Contrastive Representations for Unsupervised Anomaly Detection and Localization

Abstract

Unsupervised anomaly detection in medical imaging aims to detect and localize arbitrary anomalies without requiring labels during training. Often, this is achieved by learning a data distribution of normal samples and detecting anomalies as regions in the image which deviate from this distribution. In the medical imaging domain, most current state-of-the-art methods use latent variable generative models operating directly on the images. However, generative models have been shown to mostly capture low-level features s.a. pixel-intensities instead of rich semantic features, which also applies to their representations. We circumvent this problem by proposing CRADL whose core idea is to model the distribution of normal samples directly in the low-dimensional representation space of an encoder which has been trained with a contrastive pretext-task. By utilizing the representations of contrastive learning we aim to fix the over-fixation on low-level features and aim to learn more semantic-rich representations. Our experiments on the task of anomaly localization on three distinct datasets show that 1) the contrastive representations are superior to generative latent variable models and 2) the CRADL framework shows competetive or superior performance to state-of-the-art.