BMJ Open Ophthalmology最新文献

Purpose: To develop an artificial intelligence (AI) system for detecting pathological patterns of diabetic macular oedema (DME) with fine-grained image categorisation using optical coherence tomography (OCT) images.

Methods: The development of the AI system consists of two parts, a pretraining process on a public dataset (Asia Pacific Tele-Ophthalmology Society (APTOS)), and the training process on the local dataset. The local dataset was partitioned into the training set, validation set and test set in the ratio of 6:2:2. The Split Subspace Attention Network (SSA-Net) architecture was adopted to train independent models to detect the seven pathological patterns of DME: intraretinal fluid (IRF), subretinal fluid (SRF), pigment epithelial detachment (PED), hyper-reflective retinal foci (HRF), Müller cell cone disruption (MCCD), subretinal hyper-reflective material (SHRM) and intra-cystic hyper-reflective material (ICHRM). The confusion matrix, sensitivity, specificity and receiver operating characteristic (ROC) were used to evaluate the performance of the models.

Results: The APTOS public dataset consists of 33 853 OCT images and the local dataset consists of 1346 OCT images with DME. In the pretraining process on the public dataset, the accuracy was 0.652 for IRF, 0.928 for SRF, 0.936 for PED and 0.975 for HRF. After the training process on the local dataset, the SSA-Net architecture showed better performance in fine-grained image categorisation on the test set. The area under the ROC curve was 0.980 for IRF, 0.995 for SRF, 0.999 for PED, 0.908 for MCCD, 0.887 for HRF, 0.990 for SHRM and 0.972 for ICHRM. The AI system outputs a heatmap for each result, which can give a visual explanation for the automated identification process. The heatmaps revealed that the regions of interest of the AI system were consistent with the retinal experts.

Conclusions: The proposed SSA-Net architecture for detecting the pathological patterns of DME achieved satisfactory accuracy. However, this study was conducted in one medical centre, and multicentre trials will be needed in the future.

Objective: Facial nerve paralysis (FNP) affects eye muscle function causing lagophthalmos. This leaves the cornea vulnerable to irritation and injury which, if untreated, may cause permanent vision impairment. Decreased eye muscle control impacts facial expressions and quality of life. Although non-surgical interventions such as lubrication are routinely used in standard eyecare, patient perspectives regarding these interventions have not been extensively investigated. This study aims to investigate patients' perspectives of conservative ophthalmic management for FNP.

Methods and analysis: Participants with FNP completed a survey including seven conservative lagophthalmos management interventions: artificial tears, lubricating ointment, taping the eye shut, tape to lift the lower eyelid, scleral contact lens, moisture chamber and manual eye closure. Open-ended questions and Likert scales covered six domains: discomfort, appearance-related concerns, application difficulties, impact on activities of daily living and social activities, and tolerance to outdoor settings. Descriptive, bivariate, correlation and thematic analyses were undertaken.

Results: 49 participants completed the survey (17 males, 32 females; mean age 55 years, SD 14.94). Mean duration of FNP was 8 years, 5 months (range 12 months to 47 years). Artificial tears and lubricating ointments were most frequently used with highest overall satisfaction. Participants were less satisfied with interventions involving the use of tape. Manual eye closure had a small positive impact. Main themes were comfort, vision and appearance.

Conclusions: Functional, social and psychological impairments associated with lagophthalmos are managed using conservative interventions. A better understanding of a patient's capacity, priorities and preferences underpins a collaborative approach when selecting eye-protective interventions and optimises patient outcomes.The main limitation of this study was that participants were recruited from a specialty facial nerve clinic, resulting in an over-representation of individuals with long-standing, complex conditions.

Purpose: Bietti crystalline corneoretinal dystrophy (BCD) is an autosomal recessive progressive retinal degenerative disease due to mutations in the CYP4V2 gene. Best-corrected visual acuity (BCVA) is a common primary endpoint in clinical trials for retinal diseases, but the natural history of BCVA loss remains unclear because of the heterogeneity of manifestations in BCD patients.

Methods: Based on the individual data of untreated BCD patients, a disease progression model was established using the change in BCVA from baseline as an index, and covariates including age of onset, age, duration of disease, baseline BCVA, gender, race (East Asian/non-East Asian), genotype, and family history. Then, based on the final model, the natural disease progression characteristics of BCD were simulated.

