Predicting Intraocular Pressure using Neural Networks: Incorporating Eye Fundus Images and Clinical Data from PAPILA Dataset

Importance: Glaucoma, a leading cause of irreversible blindness, necessitates accurate intraocular pressure (IOP) prediction for early detection and management. The integration of deep learning algorithms with clinical data presents a novel approach to enhance diagnostic accuracy and patient care in ophthalmology, addressing a critical gap in current diagnostic methodologies. Objective: To assess the effectiveness of deep learning models, specifically Segformer and EfficientNetV2B0, in predicting IOP when combined with clinical data and eye fundus images, aiming to improve diagnostic accuracy and management of glaucoma. Design, Setting, and Participants: This cross-sectional study utilized the PAPILA database. The study employed publicly available databases G1020, ORIGA, and PAPILA, incorporating retinal fundus images from patients diagnosed with glaucoma. It focused on leveraging clinical data such as central corneal thickness, age, gender, axial length, and refractive defect for predictive analysis. Exposure(s): Participants were exposed to deep learning algorithm-based analysis, integrating clinical data with retinal fundus images to predict IOP. Main Outcome(s) and Measure(s): The primary outcome was the accuracy of the IOP predictions, evaluated using Mean Absolute Error (MAE), Coefficient of Determination (R-squared), and Root Mean Squared Error (RMSE). The model's performance was assessed based on its ability to accurately predict actual measured IOP values. Results: The study analyzed images and clinical data from patients within the PAPILA database. The deep learning model achieved an MAE of 2.52, indicating moderate accuracy in predicting IOP. The R-squared value was reported at 0.10, reflecting the model's limited capacity to explain variance in IOP values among the study population. Conclusions and Relevance: The findings suggest that deep learning algorithms, when integrated with clinical data, have the potential to predict IOP with a moderate level of accuracy. This innovative approach could significantly impact the management and diagnosis of glaucoma. This study underscores the potential of AI in revolutionizing ophthalmic diagnostics, particularly for glaucoma, although further validation and improvement of these models are necessary before clinical application.