Unsupervised Anomaly Localization Using Variational Auto-Encoders

Abstract

An assumption-free automatic check of medical images for potentially overseen anomalies would be a valuable assistance for a radiologist. Deep learning and especially Variational Auto-Encoders (VAEs) have shown great potential in the unsupervised learning of data distributions. In principle, this allows for such a check and even the localization of parts in the image that are most suspicious. Currently, however, the reconstruction-based localization by design requires adjusting the model architecture to the specific problem looked at during evaluation. This contradicts the principle of building assumption-free models. We propose complementing the localization part with a term derived from the Kullback-Leibler (KL)-divergence. For validation, we perform a series of experiments on FashionMNIST as well as on a medical task including >1000 healthy and >250 brain tumor patients. Results show that the proposed formalism outperforms the state of the art VAE-based localization of anomalies across many hyperparameter settings and also shows a competitive max performance.

Label-Free Segmentation of COVID-19 Lesions in Lung CT

“…Anomaly detection or outlier detection is a lasting yet active research area in machine learning [35]- [37], which is a key technique to overcome the data bottleneck [38]. A natural choice for handling this problem is one-class classification methods, such as OC-SVM [39], SVDD [40], Deep SVDD [41] and 1-NN.…”

“…Several extensions such as context encoder [47], constrained VAE [48], adversarial autoencoder [48], GMVAE [49], Bayesian VAE [50] and anoVAEGAN [51] improved the accuracy of the projection. Based on the pretrained projection, You et al [49] restored the lesion area by involving an optimization on the latent manifold, while Zimmerer et al [38] located the anomaly with a term derived from the Kullback-Leibler (KL)-divergence.…”

Unsupervised brain tumor segmentation using a symmetric-driven adversarial network

“…So, they enhanced the representative ability of an auto-encoder based model by imposing a consistency in the latent space to constrain the encoder to find a latent space where the projections of the input image and the reconstructed image are close to each other. Zimmerer et al [43] used a variational auto-encoder with the Kullback-Leibler divergence to measure reconstruction errors. The AEs are able to simulate non-linear transformations from the latent space to input data, and then to detect anomalies as a deviation from the transforms by measuring the reconstruction error.…”

Anomaly Detection Through Latent Space Restoration Using Vector Quantized Variational Autoencoders

“…The default approach consists in using the reconstruction loss to identify samples with anomalies, based on the assumption that the VAE will reconstruct their anomaly-free versions. However, recent results suggest that the KL divergence is actually a better anomaly score [3]. This can be caused by the high representational power of VAEs, which can reconstruct even (previously unseen) anomalies.…”

“…Recently, methods based on Variational Auto-Encoders (VAEs) have been proposed to identify and localize anomalies in medical images [2,3,4]. VAEs are generative models trained by minimizing a loss function composed of a reconstruction term (measuring the distance between original images and reconstructions) and a Kullback-Leibler (KL) divergence term (measuring the distance between the latent distribution and a prior, generally assumed to be Gaussian).…”