Data-driven machine learning, as a promising approach, possesses the capability to build high-quality, exact, and robust models from ophthalmic medical data. Ophthalmic medical data, however, presently exist across disparate data silos with privacy limitations, making centralized training challenging. While ophthalmologists may not specialize in machine learning and artificial intelligence (AI), considerable impediments arise in the associated realm of research. To address these issues, we design and develop FedEYE, a scalable and flexible end-to-end ophthalmic federated learning platform. During FedEYE design, we adhere to four fundamental design principles, ensuring that ophthalmologists can effortlessly create independent and federated AI research tasks. Benefiting from the design principles and architecture of FedEYE, it encloses numerous key features, including rich and customizable capabilities, separation of concerns, scalability, and flexible deployment. We also validated the applicability of FedEYE by employing several prevalent neural networks on ophthalmic disease image classification tasks.

Early detection, regular monitoring of eyelid tumors and post-surgery recurrence monitoring are crucial for patients. However, frequent hospital visits are burdensome for patients with poor medical conditions. This study validates a novel deep learning-based mobile application, based on YOLOv5 and Efficient-Net v2-B architectures, for self-diagnosing eyelid tumors, enabling improved health support systems for such patients. 1195 preprocessed clinical ocular photographs and biopsy results were collected for model training. The best-performing model was chosen and converted into a smartphone-based application, then further evaluated based on external validation dataset, achieved 0.921 accuracy for triple classification outcomes (benign/malignant eyelid tumors or normal eye), generally superior to that of general physicians, resident doctors, and ophthalmology specialists. Intelligent Eyelid Tumor Screening application exhibited a straightforward detection process, user-friendly interface and treatment recommendation scheme, provides preliminary evidence for recognizing eyelid tumors and could be used by healthcare professionals, patients and caregivers for detection and monitoring purposes.

Purpose: The aim of this study is to apply deep learning techniques for the development and validation of a system that categorizes various phases of dry age-related macular degeneration (AMD), including nascent geographic atrophy (nGA), through the analysis of optical coherence tomography (OCT) images.Methods: A total of 3,401 OCT macular images obtained from 338 patients admitted to Shenyang Aier Eye Hospital in 2019–2021 were collected for the development of the classification model. We adopted a convolutional neural network (CNN) model and introduced hierarchical structure along with image enhancement techniques to train a two-step CNN model to detect and classify normal and three phases of dry AMD: atrophy-associated drusen regression, nGA, and geographic atrophy (GA). Five-fold cross-validation was used to evaluate the performance of the multi-label classification model.Results: Experimental results obtained from five-fold cross-validation with different dry AMD classification models show that the proposed two-step hierarchical model with image enhancement achieves the best classification performance, with a f1-score of 91.32% and a kappa coefficients of 96.09% compared to the state-of-the-art models. The results obtained from the ablation study demonstrate that the proposed method not only improves accuracy across all categories in comparison to a traditional flat CNN model, but also substantially enhances the classification performance of nGA, with an improvement from 66.79 to 81.65%.Conclusion: This study introduces a novel two-step hierarchical deep learning approach in categorizing dry AMD progression phases, and demonstrates its efficacy. The high classification performance suggests its potential for guiding individualized treatment plans for patients with macular degeneration.

Myopia is a growing global health concern, and early detection and intervention are crucial for preventing its onset and progression. This study aims to predict myopia onset one year in advance by integrating biometric measurements with multimodal imaging data, including 3D optical coherence tomography (OCT) and color fundus photography (CFP). A dataset of 472 eyes from 347 subjects aged 6–14 years was collected, encompassing demographic information, biometric data, OCT, and CFP. Deep learning models were trained on OCT and CFP images to extract relevant features. A semi-supervised approach was employed to segment the choroid layer in OCT images, and the segmented images were then used to generate Early Treatment Diabetic Retinopathy Study (ETDRS) grid thickness value. An XGBoost model was developed to integrate image scores, ETDRS grid values, and biometric data for predicting myopia onset. The model incorporating all available data achieved an area under the receiver operating characteristic curve (AUROC) of 0.845 ± 0.050. Permutation importance analysis revealed that spherical equivalent was the most influential variable, followed by CFP scores, OCT scores, and ETDRS variables. The mixed model with multimodal information effectively captured the complex interactions and combined effects of the variables. These findings demonstrate the potential of integrating multimodal data to enhance the accuracy of myopia onset prediction, paving the way for personalized myopia management strategies.

Model fusion is becoming a crucial component in the context of model-as-a-service scenarios, enabling the delivery of high-quality model services to local users. However, this approach introduces privacy risks and imposes certain limitations on its applications. Ensuring secure model exchange and knowledge fusion among users becomes a significant challenge in this setting. To tackle this issue, we propose PrivFusion, a novel architecture that preserves privacy while facilitating model fusion under the constraints of local differential privacy. PrivFusion leverages a graph-based structure, enabling the fusion of models from multiple parties without additional training. By employing randomized mechanisms, PrivFusion ensures privacy guarantees throughout the fusion process. To enhance model privacy, our approach incorporates a hybrid local differentially private mechanism and decentralized federated graph matching, effectively protecting both activation values and weights. Additionally, we introduce a perturbation filter adapter to alleviate the impact of randomized noise, thereby recovering the utility of the fused model. Through extensive experiments conducted on diverse image datasets and real-world healthcare applications, we provide empirical evidence showcasing the effectiveness of PrivFusion in maintaining model performance while preserving privacy. Our contributions offer valuable insights and practical solutions for secure and collaborative data analysis within the domain of privacy-preserving model fusion.

This study aimed to evaluate and optimize intraocular lens (IOL) power selection for cataract patients with high axial myopia receiving trifocal IOLs. A multi-center, retrospective observational case series was conducted. Patients having an axial length ≥26 mm and undergoing cataract surgery with trifocal IOL implanted were studied. Preoperative biometric and postoperative outcome data from 139 eyes were collected to train and test various machine learning (ML) models (support vector machine, linear regression, and stacking regressor) using five-fold cross-validation. The models' performance was further validated externally using data from 48 eyes enrolled from other hospitals. Performance of seven IOL calculation formulas (BUII, Kane, EVO, K6, DGS, Holladay I, and SRK/T) were examined with and without ML models. The results of cross-validation revealed improvements across all IOL calculation formulas, especially for K6 and Holladay I. The model increased the percentage of eyes with a prediction error (PE) within ±0.50 D from 71.94% to 79.14% for K6, and from 35.25% to 51.80% for Holladay I. In external validation involving 48 patients from other centers, six out of seven formulas demonstrated a reduction in the mean absolute error (MAE). K6's PE within ±0.50 D improved from 62.50% to 77.08%, and Holladay I from 16.67% to 58.33%. In this study, we conducted a comprehensive evaluation of seven IOL power calculation formulas in high axial myopia cases and explored the effectiveness of the Stacking Regressor model in augmenting their accuracy. Of these formulas, K6 and Holladay I exhibited the most significant improvements, suggesting that integrating ML may have varying levels of effectiveness across different formulas but holds substantial promise in improving the predictability of IOL power calculations in patients with long eyes.