Leveraging Eye Fundus Images and Metadata Fusion for Glaucoma Detection with Artificial Intelligence

Background/Aims: Accurate diagnosis of glaucoma relies on precise imaging techniques and comprehensive clinical data. While deep learning methods hold potential for enhancing diagnostic accuracy, the incorporation of clinical data into these models remains relatively unexplored. Methods: In this study, we utilized the PAPILA dataset to investigate the integration of clinical data into machine learning models for glaucoma diagnosis. Two neural network architectures were compared: one trained solely on retinal fundus images and another incorporating both images and clinical data. The performance of these models was evaluated using standard metrics, including the DeLong test for statistical significance. Results: Our findings reveal that the inclusion of clinical data resulted in a modest improvement in classification performance. However, the difference in performance between models using only images and those incorporating clinical data was not statistically significant according to the DeLong test. Conclusion: Integrating clinical data into machine learning models for glaucoma diagnosis holds promise for enhancing diagnostic accuracy. While our study demonstrates a positive trend in classification performance with the inclusion of clinical data, further research is warranted to fully understand its impact and explore additional avenues for improvement.