[Paper Review] Diffusion Models for Medical Image Analysis: A Comprehensive Survey

Amirhossein Kazerouni, Ehsan Khodapanah Aghdam+5 more|arXiv (Cornell University)|Nov 14, 2022

Mathematical Biology Tumor Growth46 citations

TL;DR

This survey comprehensively reviews diffusion models in medical imaging, offering a taxonomy, applications across modalities, clinical relevance, limitations, and open-source resources.

ABSTRACT

Denoising diffusion models, a class of generative models, have garnered immense interest lately in various deep-learning problems. A diffusion probabilistic model defines a forward diffusion stage where the input data is gradually perturbed over several steps by adding Gaussian noise and then learns to reverse the diffusion process to retrieve the desired noise-free data from noisy data samples. Diffusion models are widely appreciated for their strong mode coverage and quality of the generated samples despite their known computational burdens. Capitalizing on the advances in computer vision, the field of medical imaging has also observed a growing interest in diffusion models. To help the researcher navigate this profusion, this survey intends to provide a comprehensive overview of diffusion models in the discipline of medical image analysis. Specifically, we introduce the solid theoretical foundation and fundamental concepts behind diffusion models and the three generic diffusion modelling frameworks: diffusion probabilistic models, noise-conditioned score networks, and stochastic differential equations. Then, we provide a systematic taxonomy of diffusion models in the medical domain and propose a multi-perspective categorization based on their application, imaging modality, organ of interest, and algorithms. To this end, we cover extensive applications of diffusion models in the medical domain. Furthermore, we emphasize the practical use case of some selected approaches, and then we discuss the limitations of the diffusion models in the medical domain and propose several directions to fulfill the demands of this field. Finally, we gather the overviewed studies with their available open-source implementations at https://github.com/amirhossein-kz/Awesome-Diffusion-Models-in-Medical-Imaging.

Motivation & Objective

• Introduce the theoretical foundations of diffusion models and their relevance to medical imaging.
• Provide a systematic taxonomy of diffusion-model approaches and applications in medicine.
• Survey diffusion-model applications across modalities, organs, and tasks.
• Discuss limitations, challenges, and future directions for diffusion models in clinical settings.

Proposed method

• Explain diffusion model theory and frameworks: diffusion probabilistic models, noise-conditioned score networks, and stochastic differential equations.
• Propose a multi-perspective taxonomy based on application, imaging modality, organ, and algorithm.
• Review literature up to October 2022 and through April 2023 to cover diffusion-model applications in medicine.
• Summarize practical use cases and provide a GitHub repository of open-source implementations.
• Highlight clinical relevance and potential of synthetic data and diffusion priors in inverse imaging problems.

Experimental results

Research questions

• RQ1 What are the core theoretical frameworks of diffusion models applicable to medical imaging?
• RQ2 How are diffusion models categorized across applications, modalities, and organs in medicine?
• RQ3 What are the main clinical benefits, limitations, and open challenges of using diffusion models in medical imaging?
• RQ4 What guidance and resources (e.g., open-source implementations) exist to advance diffusion-model research in healthcare?

Key findings

• Diffusion models offer strong sample quality and mode coverage, addressing some limitations of GANs and VAEs in medical imaging.
• The literature on diffusion models in medicine has grown rapidly, totaling 103 papers at the time of the survey.
• This is the first survey to comprehensively cover diffusion-model applications in medical imaging and provides a multi-perspective taxonomy.
• The study discusses clinical relevance, data scarcity, privacy, and the potential of synthetic data and diffusion priors for inverse imaging problems.
• An open-resource GitHub repository aggregates open-source implementations of diffusion-model methods in medical imaging.
• The survey identifies nine application categories and emphasizes the applicability across modalities and organs.