Papers
Topics
Authors
Recent
AI Research Assistant
AI Research Assistant
Well-researched responses based on relevant abstracts and paper content.
Custom Instructions Pro
Preferences or requirements that you'd like Emergent Mind to consider when generating responses.
Gemini 2.5 Flash
Gemini 2.5 Flash 80 tok/s
Gemini 2.5 Pro 49 tok/s Pro
GPT-5 Medium 26 tok/s Pro
GPT-5 High 32 tok/s Pro
GPT-4o 92 tok/s Pro
Kimi K2 182 tok/s Pro
GPT OSS 120B 438 tok/s Pro
Claude Sonnet 4 38 tok/s Pro
2000 character limit reached

Towards Universal Unsupervised Anomaly Detection in Medical Imaging (2401.10637v1)

Published 19 Jan 2024 in eess.IV, cs.CV, and cs.LG

Abstract: The increasing complexity of medical imaging data underscores the need for advanced anomaly detection methods to automatically identify diverse pathologies. Current methods face challenges in capturing the broad spectrum of anomalies, often limiting their use to specific lesion types in brain scans. To address this challenge, we introduce a novel unsupervised approach, termed \textit{Reversed Auto-Encoders (RA)}, designed to create realistic pseudo-healthy reconstructions that enable the detection of a wider range of pathologies. We evaluate the proposed method across various imaging modalities, including magnetic resonance imaging (MRI) of the brain, pediatric wrist X-ray, and chest X-ray, and demonstrate superior performance in detecting anomalies compared to existing state-of-the-art methods. Our unsupervised anomaly detection approach may enhance diagnostic accuracy in medical imaging by identifying a broader range of unknown pathologies. Our code is publicly available at: \url{https://github.com/ci-ber/RA}.

