Synthesizing Filamentary Structured Images with GANs


This paper aims at synthesizing filamentary structured images such as retinal fundus images and neuronal images, as follows: Given an unseen ground-truth, to generate one or multiple realistic looking phantoms. A ground-truth could be a binary image containing the filamentary structured morphology, while the synthesized output image is of the same size as of the ground-truth and has similar visual appearance to what has been presented in the training set. Our approach is inspired by the recent progresses in generative adversarial nets (GANs) as well as image style transfer. In particular, it is dedicated to our problem context with the following properties: Rather than large-scale dataset, it works well in the presence of only 10 or 20 training examples, which is common in medical image analysis; It is capable of synthesizing diverse images from the same ground-truth. Last and importantly, the synthetic images produced by our approach are demonstrated to be useful in boosting image analysis performance.



He ZHAO, Huiqi LI, Li CHENG. Synthesizing Filamentary Structured Images with GANs, In submission, 2017, with an arXiv entry