Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Unsupervised Detection of Fetal Brain Anomalies using Denoising Diffusion Models

Markus Ditlev Sjøgren Olsen, Jakob Ambsdorf, Manxi Lin, Caroline Taksøe-Vester, Morten Bo Søndergaard Svendsen, Anders Nymark Christensen, Mads Nielsen, Martin Grønnebæk Tolsgaard, Aasa Feragen, Paraskevas Pegios

TL;DR

The study tackles fetal brain anomaly detection in ultrasound, where labeled abnormal data are scarce and anomalies are diverse. It introduces iNAAD, an unsupervised framework built on denoising diffusion probabilistic models (DDPMs) that uses inpainting and multi-noise reconstruction to identify deviations from normal brain anatomy, trained solely on normal images. The method evaluates three noise distributions (Gaussian, Simplex, Pyramid) and demonstrates competitive or superior performance to a supervised ResNet-18 baseline for localized anomalies, while providing explainable heatmaps of reconstruction differences. These results highlight diffusion-based unsupervised anomaly detection as a viable approach for clinical fetal ultrasound with potential for broader application after further clinical validation.

Abstract

Congenital malformations of the brain are among the most common fetal abnormalities that impact fetal development. Previous anomaly detection methods on ultrasound images are based on supervised learning, rely on manual annotations, and risk missing underrepresented categories. In this work, we frame fetal brain anomaly detection as an unsupervised task using diffusion models. To this end, we employ an inpainting-based Noise Agnostic Anomaly Detection approach that identifies the abnormality using diffusion-reconstructed fetal brain images from multiple noise levels. Our approach only requires normal fetal brain ultrasound images for training, addressing the limited availability of abnormal data. Our experiments on a real-world clinical dataset show the potential of using unsupervised methods for fetal brain anomaly detection. Additionally, we comprehensively evaluate how different noise types affect diffusion models in the fetal anomaly detection domain.

Unsupervised Detection of Fetal Brain Anomalies using Denoising Diffusion Models

TL;DR

The study tackles fetal brain anomaly detection in ultrasound, where labeled abnormal data are scarce and anomalies are diverse. It introduces iNAAD, an unsupervised framework built on denoising diffusion probabilistic models (DDPMs) that uses inpainting and multi-noise reconstruction to identify deviations from normal brain anatomy, trained solely on normal images. The method evaluates three noise distributions (Gaussian, Simplex, Pyramid) and demonstrates competitive or superior performance to a supervised ResNet-18 baseline for localized anomalies, while providing explainable heatmaps of reconstruction differences. These results highlight diffusion-based unsupervised anomaly detection as a viable approach for clinical fetal ultrasound with potential for broader application after further clinical validation.

Abstract

Congenital malformations of the brain are among the most common fetal abnormalities that impact fetal development. Previous anomaly detection methods on ultrasound images are based on supervised learning, rely on manual annotations, and risk missing underrepresented categories. In this work, we frame fetal brain anomaly detection as an unsupervised task using diffusion models. To this end, we employ an inpainting-based Noise Agnostic Anomaly Detection approach that identifies the abnormality using diffusion-reconstructed fetal brain images from multiple noise levels. Our approach only requires normal fetal brain ultrasound images for training, addressing the limited availability of abnormal data. Our experiments on a real-world clinical dataset show the potential of using unsupervised methods for fetal brain anomaly detection. Additionally, we comprehensively evaluate how different noise types affect diffusion models in the fetal anomaly detection domain.
Paper Structure (17 sections, 4 equations, 4 figures, 4 tables, 1 algorithm)

This paper contains 17 sections, 4 equations, 4 figures, 4 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Method
  4. Learning Distribution of Normal Brain Images with DDPMs
  5. iNAAD: Inpainting-based Noise Agnostic Anomaly Detection
  6. Experiments and Results
  7. Dataset.
  8. Models and implementation.
  9. Evaluation of DDPMs.
  10. Supervised baseline.
  11. Results.
  12. Ablation study.
  13. Localization and explanability.
  14. Discussion and Conclusion
  15. Limitations.
  16. ...and 2 more sections

Figures (4)

  • Figure 1: Overview of iNAAD for unsupervised detection of fetal brain anomalies.
  • Figure 2: Reconstruction of a normal fetal brain from corruption level $t=150$.
  • Figure 3: ROC curves on the test set for the different models.
  • Figure 4: Heatmaps and annotated anomalies by an MD with 3 years of experience in prenatal ultrasound imaging. Top: Abnormal cases. Bottom: Normal cases. Anomalies were annotated and localized only for visualization purposes.