Result: A total of 14 studies met the inclusion criteria, with a total sample size of 117 cases, including 6 studies (N=80) with East Asian populations and 9 studies (N=37) with non-East Asian populations. The change of BCVA from baseline increased linearly with time, and the disease progression model of BCD was successfully established. BCVA increased by 0.06 logarithm of the minimum angle of resolution (LogMAR) per year in BCD patients. BCVA increased by 0.09 LogMAR per year in patients with BCVA≥0.5LogMAR and disease duration more than 10 years.

Conclusions: For the first time, we successfully established a BCD disease progression model based on the change in BCVA from baseline. The mean visual acuity loss increased linearly with the progression of the disease. A sharper loss of BCVA may be expected in patients with BCVA≥0.5LogMAR and disease duration ≥10 years.

Objective: Glaucoma is a complex neurodegenerative ocular disorder accompanied by brain functional abnormalities that extend beyond the visual system. However, the causal association between the two remains unclear at present. This study aimed to investigate the potential causal relationships between glaucoma and brain functional networks in order to provide novel insights into the neuropathic mechanism of glaucoma.

Methods and analysis: Based on the genome-wide association studies data of glaucoma and resting-state functional MRI (Rs-fMRI), a bidirectional Mendelian randomisation (MR) analysis was conducted between glaucoma and brain functional networks. Inverse variance weighting was applied as the primary method to estimate causality with false discovery rate correction. Additional sensitivity analyses were conducted to evaluate the robustness of the results.

Results: Forward MR analysis suggested that glaucoma was causally associated with two brain networks between the subcortical cerebellum and the attention or visual network (p=0.022), as well as the default mode and central executive network (p=0.008), but without significance after false discovery rate correction (q>0.1). Reverse MR analysis revealed 19 Rs-fMRI traits related to glaucoma risk, including the salience or central executive network in the frontal region (p=0.0005, q=0.08) and the motor network (p=0.0009, q=0.08) with significant causality.

Conclusions: This MR study revealed potentially causal relationships between glaucoma and brain functional networks. Especially, the functional connectivity of the motor network between the postcentral or precentral areas may potentially lead to increased risk of glaucoma.

Objective: To compare the genetic characteristics of the normal control group to those of neovascular age-related macular degeneration (AMD) patients and to detect single-nucleotide polymorphisms (SNPs) related to the pathogenesis of neovascular AMD and the sensitivity to anti-VEGF drug, combercept.

Method: This is a prospective case-controlled study. A total of 104 neovascular AMD patients were treated with combercept and 106 normal subjects were served as the control group. SNPs associated with neovascular AMD and disease susceptibility and drug sensitivity were analysed.

Results: Significant differences existed between neovascular AMD patients and normal subjects among genotypes of the SNPs of two genes, ARMS2 (rs10490924 T) and HTRA 1 (rs11200638 A). The T alleles in rs1065489 of CFH and the rs2230205 of C3 significantly promoted neovascular AMD in males while having no significant effect in females. Six SNPs of five genes, including C3 (rs2250656 G), CFB (rs2072633 G), CFH (rs2274700 A, rs3766405 T), KDR (rs6828477 A) and FZD 4 (rs10898563 T), had significant impact in reducing neovascular AMD. Two SNPs of the CFH gene (rs2274700 A and rs3766405 T) and one SNP of the CFB gene, rs2072633 G, were statistically significantly associated with good response to combercept. Conversely, the other two SNPs of the CFH gene, rs1065489 T and rs3753396 G, and the rs7412 T of the APOE gene were associated with a relatively poor patient response to drug action. Two sets of SNPs of CFB have a combined positive effect on disease. The two SNPs of CFH (rs1065489 T and rs3753396 G) and the combination of the two SNPs of CFH and rs7412T of APOE have negative effects on the drug effectiveness.

Conclusions: These genotype differences facilitate the selection of individualised treatment options towards obtaining the most efficacious clinical treatment. These findings need to be validated by studies with different ethnic populations and/or larger samples.

Background/aims: Charles Bonnet syndrome (CBS) is a common complication of visual impairment. However, demographic and clinical characteristics may modify the prevalence and impact of the condition. The aim was to investigate the prevalence of CBS among visually impaired military veterans and the associated impact of visual hallucinations.

Methods: Cross-sectional screening and survey study of members at Blind Veterans UK, a national charity supporting ex-armed forces men and women with a registered visual impairment. Data were analysed using membership records followed by a prospective CBS screening measure and survey.

