OBJECTIVE: To develop and externally test deep learning (DL) models for assessing the image quality of three-dimensional (3D) macular scans from Cirrus and Spectralis optical coherence tomography devices.
METHODS: We retrospectively collected two data sets including 2277 Cirrus 3D scans and 1557 Spectralis 3D scans, respectively, for training (70%), fine-tuning (10%) and internal validation (20%) from electronic medical and research records at The Chinese University of Hong Kong Eye Centre and the Hong Kong Eye Hospital. Scans with various eye diseases (eg, diabetic macular oedema, age-related macular degeneration, polypoidal choroidal vasculopathy and pathological myopia), and scans of normal eyes from adults and children were included. Two graders labelled each 3D scan as gradable or ungradable, according to standardised criteria. We used a 3D version of the residual network (ResNet)-18 for Cirrus 3D scans and a multiple-instance learning pipline with ResNet-18 for Spectralis 3D scans. Two deep learning (DL) models were further tested via three unseen Cirrus data sets from Singapore and five unseen Spectralis data sets from India, Australia and Hong Kong, respectively.
RESULTS: In the internal validation, the models achieved the area under curves (AUCs) of 0.930 (0.885-0.976) and 0.906 (0.863-0.948) for assessing the Cirrus 3D scans and Spectralis 3D scans, respectively. In the external testing, the models showed robust performance with AUCs ranging from 0.832 (0.730-0.934) to 0.930 (0.906-0.953) and 0.891 (0.836-0.945) to 0.962 (0.918-1.000), respectively.
CONCLUSIONS: Our models could be used for filtering out ungradable 3D scans and further incorporated with a disease-detection DL model, allowing a fully automated eye disease detection workflow.

OBJECTIVE: To investigate the effect of preretinal tractional structures (PTS) and posterior scleral structures (PSS) on myopic traction maculopathy (MTM) progression.
METHODS: This retrospective cohort study included 185 fellow highly myopic eyes of 185 participants who underwent surgery for MTM. PTS included epiretinal membrane, incomplete posterior vitreous detachment and their combination. PSS included posterior staphyloma and dome-shaped macula (DSM). The MTM stage was graded according to the Myopic Traction Maculopathy Staging System. Optical coherence tomography was used to identify MTM progression, defined as an upgrade of MTM. The Kaplan-Meier method with log-rank test was used to assess MTM progression over the 3-year follow-up period. Risk factors for progression were identified using Cox regression analysis.
RESULTS: MTM progression was observed in 48 (25.9%) eyes. Three-year progression-free survival (PFS) rates for eyes with PTS, staphyloma and DSM were 53.7%, 58.2% and 90.7%, respectively. Eyes with PTS and staphyloma exhibited lower 3-year PFS rates than those without PTS or staphyloma (P log-rank test =0.002 and <0.001), while eyes with DSM had a higher 3-year PFS rate than eyes without DSM (P log-rank test=0.01). Multivariate Cox regression analysis showed that PTS (HR, 3.23; p<0.001) and staphyloma (HR, 7.91; p<0.001) were associated with MTM progression, whereas DSM (HR, 0.23; p=0.046) was a protective factor.
CONCLUSIONS: Both PTS and PSS play a critical role in the progression of MTM. Addressing these factors can aid in the management of MTM.

OBJECTIVE: To compare the performance of generative versus retrieval-based chatbots in answering patient inquiries regarding age-related macular degeneration (AMD) and diabetic retinopathy (DR).
METHODS: We evaluated four chatbots: generative models (ChatGPT-4, ChatGPT-3.5 and Google Bard) and a retrieval-based model (OcularBERT) in a cross-sectional study. Their response accuracy to 45 questions (15 AMD, 15 DR and 15 others) was evaluated and compared. Three masked retinal specialists graded the responses using a three-point Likert scale: either 2 (good, error-free), 1 (borderline) or 0 (poor with significant inaccuracies). The scores were aggregated, ranging from 0 to 6. Based on majority consensus among the graders, the responses were also classified as \'Good\', \'Borderline\' or \'Poor\' quality.
RESULTS: Overall, ChatGPT-4 and ChatGPT-3.5 outperformed the other chatbots, both achieving median scores (IQR) of 6 (1), compared with 4.5 (2) in Google Bard, and 2 (1) in OcularBERT (all p ≤8.4×10-3). Based on the consensus approach, 83.3% of ChatGPT-4\'s responses and 86.7% of ChatGPT-3.5\'s were rated as \'Good\', surpassing Google Bard (50%) and OcularBERT (10%) (all p ≤1.4×10-2). ChatGPT-4 and ChatGPT-3.5 had no \'Poor\' rated responses. Google Bard produced 6.7% Poor responses, and OcularBERT produced 20%. Across question types, ChatGPT-4 outperformed Google Bard only for AMD, and ChatGPT-3.5 outperformed Google Bard for DR and others.
CONCLUSIONS: ChatGPT-4 and ChatGPT-3.5 demonstrated superior performance, followed by Google Bard and OcularBERT. Generative chatbots are potentially capable of answering domain-specific questions outside their original training. Further validation studies are still required prior to real-world implementation.

