Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Computationally Efficient Diffusion Models in Medical Imaging: A Comprehensive Review

Abdullah, Tao Huang, Ickjai Lee, Euijoon Ahn

TL;DR

Diffusion models are powerful but computationally intensive, posing challenges for medical imaging where speed and reliability are critical. This survey comprehensively analyzes three efficiency-focused families—DDPM, LDM, and WDM—and their applications in natural and medical imaging, highlighting how latent-space and wavelet-domain approaches address the speed–quality trade-off. It details the foundations, advantages, and limitations of each family, along with extensive applications across generation, translation, reconstruction, segmentation, and denoising in medical imaging. The work offers practical guidance on model selection for fast, high-quality medical image synthesis, reconstruction, and analysis, and outlines future research directions to further reduce latency while preserving diagnostic fidelity.

Abstract

The diffusion model has recently emerged as a potent approach in computer vision, demonstrating remarkable performances in the field of generative artificial intelligence. Capable of producing high-quality synthetic images, diffusion models have been successfully applied across a range of applications. However, a significant challenge remains with the high computational cost associated with training and generating these models. This study focuses on the efficiency and inference time of diffusion-based generative models, highlighting their applications in both natural and medical imaging. We present the most recent advances in diffusion models by categorizing them into three key models: the Denoising Diffusion Probabilistic Model (DDPM), the Latent Diffusion Model (LDM), and the Wavelet Diffusion Model (WDM). These models play a crucial role in medical imaging, where producing fast, reliable, and high-quality medical images is essential for accurate analysis of abnormalities and disease diagnosis. We first investigate the general framework of DDPM, LDM, and WDM and discuss the computational complexity gap filled by these models in natural and medical imaging. We then discuss the current limitations of these models as well as the opportunities and future research directions in medical imaging.

Computationally Efficient Diffusion Models in Medical Imaging: A Comprehensive Review

TL;DR

Diffusion models are powerful but computationally intensive, posing challenges for medical imaging where speed and reliability are critical. This survey comprehensively analyzes three efficiency-focused families—DDPM, LDM, and WDM—and their applications in natural and medical imaging, highlighting how latent-space and wavelet-domain approaches address the speed–quality trade-off. It details the foundations, advantages, and limitations of each family, along with extensive applications across generation, translation, reconstruction, segmentation, and denoising in medical imaging. The work offers practical guidance on model selection for fast, high-quality medical image synthesis, reconstruction, and analysis, and outlines future research directions to further reduce latency while preserving diagnostic fidelity.

Abstract

The diffusion model has recently emerged as a potent approach in computer vision, demonstrating remarkable performances in the field of generative artificial intelligence. Capable of producing high-quality synthetic images, diffusion models have been successfully applied across a range of applications. However, a significant challenge remains with the high computational cost associated with training and generating these models. This study focuses on the efficiency and inference time of diffusion-based generative models, highlighting their applications in both natural and medical imaging. We present the most recent advances in diffusion models by categorizing them into three key models: the Denoising Diffusion Probabilistic Model (DDPM), the Latent Diffusion Model (LDM), and the Wavelet Diffusion Model (WDM). These models play a crucial role in medical imaging, where producing fast, reliable, and high-quality medical images is essential for accurate analysis of abnormalities and disease diagnosis. We first investigate the general framework of DDPM, LDM, and WDM and discuss the computational complexity gap filled by these models in natural and medical imaging. We then discuss the current limitations of these models as well as the opportunities and future research directions in medical imaging.
Paper Structure (62 sections, 9 equations, 6 figures, 9 tables)

This paper contains 62 sections, 9 equations, 6 figures, 9 tables.

Table of Contents

  1. Introduction
  2. Preliminary: Diffusion Models Concept
  3. Diffusion Models
  4. Foundation of Denoising Diffusion Probabilistic Model (DDPM)
  5. Latent Diffusion Model (LDM)
  6. Wavelet Diffusion Model (WDM)
  7. Critical Analysis and Discussion
  8. Efficiency:
  9. Stability:
  10. Fast Convergence:
  11. Image Generation Quality:
  12. Efficient Sampling:
  13. Generating Large Images:
  14. Application in Natural and Medical Imaging:
  15. Applications
  16. ...and 47 more sections

Figures (6)

  • Figure 1: This reflects the evolving trends in landscape research. (a) The annual number of papers published related to DDPM, LDM, and WDM; (b) A total of 229 papers were identified. Within these categories, 130 papers focus on DDPM, 68 papers on LDM, and 31 papers on WDM.
  • Figure 2: An Illustration of the DDPM forward process $q(x_t| x_{t_1})$ add noise and reverse process $p_{\theta}(x_{T-1}|x_t)$ removing noise.
  • Figure 3: LDM basic framework encodes data from the pixel space into a latent space. Within this latent space, the diffusion process is carried out, guided by the conditional input modality for synthesis. Finally, the model decodes the data back into the image domain RN199.
  • Figure 4: The wavelet transform method decomposes an image into its frequency components at different scales to capture both contextual and wide-range features efficiently: Implementation of 2D-DWT.
  • Figure 5: We categorize diffusion models into models, applications, and future directions. Notably, DDPM, LDM, and WDM are highlighted for their efficiency and stability. Applications are further categorized into two mainstream natural and medical Images with their specific tasks. For clarity, paper references are indicated by ascending prefix numbers: (i) RN259, (ii) RN447, (iii) RN233, (iv) RN448, (v) RN267, (vi) RN327, (vii) RN265, (viii) RN266, (ix) RN186, (x) RN416, (xi) RN417, (xii) RN418, (xiii) RN477, (xiv) RN264, (xv) RN427, (xvi) RN335, (xvii) RN257, (xviii) RN393, (xix)RN478, (xx) RN258, (xxi) RN479, (xxii) RN480, (xxiii) RN396, (xxiv) RN445, (xxv) RN481, (xxvi) RN482, (xxvii) RN483, (xxviii) RN273, (xxix) L5, (xxx) RN389, (xxxi) RN459, (xxxii) RN269, (xxxiii) RN484, (xxxiv) RN270, (xxxv) RN412, (xxxvi) RN413, (xxxvii )RN253, (xxxviii) RN306, (xxxix) RN464, (xl) RN486, (xli) RN487, (xlii) RN401.
  • ...and 1 more figures