Results: 461 military veterans with CBS were identified from 4109 individual records, representing 11.2% (95% CI 10.2% to 12.2%) of the population. From this, 115 members (24.9%) participated in the survey. The average age was 82.5 (±12.3) years and 89.6% were male. The most common ophthalmic condition was age-related macular degeneration (58.3%) and participants had severely reduced visual acuity (average better eye 1.2 (±1.4) LogMAR). Reporting 'bothersome' hallucinations was associated with living with CBS for over 3 years (p=0.01) and hallucinating at least once per week (p=0.05). Diverting attention elsewhere was considered an effective relief strategy among 64.9% (95% CI 56.2% to 73.6%) of those who had used this. Fewer than half (44.3%) had disclosed their symptoms to a hospital doctor and corresponding clinical management was variable.

Conclusion: CBS was common among visually impaired military veterans and was associated with negative outcomes. However, limitations include identifying cases through retrospective screening and not controlling for visual acuity and rehabilitation. Findings may not generalise beyond veterans in the charity's database. Yet, the findings yield evidence for a therapeutic benefit of relief strategies in managing symptoms and the need to promote dialogue about visual hallucinations between patients and clinicians.

Background/aims: To determine the accuracy of an innovative paper-based, low-cost, easily accessed, portable method of evaluating the visual field (VF).

Methods: Two groups of consecutive patients with glaucoma and controls were enrolled. Both eyes were assessed, and the Humphrey field analyser (HFA) (24-2) was used to identify patients with and without VF defects. The TsiogkaSpaeth grid (TS) grid test was performed in each eye of all participants. The Humphrey VF examination and the TS grid examination were conducted consecutively on the same day by two different examiners, prior to any other eye examinations. Sensitivity, specificity, and positive and negative predictive values of the TS grid scotoma area were assessed, using the results found with the HFA as the reference standard. A mixed-effects logistic regression model with a random intercept per participant was used to account for the correlation between eyes.

Results: This prospective, cross-sectional study recruited 51 participants in a tertiary care hospital-based glaucoma department. Of them, 19 had glaucomatous defects in both eyes, 12 had defects in 1 eye only and 21 had no defects in either eye. The sensitivity and specificity of the TS grid test were 81.63% (95% CI 68% to 91.2%) and 84.91% (95% CI 72.4% to 93.3%), respectively, positive predicted value was 83.33% (95% CI 69.8% to 92.5%) and negative predictive value was 83.33% (95% CI 70.7% to 92.1%). The area under the curve was 0.83 (95% CI 0.76 to 0.91). There was a significant correlation between the TS grid test score and the VF Index of the HFA 24-2 (r=0.87, p<0.0001) and the mean deviation of the HFA 24-2 (r=0.85, p<0.0001).

Conclusions: The TS grid test offers a reliable, low-cost alternative for VF examination in glaucoma patients. This method could be advantageous in clinical settings with limited access to traditional testing equipment. The TS grid has limitations, including difficulty of fixation loss testing and ensuring the correct fixation distance. The test's self-administration reliability is uncertain, and some patients were unable to perform it. Other limitations concern the grid's design and the small sample size of the study.

Objective: Retinopathy of prematurity (ROP) registers enable population-based studies to monitor ROP screening programmes to improve their effectiveness. The aim of this study was to determine the frequency of ROP and the coverage of screening in a South African cohort using a prospective ROP South African (ROPSA) register.

Methods and analysis: Infants born from 1 May 2022 to 31 January 2023 and screened prospectively for ROP at five neonatal intensive care units in Cape Town were included. The screening criteria were a gestational age (GA) <32 weeks or birth weight (BW) <1250 g. Data were extracted from the ROPSA register and analysed.

Results: 696 of 1154 (60.3%) eligible infants were screened, almost half of whom (45.7%) did not complete screening. ROP was detected in 220 infants (31.6%, 95% CI 28.3% to 35.3%), 7 (1.0%) of whom required treatment. Infants with incomplete screening had a lower mean GA than those who completed screening; 28.7 (SD 1.6, range 25-33) and 29.1 (SD 1.7, range 24-36) weeks, respectively (p=0.004) and a lower mean BW; 1048 (SD 203, range 650-1690) g and 1108.5 (SD 227, range 640-1840) g, respectively (p<0.001).

Conclusions: Data from the ROPSA register on the frequency of any ROP and treatment-requiring ROP may be biased due to low screening coverage and high incomplete screening. Reasons need to be explored and corrective interventions initiated. The ROPSA register will enable the impact of these interventions to be monitored. The findings of this study will contribute to the ongoing revision of South African national ROP screening guidelines.