OBJECTIVE: To distinguish the clinical feature of nanophthalmos (NNO) caused by mutations in protease serine 56 (PRSS56), membrane-type frizzled-related protein (MFRP), myelin regulatory factor (MYRF) and transmembrane protein 98 (TMEM98) and to evaluate the association between angle-closure glaucoma (ACG) and NNO.
METHODS: Variants in those four genes were identified through exome sequencing/whole genome sequencing data, and bioinformatic analysis was conducted to identify pathogenic/likely pathogenic (P/LP) variants. This observational study comprehensively summarised ophthalmological data of 67 patients with NNO from 63 families. Ocular parameters from 68 eyes without surgical treatment were subjected to further analysis.
RESULTS: Totally, 67 patients from 63 families harboured 57 P/LP variants in the four genes, including 30 in PRSS56 (47.6%), 23 in MFRP (36.5%), 5 in TMEM98 (7.9%) and 5 in MYRF (7.9%). ACG was present in 79.1% of patients. An analysis of ocular parameters from 68 eyes revealed that shorter axial length (AL), lower vitreous-to-AL ratios and severe foveal hypoplasia were associated with variants in PRSS56 and MFRP. Uveal effusion was more common in patients with PRSS56 variants, while retinitis pigmentosa was frequently observed in patients with MFRP variants. Patients with MYRF variants exhibited the thinnest retinal nerve fibre layer thickness. Patients with TMEM98 variants had an earlier average onset age of glaucoma.
CONCLUSIONS: Variants in PRSS56 and MFRP are the most common genetic cause of NNO. ACG is a severe complication frequently observed in these patients. Earlier onset of ACG is observed in patients with dominant NNO, while foveal hypoplasia is more common in patients with recessive disease. Recognising these features is helpful in clinical care and genetic counselling.

OBJECTIVE: To evaluate the diagnostic accuracy of spectral-domain optical coherence tomography (SD OCT) combined with OCT angiography (OCTA) for myopic myopic macular neovascularisation (MNV) activity.
METHODS: Both eyes of patients with myopic MNV diagnosed with fluorescein angiography (FA), SD OCT and OCTA were assessed by unmasked investigators. The images were deidentified and randomised before graded by masked investigators, who determined the presence of active myopic MNV by using SD OCT together with OCTA without FA and by FA alone, respectively. The findings of masked investigators were compared with unmasked investigators.
RESULTS: 213 eyes of 110 patients comprising 499 imaging episodes were eligible for grading. For diagnosing new-onset myopic MNV without FA, combined use of SD OCT and OCTA had a sensitivity of 0.94, specificity of 0.84 and area under the curve (AUC) of 0.92. FA had a sensitivity of 0.52 (p<0.01), specificity of 0.80 (p=0.38) and AUC of 0.66 (p<0.01). For recurrent myopic MNV, the combination of SD OCT and OCTA had a sensitivity of 0.98, specificity of 0.78 and AUC of 0.88. FA had a sensitivity of 0.50 (p=0.04), specificity of 0.76 (p=0.85) and AUC of 0.63 (p=0.01). Myopic traction maculopathy was more frequently associated with recurrent myopic MNV (p<0.01).
CONCLUSIONS: SD OCT with dense volumetric scan was highly sensitive for diagnosing myopic MNV. The addition of OCTA improved the diagnostic specificity without FA. Monitoring of the longitudinal changes on SD OCT and judicious use of FA is a reliable surveillance strategy for myopic MNV.

