Deep Autoencoding Models for Unsupervised Anomaly Segmentation in Brain MR Images

TL;DR

This paper introduces deep spatial autoencoding models for unsupervised anomaly segmentation in brain MR images, effectively capturing normal anatomical variability and detecting anomalies like MS lesions through reconstruction comparison.

Contribution

It demonstrates that deep spatial autoencoders can efficiently model entire brain MR images for anomaly detection, outperforming patch-based methods.

Findings

Deep spatial autoencoders effectively model whole brain MR images.

Constraints on latent space improve segmentation accuracy.

Adversarial training enhances anomaly detection performance.

Abstract

Reliably modeling normality and differentiating abnormal appearances from normal cases is a very appealing approach for detecting pathologies in medical images. A plethora of such unsupervised anomaly detection approaches has been made in the medical domain, based on statistical methods, content-based retrieval, clustering and recently also deep learning. Previous approaches towards deep unsupervised anomaly detection model patches of normal anatomy with variants of Autoencoders or GANs, and detect anomalies either as outliers in the learned feature space or from large reconstruction errors. In contrast to these patch-based approaches, we show that deep spatial autoencoding models can be efficiently used to capture normal anatomical variability of entire 2D brain MR images. A variety of experiments on real MR data containing MS lesions corroborates our hypothesis that we can detect and even delineate anomalies in brain MR images by simply comparing input images to their reconstruction. Results show that constraints on the latent space and adversarial training can further improve the segmentation performance over standard deep representation learning.