mDDPM: Unsupervised Anomaly Detection in Medical Images Using Masked Diffusion Model

Hasan Iqbal1* Umar Khalid2*, Jing Hua1, Chen Chen2,
1Wayne State University 2University of Centra Florida
*Equal Contribution
arXiv Code Data

Abstract

It can be challenging to identify brain MRI anomalies using supervised deep-learning techniques due to anatomical heterogeneity and the requirement for pixel-level labeling. Unsupervised anomaly detection approaches provide an alternative solution by relying only on sample-level labels of healthy brains to generate a desired representation to identify abnormalities at the pixel level. Although, generative models are crucial for generating such anatomically consistent representations of healthy brains, accurately generating the intricate anatomy of the human brain remains a challenge. In this study, we present a method called masked-DDPM (mDPPM), which introduces masking-based regularization to reframe the generation task of diffusion models. Specifically, we introduce Masked Image Modeling (MIM) and Masked Frequency Modeling (MFM) in our self-supervised approach that enables models to learn visual representations from unlabeled data. To the best of our knowledge, this is the first attempt to apply MFM in DPPM models for medical applications. We evaluate our approach on datasets containing tumors and numerous sclerosis lesions and exhibit the superior performance of our unsupervised method as compared to the existing fully/weakly supervised baselines

Schematic diagram of our framework. (a) During training, only healthy images are used, and no classifier guidance is required. The healthy image is passed through the Masking Block before feeding into the DDPM. The reconstruction loss is calculated between the original image and the image generated by the DDPM. Here, Masking Block plays the role of regularizer and eliminates the need for classifier guidance (b) During inference, DDPM considers the tumor in the unhealthy image as an augmented patch and eliminates it to generate a healthy image. The difference between the generated image and the given unhealthy image is then calculated to report the anomaly map. No masking mechanism is employed at inference.
Proposed data augmentation techniques which are implemented in the Masking Block.
Comparison of mDDPM with state-of-the-art technieques.

BibTeX

@misc{iqbal2023unsupervised,https://github.com/orgs/community/discussions/54546
  author    = {Hasan Iqbal and Umar Khalid and Jing Hua and Chen Chen},
  title     = {Unsupervised Anomaly Detection in Medical Images Using Masked Diffusion Model},
  year      = {2023},
  eprint    = {2305.198671},
  archivePrefix={arXiv}, 
  primaryClass={eess.IV}
}