OBJECTIVE: To investigate macula structure thickness and volume changes in patients with moyamoya disease (MMD).
METHODS: In this cross-sectional study, we used artificial intelligence (AI) -assisted optical coherence tomography (OCT) to analyze the thickness and volume of macula in Moyamoya patients. ETDRS zoning divides the macula into nine different regions. In 15 patients with radial scanning OCT, the average thickness and volume of retina, RNFL, GCL, and choroid in these regions were measured. In 30 patients with radial or horizontal scanning OCT, based on the anatomical structure, the macula is divided into seven segments. Mean Sattler layer-choriocapillaris complex thickness (SLCCT), Haller layer thickness, and total choroidal thickness were measured for each segment using AI-assisted OCT.
RESULTS: We recruited 30 patients (59 eyes) with MMD. In the 15 patients (29 eyes) who underwent radial scanning OCT, no significant change in retina, RNFL, GCL, and choroidal thickness was identified between the two groups (p > 0.05). There was no significant change in retina, RNFL, or choroidal volume between the two groups in different ETDRS macula regions (p > 0.05). The GCL volume in the macula\'s inner ring nasal portion (IN) was significantly lower. SLCCTs were considerably reduced in six macula regions in moyamoya groups (p < 0.05). There was no statistically significant change in Haller layer thickness. Only the nasal perifovea (PE_N) exhibited a significant variation in choroidal thickness. The Moyamoya group showed reduced choroidal thickness in PE_N segment.
CONCLUSIONS: In patients with MMD, there is thinning of the Sattler layer-choriocapillaris complex in the choroid.

To develop a Vision Transformer model to detect different stages of diabetic maculopathy (DM) based on optical coherence tomography (OCT) images.
After removing images with poor quality, a total of 3319 OCT images were extracted from the Eye Center of the Renmin Hospital of Wuhan University and randomly split the images into training and validation sets in a 7:3 ratio. All macular cross-sectional scan OCT images were collected retrospectively from the eyes of DM patients from 2016 to 2022. One of the OCT stages of DM, including early diabetic macular oedema (DME), advanced DME, severe DME and atrophic maculopathy, was labelled on the collected images, respectively. A deep learning (DL) model based on Vision Transformer was trained to detect four OCT grading of DM.
The model proposed in our paper can provide an impressive detection performance. We achieved an accuracy of 82.00%, an F1 score of 83.11%, an area under the receiver operating characteristic curve (AUC) of 0.96. The AUC for the detection of four OCT grading (ie, early DME, advanced DME, severe DME and atrophic maculopathy) was 0.96, 0.95, 0.87 and 0.98, respectively, with an accuracy of 90.87%, 89.96%, 94.42% and 95.13%, respectively, a precision of 88.46%, 80.31%, 89.42% and 87.74%, respectively, a sensitivity of 87.03%, 88.18%, 63.39% and 89.42%, respectively, a specificity of 93.02%, 90.72%, 98.40% and 96.66%, respectively and an F1 score of 87.74%, 84.06%, 88.18% and 88.57%, respectively.
Our DL model based on Vision Transformer demonstrated a relatively high accuracy in the detection of OCT grading of DM, which can help with patients in a preliminary screening to identify groups with serious conditions. These patients need a further test for an accurate diagnosis, and a timely treatment to obtain a good visual prognosis. These results emphasised the potential of artificial intelligence in assisting clinicians in developing therapeutic strategies with DM in the future.