Definition Search Book Streamline Icon: https://streamlinehq.com
References (41)
  1. Mask, stitch, and re-sample: Enhancing robustness and generalizability in anomaly detection through automatic diffusion models. International Conference on Machine learning Workshops - 3rd Workshop on Interpretable Machine Learning in Healthcare (IMLH) URL: https://openreview.net/forum?id=kTpafpXrqa.
  2. What do AEs learn? Challenging Common Assumptions in Unsupervised Anomaly Detection, in: International Conference on Medical Image Computing and Computer-Assisted Intervention, Springer. pp. 304–314.
  3. Generalizing unsupervised anomaly detection: Towards unbiased pathology screening. International Conference on Medical Imaging with Deep Learning .
  4. Uninformed students: Student-teacher anomaly detection with discriminative latent embeddings, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 4183–4192.
  5. Padchest: A large chest X-ray image dataset with multi-label annotated reports. Medical Image Analysis 66.
  6. Soft-IntroVAE: Analyzing and improving the introspective variational autoencoder, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 4391–4400.
  7. Generative adversarial nets. Advances in Neural Information Processing Systems 27.
  8. Denoising diffusion models for out-of-distribution detection. arXiv preprint arXiv:2211.07740 .
  9. Masked autoencoders are scalable vision learners. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 16000–16009.
  10. Automated screening of computed tomography using weakly supervised anomaly detection. International Journal of Computer Assisted Radiology and Surgery .
  11. Denoising diffusion probabilistic models. Advances in Neural Information Processing Systems 33, 6840–6851.
  12. Masked swin transformer unet for industrial anomaly detection. IEEE Transactions on Industrial Informatics 19, 2200–2209. doi:10.1109/TII.2022.3199228.
  13. Denoising autoencoders for unsupervised anomaly detection in brain MRI, in: International Conference on Medical Imaging with Deep Learning, pp. 653–664.
  14. Fool me twice: delayed diagnoses in radiology with emphasis on perpetuated errors. American Journal of Roentgenology 202, 465–470.
  15. Auto-encoding variational Bayes. arXiv preprint arXiv:1312.6114 .
  16. Normalizing flows: An introduction and review of current methods. IEEE Transactions on Pattern Analysis and Machine Intelligence 43, 3964–3979.
  17. Standardized assessment of automatic segmentation of white matter hyperintensities and results of the wmh segmentation challenge. IEEE Transactions on Medical Imaging 38, 2556–2568.
  18. 3d masked autoencoders with application to anomaly detection in non-contrast enhanced breast mri. arXiv preprint arXiv:2303.05861 .
  19. Cutpaste: Self-supervised learning for anomaly detection and localization, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 9664–9674.
  20. A large, curated, open-source stroke neuroimaging dataset to improve lesion segmentation algorithms. Scientific Data 9.
  21. The multimodal brain tumor image segmentation benchmark (brats). IEEE Transactions on Medical Imaging 34, 1993–2024.
  22. A pediatric wrist trauma x-ray dataset (grazpedwri-dx) for machine learning. Scientific Data 9, 222. doi:10.1038/s41597-022-01328-z.
  23. Unsupervised brain imaging 3d anomaly detection and segmentation with transformers. Medical Image Analysis 79, 102475.
  24. Multiresolution knowledge distillation for anomaly detection, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 14902–14912.
  25. f-AnoGAN: Fast unsupervised anomaly detection with generative adversarial networks. Medical Image Analysis 54, 30–44.
  26. Natural synthetic anomalies for self-supervised anomaly detection and localization, in: Avidan, S., Brostow, G., Cissé, M., Farinella, G.M., Hassner, T. (Eds.), European Conference on Computer Vision, Springer Nature Switzerland, Cham. pp. 474–489.
  27. Maeday: Mae for few and zero shot anomaly-detection. arXiv preprint arXiv:2211.14307 .
  28. Augmenting the national institutes of health chest radiograph dataset with expert annotations of possible pneumonia. Radiology: Artificial Intelligence 1, e180041.
  29. Detecting outliers with foreign patch interpolation. Machine Learning for Biomedical Imaging 1, 1–27.
  30. Looking for abnormalities in mammograms with self- and weakly supervised reconstruction. IEEE Transactions on Medical Imaging 40, 2711–2722. doi:10.1109/TMI.2021.3050040.
  31. Weakly supervised anomaly segmentation in retinal OCT images using an adversarial learning approach. Biomedical Optical Express 12, 4713–4729.
  32. A weakly supervised gradient attribution constraint for interpretable classification and anomaly detection. IEEE Transactions on Medical Imaging , 1–1doi:10.1109/TMI.2023.3282789.
  33. Diffusion models for medical anomaly detection, in: International Conference on Medical Image Computing and Computer-Assisted Intervention, Springer. pp. 35–45.
  34. Anoddpm: Anomaly detection with denoising diffusion probabilistic models using simplex noise, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, pp. 650–656.
  35. Anatomy-guided weakly-supervised abnormality localization in chest x-rays, in: Wang, L., Dou, Q., Fletcher, P.T., Speidel, S., Li, S. (Eds.), International Conference on Medical Image Computing and Computer-Assisted Intervention, Springer Nature Switzerland, Cham. pp. 658–668.
  36. The unreasonable effectiveness of deep features as a perceptual metric, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 586–595.
  37. Anomaly detection for medical images using self-supervised and translation-consistent features. IEEE Transactions on Medical Imaging 40, 3641–3651. doi:10.1109/TMI.2021.3093883.
  38. fastmri+: Clinical pathology annotations for knee and brain fully sampled multi-coil MRI data. arXiv preprint arXiv:2109.03812 .
  39. AE-FLOW: Autoencoders with normalizing flows for medical images anomaly detection. International Conference on Learning Representations URL: https://openreview.net/forum?id=9OmCr1q54Z.
  40. Unsupervised anomaly localization using variational auto-encoders, in: International Conference on Medical Image Computing and Computer-Assisted Intervention, Springer. pp. 289–297.
  41. Context-encoding variational autoencoder for unsupervised anomaly detection. arXiv preprint arXiv:1812.05941 .
Citations (1)
View on Semantic Scholar

Summary

  • The paper introduces a reversed auto-encoder technique that creates pseudo-healthy reconstructions to detect a wide range of medical anomalies without supervision.
  • It integrates ELBO regularization, adversarial training, and a reversed loss mechanism to enhance anomaly localization across multiple imaging modalities.
  • The approach shows superior performance in brain MRIs, pediatric wrist, and chest X-rays, highlighting its potential for clinical diagnosis improvement.

