Visual explanations for the detection of diabetic retinopathy from retinal fundus images

Abstract

In medical image classification tasks like the detection of diabetic retinopathy from retinal fundus images, it is highly desirable to get visual explanations for the decisions of black-box deep neural networks (DNNs). However, gradient-based saliency methods often fail to highlight the diseased image regions reliably. On the other hand, adversarially robust models have more interpretable gradients than plain models but suffer typically from a significant drop in accuracy, which is unacceptable for clinical practice. Here, we show that one can get the best of both worlds by ensembling a plain and an adversarially robust model: maintaining high accuracy but having improved visual explanations. Also, our ensemble produces meaningful visual counterfactuals which are complementary to existing saliency-based techniques. Code is available under \url{https://github.com/valentyn1boreiko/Fundus_VCEs}.

Sparse Activations for Interpretable Disease Grading

“…Numerous high-performing black-box DR detection methods have been proposed (Rao et al, 2020;Alyoubi et al, 2020;Tavakoli and Kelley, 2021;Huang et al, 2021). For such methods, interpretation is mostly aided by saliency maps (Wang and Yang, 2019;Chetoui and Akhloufi, 2020) or the generation of counterfactual images (González-Gonzalo et al, 2020;Boreiko et al, 2022). A more interpretable approach for detecting DR is a multipleinstance learning model which combines features extracted from different image patches with attention weights (Papadopoulos et al, 2021).…”

Dynamic Graph Clustering Learning for Unsupervised Diabetic Retinopathy Classification

“…Therefore, the development of an automated diagnostic system for diabetic retinopathy is essential. Significant progress has been made in this field, as evidenced by notable studies [3][4][5]. Traditional image identification training methods heavily rely on high-quality, manually annotated labels.…”