Topological Persistent Homology for AI-Powered Retinal Image Classification

As many systemic diseases manifest in the retina, retinal imaging can be used as a non-invasive diagnostic tool beyond retinopathy. Automatic processing of medical imaging is highly desirable for efficient screening and diagnosis, particularly in AI-driven healthcare applications. Most current methods rely on intensive image processing or learning techniques. This paper proposes a method for automatic retinal image analysis using persistent homology, a Topological Data Analysis (TDA) framework that can serve as a feature extraction technique for AI models. The framework computes the number of topological components at each grayscale level and represents their lifespan as persistence diagrams, which can be further processed into persistence landscapes for statistical analysis. These extracted features provide valuable inputs for AI-based classification models. The method was validated using a <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$2 D$</tex> fundus image dataset consisting of 20 normal and 20 pathological retinal images. Results show that the zeroth homology successfully distinguishes between both groups, and Principal Component Analysis (PCA) achieves an 87.5% clustering accuracy. Persistent homology is a promising approach for efficient AI-driven medical image analysis and automatic diagnosis.