Looking at the whole picture: constrained unsupervised anomaly segmentation

Abstract

Current unsupervised anomaly localization approaches rely on generative models to learn the distribution of normal images, which is later used to identify potential anomalous regions derived from errors on the reconstructed images. However, a main limitation of nearly all prior literature is the need of employing anomalous images to set a class-specific threshold to locate the anomalies. This limits their usability in realistic scenarios, where only normal data is typically accessible. Despite this major drawback, only a handful of works have addressed this limitation, by integrating supervision on attention maps during training. In this work, we propose a novel formulation that does not require accessing images with abnormalities to define the threshold. Furthermore, and in contrast to very recent work, the proposed constraint is formulated in a more principled manner, leveraging well-known knowledge in constrained optimization. In particular, the equality constraint on the attention maps in prior work is replaced by an inequality constraint, which allows more flexibility. In addition, to address the limitations of penalty-based functions we employ an extension of the popular log-barrier methods to handle the constraint. Comprehensive experiments on the popular BRATS'19 dataset demonstrate that the proposed approach substantially outperforms relevant literature, establishing new state-of-the-art results for unsupervised lesion segmentation. IntroductionUnder the supervised learning paradigm, deep learning models have achieved astonishing performance in a wide range of applications. Nevertheless, a main limitation of these models is the large amount of labeled data required for training. Obtaining such curated labeled datasets is a cumbersome process prone to subjectivity, which makes access to sufficient training data difficult in practice. This problem is further magnified in the context of medical image segmentation, where labeling involves assigning a category to each image pixel

An efficient semi-supervised quality control system trained using physics-based MRI-artefact generators and adversarial training

“…Accurate thresholds are not easily identified. To tackle this problem [Silva-Rodríguez et al, 2021] proposed a novel formulation that does not require accessing images with artefacts to define these thresholds. In particular, they obtain this by an inequality constraint, implemented by extending a log-barrier method.…”