Context-encoding Variational Autoencoder for Unsupervised Anomaly Detection
This addresses the need for better anomaly detection in medical imaging without annotations, offering a novel scoring approach that improves over existing methods, though it is incremental as it builds on autoencoder frameworks.
The paper tackled the problem of unsupervised anomaly detection in medical images by proposing the Context-encoding Variational Autoencoder (ceVAE), which combines reconstruction- and density-based scoring to improve sample- and pixel-wise results, achieving ROC-AUCs of 0.95 and 0.89 on BraTS-2017 and ISLES-2015 benchmarks, outperforming state-of-the-art methods.
Unsupervised learning can leverage large-scale data sources without the need for annotations. In this context, deep learning-based auto encoders have shown great potential in detecting anomalies in medical images. However, state-of-the-art anomaly scores are still based on the reconstruction error, which lacks in two essential parts: it ignores the model-internal representation employed for reconstruction, and it lacks formal assertions and comparability between samples. We address these shortcomings by proposing the Context-encoding Variational Autoencoder (ceVAE) which combines reconstruction- with density-based anomaly scoring. This improves the sample- as well as pixel-wise results. In our experiments on the BraTS-2017 and ISLES-2015 segmentation benchmarks, the ceVAE achieves unsupervised ROC-AUCs of 0.95 and 0.89, respectively, thus outperforming state-of-the-art methods by a considerable margin.