CNN Layer-wise explanations for the prediction of Diabetic Retinopathy from OCTA images

Optical Coherence Tomography Angiography (OCTA) is a promising new imaging technique in ophthalmology for diagnosing Diabetic Retinopathy (DR) in diabetic patients. Early detection and treatment of DR is crucial to prevent vision loss and blindness. Deep Learning-based approaches to image classification have proven to be very effective in image analysis tasks, but their adoption hinges on the ability of the medical professionals to understand how these networks make their decisions. This study focuses on exploring explainable artificial intelligence techniques to better understand how a Convolutional Neural Network (CNN) classifies OCTA images of diabetic patients with varying levels of DR. This exploration is based on Signature Activation, a technique that generates holistic and class-agnostic explanations of the CNN’s decisions.