Automated artificial intelligence-based system for clinical follow-up of patients with age-related macular degeneration

Inter-grader reliability in the Danish screening programme for diabetic retinopathy

PURPOSE

The Danish Registry of Diabetic Retinopathy includes information from >200 000 patients who attends diabetic retinopathy (DR) screening in Denmark. Screening of patients with uncomplicated type 2 diabetes is often performed by practicing ophthalmologists, while patients with type 1 and complicated type 2 diabetes attends screening at hospitals. We performed a clinical reliability study of retinal images from Danish screening facilities to explore the inter-grader agreement between the primary screening ophthalmologist and a blinded, certified grader.

METHODS

Invitations to participate were sent to screening facilities across Denmark. The primary grader uploaded fundus photographs with information on estimated level of DR (International Clinical Diabetic Retinopathy scale as 0 [no DR], 1-3 [mild, moderate or severe nonproliferative DR {NPDR}], or 4 [proliferative DR {PDR}]), region of screening, image style, and screening facility. Images were then regraded by a blinded, certified, secondary grader. Weighted kappa analysis was performed to evaluate agreement.

RESULTS

Fundus photographs from 230 patients (458 eyes) were received from practicing ophthalmologists (52.6%) and hospital-based grading centres (47.4%) from all Danish regions. Reported levels of DR by the primary graders were 66.8%, 12.2%, 13.1%, 1.3% and 5.5% for DR levels 0-4. The overall agreement between primary and secondary graders was 93% (κ = 0.83). Based on screening facility agreement was 96% (κ = 0.89) and 90% (κ = 0.76) for practicing ophthalmologists and hospital-based graders.

CONCLUSION

In this nationwide study, we observed a high overall inter-grader agreement and based on this, it is reasonable to assume that reported DR gradings in the screening programme in Denmark, accurately reflect the truth.

Artificial intelligence in ophthalmology: The path to the real-world clinic

Artificial intelligence (AI) has great potential to transform healthcare by enhancing the workflow and productivity of clinicians, enabling existing staff to serve more patients, improving patient outcomes, and reducing health disparities. In the field of ophthalmology, AI systems have shown performance comparable with or even better than experienced ophthalmologists in tasks such as diabetic retinopathy detection and grading. However, despite these quite good results, very few AI systems have been deployed in real-world clinical settings, challenging the true value of these systems. This review provides an overview of the current main AI applications in ophthalmology, describes the challenges that need to be overcome prior to clinical implementation of the AI systems, and discusses the strategies that may pave the way to the clinical translation of these systems.

Prediction of anti-vascular endothelial growth factor agent-specific treatment outcomes in neovascular age-related macular degeneration using a generative adversarial network

To develop an artificial intelligence (AI) model that predicts anti-vascular endothelial growth factor (VEGF) agent-specific anatomical treatment outcomes in neovascular age-related macular degeneration (AMD), thereby assisting clinicians in selecting the most suitable anti-VEGF agent for each patient. This retrospective study included patients diagnosed with neovascular AMD who received three loading injections of either ranibizumab or aflibercept. Training was performed using optical coherence tomography (OCT) images with an attention generative adversarial network (GAN) model. To test the performance of the AI model, the sensitivity and specificity to predict the presence of retinal fluid after treatment were calculated for the AI model, an experienced (Examiner 1), and a less experienced (Examiner 2) human examiners. A total of 1684 OCT images from 842 patients (419 treated with ranibizumab and 423 treated with aflibercept) were used as the training set. Testing was performed using images from 98 patients. In patients treated with ranibizumab, the sensitivity and specificity, respectively, were 0.615 and 0.667 for the AI model, 0.385 and 0.861 for Examiner 1, and 0.231 and 0.806 for Examiner 2. In patients treated with aflibercept, the sensitivity and specificity, respectively, were 0.857 and 0.881 for the AI model, 0.429 and 0.976 for Examiner 1, and 0.429 and 0.857 for Examiner 2. In 18.5% of cases, the fluid status of synthetic posttreatment images differed between ranibizumab and aflibercept. The AI model using GAN might predict anti-VEGF agent-specific short-term treatment outcomes with relatively higher sensitivity than human examiners. Additionally, there was a difference in the efficacy in fluid resolution between the anti-VEGF agents. These results suggest the potential of AI in personalized medicine for patients with neovascular AMD.

Artificial intelligence-based pathologic myopia identification system in the ophthalmology residency training program

Background: Artificial intelligence (AI) has been successfully applied to the screening tasks of fundus diseases. However, few studies focused on the potential of AI to aid medical teaching in the residency training program. This study aimed to evaluate the effectiveness of the AI-based pathologic myopia (PM) identification system in the ophthalmology residency training program and assess the residents’ feedback on this system.

Materials and Methods: Ninety residents in the ophthalmology department at the Second Affiliated Hospital of Zhejiang University were randomly assigned to three groups. In group A, residents learned PM through an AI-based PM identification system. In group B and group C, residents learned PM through a traditional lecture given by two senior specialists independently. The improvement in resident performance was evaluated by comparing the pre-and post-lecture scores of a specifically designed test using a paired t-test. The difference among the three groups was evaluated by one-way ANOVA. Residents’ evaluations of the AI-based PM identification system were measured by a 17-item questionnaire.

Results: The post-lecture scores were significantly higher than the pre-lecture scores in group A (p < 0.0001). However, there was no difference between pre-and post-lecture scores in group B (p = 0.628) and group C (p = 0.158). Overall, all participants were satisfied and agreed that the AI-based PM identification system was effective and helpful to acquire PM identification, myopic maculopathy (MM) classification, and “Plus” lesion localization.

Conclusion: It is still difficult for ophthalmic residents to promptly grasp the knowledge of identification of PM through a single traditional lecture, while the AI-based PM identification system effectively improved residents’ performance in PM identification and received satisfactory feedback from residents. The application of the AI-based PM identification system showed advantages in promoting the efficiency of the ophthalmology residency training program.