Background: The advent of generative artificial intelligence has led to the emergence of multiple vision large language models (VLLMs). This study aimed to evaluate the capabilities of commonly available VLLMs, such as OpenAI's GPT-4V and Google's Gemini, in detecting and diagnosing ocular diseases from retinal images.

Methods and analysis: From the Singapore Epidemiology of Eye Diseases (SEED) study, we selected 44 representative retinal photographs, including 10 healthy and 34 representing six eye diseases (age-related macular degeneration, diabetic retinopathy, glaucoma, visually significant cataract, myopic macular degeneration and retinal vein occlusion). OpenAI's GPT-4V (both default and data analyst modes) and Google Gemini were prompted with each image to determine if the retina was normal or abnormal and to provide diagnostic descriptions if deemed abnormal. The outputs from the VLLMs were evaluated for accuracy by three attending-level ophthalmologists using a three-point scale (poor, borderline, good).

Results: GPT-4V default mode demonstrated the highest detection rate, correctly identifying 33 out of 34 detected correctly (97.1%), outperforming its data analyst mode (61.8%) and Google Gemini (41.2%). Despite the relatively high detection rates, the quality of diagnostic descriptions was generally suboptimal-with only 21.2% of GPT-4V's (default) responses, 4.8% of GPT-4V's (data analyst) responses and 28.6% for Google Gemini's responses rated as good.

Conclusions: Although GPT-4V default mode showed generally high sensitivity in abnormality detection, all evaluated VLLMs were inadequate in providing accurate diagnoses for ocular diseases. These findings emphasise the need for domain-customised VLLMs and suggest the continued need for human oversight in clinical ophthalmology.

Objective: To evaluate multimodal imaging (MMI) biomarkers for predicting progression from intermediate to advanced age-related macular degeneration (AMD).

Methods and analysis: This prospective longitudinal cohort study included patients with intermediate AMD (iAMD) enrolled in the University of Colorado AMD registry between July 2014 and August 2023, with follow-up through February 2024. At enrolment, patients' medical histories and MMI were collected. Baseline and follow-up imaging were reviewed for progression to geographic atrophy (GA) and neovascular AMD (nAMD). Univariate and multivariable Cox proportional hazard modelling with competing risks to determine HRs for progression.

Results: A total of 367 patients (733 eyes) with iAMD were included in the study, with a median follow-up of 27.8 months. During this period, 100 eyes progressed to GA, 58 to nAMD. Adjusted for age, BMI and hypertension, progression to nAMD was significantly associated with soft drusen (HR 5.31, 95% CI 1.95 to 14.4, p=0.001), pigmentary changes (HR 2.74, 95% CI 1.52 to 4.92, p=0.0008) on colour fundus photography (CFP) and subretinal hyper-reflective material (SHRM) (HR 3.36, 95% CI 1.88 to 6.02, p<0.0001) and intraretinal hyper-reflective foci (IHRF) (HR 3.12, 95% CI 1.74 to 5.57, p=0.0001) on optical coherence tomography (OCT). Adjusted for age, progression to GA was predicted by soft drusen (HR 1.90, 95% CI 1.11 to 3.27, p=0.020), drusenoid pigment epithelial detachment (PED) (HR 5.51, 95% CI 2.49 to 12.2, p<0.0001), avascular non-drusenoid PED (HR 6.59, 95% CI 1.54 to 28.1, p=0.011), pigmentary changes (HR 4.44, 95% CI 2.84 to 6.96, p<0.0001) on CFP and nnSRF (HR 6.41, 95% CI 1.39 to 29.6, p=0.017), SHRM (HR 2.55, 95% CI 1.45 to 4.49, p=0.001), drusenoid PED (HR 2.25, 95% CI 1.43 to 3.55, p=0.0005), avascular non-drusenoid PED (HR 4.67, 95% CI 2.45 to 8.92, p<0.0001), IHRF (HR 6.27, 95% CI 3.89 to 10.1, p<0.0001) and incomplete retinal pigment epithelium and outer retinal atrophy (HR 9.42, 95% CI 5.82 to 15.2, p<0.0001) on OCT (table 3).

Conclusions: Key imaging biomarkers associated with the progression were identified, which may offer prognostic information for providers. However, the study is limited by its predominantly Caucasian population and single-centre design, which may affect the generalisability of certain biomarkers.