OBJECTIVE: To investigate the changes in macular vessel density (VD) of the superficial layer of retina (SLR) and deep layer of retina (DLR) in dysthyroid optic neuropathy (DON) after high-dose intravenous pulse methylprednisolone (IVMP).
METHODS: Eighteen DON patients (29 eyes) who completed high-dose IVMP and 16 healthy individuals (32 eyes) were enrolled in this study. Optical coherence tomography angiography (OCTA) image analysis and comprehensive ophthalmic examinations were performed, including the SLR macular whole-image VD (SLR-mwiVD) and DLR-mwiVD, best-corrected visual acuity (BCVA), the mean deviation of visual field (VF-MD), pattern standard deviation of visual field (VF-PSD) and the other parameters.
RESULTS: The SLR-mwiVD (41.39 ± 4.71 vs. 48.13 ± 3.68, p < 0.001) and DLR-mwiVD (40.77 ± 5.85 vs. 49.14 ± 7.02, p < 0.001) were decreased in DON compared to control eyes. After IVMP, visual function parameters were improved, and SLR-mwiVD (49.41 ± 3.18, p < 0.001) and DLR-mwiVD (50.41 ± 4.04, p < 0.001) were increased in the DON group compared to pretreatment. The increased SLR-mwiVD and DLR-mwiVD were significantly correlated with improvements in BCVA (Log MAR: from 0.62 ± 0.49 to -0.01 ± 0.03, p < 0.001), VF-MD (from - 6.89 ± 2.89 dB to - 1.75 ± 1.29 dB, p < 0.001) and VF-PSD (from 4.38 ± 2.52 dB to 2.32 ± 1.64 dB, p < 0.001).
CONCLUSIONS: The increase in macular VD was significantly correlated with the improvement in visual function in DON after IVMP. Macular VD changes on OCTA may be a useful indicator for the response in DON after IVMP.

OBJECTIVE: To qualitatively and quantitatively characterise the genotypes and phenotypes of Bietti\'s crystalline dystrophy (BCD) in a cohort of patients.
METHODS: Cross-sectional and observational study.
METHODS: Clinically confirmed BCD patients were recruited for genotyping and phenotyping. Multiple retinal imaging modalities were employed. Atrophy in the fovea was adopted as major consideration for staging strategy, while percentage area of autofluorescence (AF) atrophy (PAFA) in the macula was determined for quantitation.
RESULTS: In 74 clinically diagnosed BCD patients, c.802-8_810del17insGC was shown the predominant variant of the CYP4V2 gene (allele frequency 55.4%). Sixty-two cases (123 eyes) with full imaging data were classified according to a modified criterion into stages 1 (n=8, 6.50%), 2A (n=9, 7.32%), 2B (n=17, 13.82%), 3A (n=30, 24.39%) and 3B (n=59, 47.97%). The eyes of the stage 2B were particularly deemed \'high risk\' due to atrophy near fovea, while in stage 3A, though with remarkable foveal atrophy, preserved retinal pigment epithelium/photoreceptor islands near the fovea were found in 14 eyes. A tendency of increase in PAFA with age was found (rs=0.31, p=0.014). Significant PAFA increase was shown through stages 1 to 3B, and best-corrected visual acuity (BCVA, Logarithm of the Minimum Angle of Resolution) was shown to moderately correlate with PAFA (rs=0.56, p<0.001).
CONCLUSIONS: The PAFA might be an efficient biomarker for BCD severities correlating with BCVA. The highly heterogeneous chorioretinopathy and BCVA of BCD cases appear to be associated with disease stages, progression types and patients\' ages. Foveal involvement should be of a major concern for consideration of potential therapeutic intervention.

Nonhuman primates are important research models for basic vision research, preclinical pathogenesis, and treatment studies due to strong similarities in retinal structure and function with humans. We compared retinal parameters between 10 healthy normal rhesus macaques (Macaca mulatta) and 10 cynomolgus macaques (Macaca fascicularis) by optical coherence tomography and electroretinography. The Heidelberg Spectralis® HRA+OCT and Roland multifocal electrophysiometer were used to analyze retinal morphology, multifocal electroretinograms (mfERGs), and full-field electroretinograms (ff-ERGs). Mean retinal thickness was lowest in the central fovea of macaques and did not differ significantly between species, but the retinal thicknesses of the nerve fiber ganglion cell layer and the inner plexiform layer were significantly different. The amplitude density of the N1 wave was lower in rhesus macaques than in cynomolgus macaques in ring and quadrant areas. Dark-adapted 3.0 oscillatory potentials (reflection of amacrine cell activity) and light-adapted 30-hz flicker ERG (a sensitive cone-pathway-driven response) waveforms of the ff-ERG were similar in both species, while the times to peaks in dark-adapted 0.01 ERG (the rod-driven response of bipolar cells) and dark-adapted 3.0 ERG (combined rod and cone system responses) as well as the implicit times of the a- and b-waves in light-adapted 3.0 ERG (the single-flash cone response) were substantially different. This study provides normative retinal parameters for nonhuman primate research on basic and clinical ophthalmology, as well as a reference for researchers in the appropriate selection of rhesus or cynomolgus macaques as models for ophthalmology studies.