Pathological Evidence Exploration in Deep Retinal Image Diagnosis
This work addresses the need for interpretable evidence in medical AI for ophthalmologists, though it is incremental as it builds on existing GAN and interpretability techniques.
The paper tackled the lack of interpretability in deep learning for medical diagnosis by proposing a method to extract pathological descriptors from diabetic retinopathy detectors and synthesize retinal images using GANs, with verification by ophthalmologists showing superior qualitative and quantitative results compared to existing methods.
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.