Unsupervised Anomaly Detection with Generative Adversarial Networks to Guide Marker Discovery

Link to paper: https://arxiv.org/abs/1703.05921

Problem: Models are typically based on large amounts of data with annotated examples of known markers aiming at automating detection. However, High annotation effort and the limitation to a vocabulary of known markers limit the power of such approaches.

Proposed Solution: unsupervised learning to identify anomalies in imaging data as candidates for markers using GANs. Generative adversarial training is performed on healthy data and testing is performed on both, unseen healthy cases and anomalous data.

GAN consists of two adversarial modules, a generator G and a discriminator D.

--> The generator G learns a distribution pg over data x via a mapping G(z) of samples z, 1D vectors of uniformly distributed input noise sampled from latent space Z, to 2D images in the image space manifold X , which is populated by healthy examples.

--> Discriminator D is a standard CNN that maps a 2D image to a single scalar value D(·). The discriminator output D(·) can be interpreted as the probability that the given input to the discriminator D was a real image x sampled from training data X or generated G(z) by the generator G.

Training Goal: The discriminator is trained to maximize the probability of assigning real training examples the “real” and samples from pg the “fake” label. The generator G is simultaneously trained to fool D via minimizing V (G) = log(1 − D(G(z))), or V (G) = D(G(z)).

where V is a two-player min-max game value function.

The loss function used: Residual loss and Discrimniation Loss.

--> The residual loss measures the visual dissimilarity between query image x and generated image G(zγ).