# Pathological Evidence Exploration in Deep Retinal Image Diagnosis

## TL;DR

This paper introduces a novel interpretability approach for deep retinal image diagnosis by extracting pathological descriptors, visualizing symptoms with GANs, and manipulating lesion features, validated by ophthalmologists.

## Contribution

It proposes a new pathological descriptor based on neural activations, combined with GAN-based visualization and manipulation, enhancing interpretability and synthesis of retinal images for diabetic retinopathy.

## Key findings

- Synthesized images accurately reflect diabetic retinopathy symptoms.
- Generated images are qualitatively and quantitatively superior to existing methods.
- Ophthalmologists verified the clinical relevance of the synthesized images.

## Abstract

Though deep learning has shown successful performance in classifying the label and severity stage of certain disease, most of them give few evidence on how to make prediction. Here, we propose to exploit the interpretability of deep learning application in medical diagnosis. Inspired by Koch's Postulates, a well-known strategy in medical research to identify the property of pathogen, we define a pathological descriptor that can be extracted from the activated neurons of a diabetic retinopathy detector. To visualize the symptom and feature encoded in this descriptor, we propose a GAN based method to synthesize pathological retinal image given the descriptor and a binary vessel segmentation. Besides, with this descriptor, we can arbitrarily manipulate the position and quantity of lesions. As verified by a panel of 5 licensed ophthalmologists, our synthesized images carry the symptoms that are directly related to diabetic retinopathy diagnosis. The panel survey also shows that our generated images is both qualitatively and quantitatively superior to existing methods.