Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Ensembled Cold-Diffusion Restorations for Unsupervised Anomaly Detection

Sergio Naval Marimont, Vasilis Siomos, Matthew Baugh, Christos Tzelepis, Bernhard Kainz, Giacomo Tarroni

TL;DR

The paper tackles unsupervised anomaly detection in medical imaging by addressing the limitations of purely generative or self-supervised approaches. It introduces DISYRE v2, a hybrid pipeline that combines Disentangled Anomaly Generation (DAG) with cold-diffusion-inspired restorations, and ensembles restorations across different anomaly severities to produce robust pixel-wise anomaly scores and localization. DAG disentangles shape, texture, and intensity bias to generate diverse, plausibly corrupted regions, improving generalization to unseen anomalies. Empirically, DISYRE v2 achieves state-of-the-art performance on three Brain MRI datasets, demonstrating strong robustness to varying anomaly severity and benefits from conditional, ensemble-based scoring and sliding-window inference applicable to high-dimensional medical images.

Abstract

Unsupervised Anomaly Detection (UAD) methods aim to identify anomalies in test samples comparing them with a normative distribution learned from a dataset known to be anomaly-free. Approaches based on generative models offer interpretability by generating anomaly-free versions of test images, but are typically unable to identify subtle anomalies. Alternatively, approaches using feature modelling or self-supervised methods, such as the ones relying on synthetically generated anomalies, do not provide out-of-the-box interpretability. In this work, we present a novel method that combines the strengths of both strategies: a generative cold-diffusion pipeline (i.e., a diffusion-like pipeline which uses corruptions not based on noise) that is trained with the objective of turning synthetically-corrupted images back to their normal, original appearance. To support our pipeline we introduce a novel synthetic anomaly generation procedure, called DAG, and a novel anomaly score which ensembles restorations conditioned with different degrees of abnormality. Our method surpasses the prior state-of-the art for unsupervised anomaly detection in three different Brain MRI datasets.

Ensembled Cold-Diffusion Restorations for Unsupervised Anomaly Detection

TL;DR

The paper tackles unsupervised anomaly detection in medical imaging by addressing the limitations of purely generative or self-supervised approaches. It introduces DISYRE v2, a hybrid pipeline that combines Disentangled Anomaly Generation (DAG) with cold-diffusion-inspired restorations, and ensembles restorations across different anomaly severities to produce robust pixel-wise anomaly scores and localization. DAG disentangles shape, texture, and intensity bias to generate diverse, plausibly corrupted regions, improving generalization to unseen anomalies. Empirically, DISYRE v2 achieves state-of-the-art performance on three Brain MRI datasets, demonstrating strong robustness to varying anomaly severity and benefits from conditional, ensemble-based scoring and sliding-window inference applicable to high-dimensional medical images.

Abstract

Unsupervised Anomaly Detection (UAD) methods aim to identify anomalies in test samples comparing them with a normative distribution learned from a dataset known to be anomaly-free. Approaches based on generative models offer interpretability by generating anomaly-free versions of test images, but are typically unable to identify subtle anomalies. Alternatively, approaches using feature modelling or self-supervised methods, such as the ones relying on synthetically generated anomalies, do not provide out-of-the-box interpretability. In this work, we present a novel method that combines the strengths of both strategies: a generative cold-diffusion pipeline (i.e., a diffusion-like pipeline which uses corruptions not based on noise) that is trained with the objective of turning synthetically-corrupted images back to their normal, original appearance. To support our pipeline we introduce a novel synthetic anomaly generation procedure, called DAG, and a novel anomaly score which ensembles restorations conditioned with different degrees of abnormality. Our method surpasses the prior state-of-the art for unsupervised anomaly detection in three different Brain MRI datasets.
Paper Structure (11 sections, 4 equations, 3 figures, 2 tables)

This paper contains 11 sections, 4 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Methods
  4. Disentangled Anomaly Generation
  5. Cold-diffusion with synthetic anomaly corruptions
  6. Anomaly localization
  7. Experiments
  8. Experimental setup
  9. Results and Discussion
  10. Conclusion
  11. Disclosure of Interests.

Figures (3)

  • Figure 1: Overview of the main components of DISYRE v2, our proposed unsupervised anomaly detection pipeline.
  • Figure 2: AP for 4 independent seeds, using single-step restoration, FPI vs DAG.
  • Figure 3: Examples of restorations and Ensembled AS vs ground truth (red outline).