Advancing Unsupervised Anomaly Detection in Medical Imaging with Reversed Auto-Encoders

Introduction to Reversed Auto-Encoders (RA)

The field of medical imaging has long been challenged by the need for rapid, accurate, and unbiased detection of pathologies across various imaging modalities. Despite advancements in supervised anomaly detection methods, their inherent bias towards known distributions of pathologies limits their utility in recognizing the wide spectrum of potential anomalies. Unsupervised anomaly detection approaches, while promising, have faced difficulties in delivering broad and unbiased detection capabilities. To address these challenges, a novel unsupervised approach termed Reversed Auto-Encoders (RA) has been introduced, aiming to significantly enhance the detection of a broader range of pathologies across different anatomies and imaging techniques.

Methodology behind RA

RA operates on the principle of creating realistic pseudo-healthy reconstructions of pathological inputs, thereby enabling a nuanced and comprehensive detection of anomalies. This approach relies on a sophisticated training regime that combines the regulative properties of the Evidence Lower Bound (ELBO) with an introspective adversarial scheme and a newly introduced 'reversed loss' mechanism. This trio of strategies ensures the model accurately reconstructs normal anatomical patterns, laying the groundwork for effective anomaly detection.

Extensive Evaluation across Modalities

RA has been rigorously evaluated across various imaging modalities, including MRI scans of the brain, pediatric wrist X-rays, and chest X-rays. Compared to existing state-of-the-art methods, RA has consistently demonstrated superior performance in detecting anomalies across all tested scenarios. For instance, in the domain of brain MRI, RA has shown a distinct ability to detect and localize anomalies, with a compelling F1 score that underscores its effectiveness.

Anomaly Detection in Pediatric Wrist X-rays

In the analysis of pediatric wrist X-rays, RA’s capability to accurately identify and localize various abnormalities underlines its potential utility in enhancing the diagnostic processes for common pediatric injuries. The method's adaptability and performance in detecting soft tissue abnormalities and fractures highlight its clinical relevance.

Enhancing Chest X-ray Analysis

Chest X-rays, a critical tool in diagnosing respiratory conditions, have also benefitted from RA's anomaly detection capabilities. The method's performance in distinguishing between normal and pathological findings, particularly in cases of pneumonia and COVID-19, showcases its potential to support faster and more accurate diagnoses, which is essential amidst global health challenges like the COVID-19 pandemic.

Future Implications and Developments in AI-Driven Medical Imaging

The introduction of RA marks a significant advancement in unsupervised anomaly detection in medical imaging. Its ability to generate pseudo-healthy reconstructions for a wide array of pathologies presents a promising avenue for facilitating more accurate and unbiased anomaly detection. Looking ahead, the continued refinement of RA and exploration of its applicability across a broader spectrum of medical imaging scenarios could revolutionize diagnostic processes. Moreover, addressing current limitations, such as the detection of very subtle anomalies, will be crucial in evolving RA's utility and effectiveness in clinical settings.

In conclusion, RA represents a substantive contribution to the field of medical imaging, demonstrating the potential of unsupervised anomaly detection methods to transform diagnostic methodologies. As this research progresses, it may pave the way for the development of universally applicable, AI-driven diagnostic tools, ultimately enhancing patient care and the efficiency of medical processes.

File Document Download Save Streamline Icon: https://streamlinehq.com PDF File Document Download Save Streamline Icon: https://streamlinehq.com Markdown
List To Do Tasks Checklist Streamline Icon: https://streamlinehq.com

Collections

Sign up for free to add this paper to one or more collections.

User Circle Single Streamline Icon: https://streamlinehq.com Sign Up