ToFi-ML: Retinal Image Screening with Topological Machine Learning

The analysis of fundus images for the early screening of eye diseases is of great clinical importance. Traditional methods for such analysis are time-consuming and expensive as it requires a trained clinician. Therefore, the need for a comprehensive and automated method of retinal image screening to diagnose and grade retinal diseases has long been recognized. In the past decade, with the substantial developments in computer vision and deep learning, machine learning methods became highly effective in this field as clinical-decision support methods. However, most of these algorithms face challenges like computational feasibility, reliability, and interpretability. In this paper, we develop a novel approach to this crucial task in retinal image screening. By employing topological data analysis tools, for the most common retinal diseases, i.e., Diabetic Retinopathy (DR), Glaucoma, and Age-related Macular Degeneration (AMD), we observe different types of topological patterns between normal and abnormal classes. These patterns enable us to produce topological fingerprints of fundus images, and we use them as feature vectors with standard ML methods. Our computationally efficient model ToFi-ML outperforms or gives highly competitive accuracy results with state-of-the-art deep learning methods. Furthermore, our topological fingerprints are both explainable and interpretable, and can easily be integrated with future early-screening ML models.