful tool for estimating cycloplegic SER and myopia prevalence with heterogeneous clinical parameters, and study in additional populations is warranted.