A Systematic Review of Diffusion Models for Medical Image-Based Diagnosis: Methods, Taxonomies, Clinical Integration,...

Area of Science:

• * Artificial Intelligence
• * Computer Vision
• * Health Informatics

Background:

• * Diffusion models are advanced generative models crucial for high-resolution image synthesis and reconstruction.
• * Their application is rapidly expanding in computer vision and health informatics, particularly in medical imaging.
• * A structured overview of diffusion models in clinical imaging is lacking, posing challenges for researchers.

Purpose of the Study:

• * To systematically review the application of diffusion models in medical imaging for diagnostic purposes.
• * To provide an integrated overview of diffusion model principles, applications, and limitations in this field.
• * To identify future research directions for clinical translation.

Main Methods:

• * Systematic review following PRISMA 2020 guidelines (2013-2024).
• * Inclusion of peer-reviewed studies using diffusion models for medical diagnostic tasks.
• * Narrative synthesis approach due to study heterogeneity; risk of bias assessment.

Main Results:

• * 68 studies included, covering anomaly detection, classification, denoising, generation, reconstruction, segmentation, super-resolution, and image-to-image translation.
• * Explainable AI (22.06%), clinician engagement (57.35%), and real-time implementation (10.30%) were noted.
• * Findings indicate strong diagnostic potential but highlight variability in reporting, methodological inconsistencies, and limited real-world validation.

Conclusions:

• * Diffusion models hold significant promise for diagnostic imaging.
• * Clinical deployment requires advancements in explainability, clinician integration, and real-time capabilities.
• * Twelve key research directions are identified to guide future development and clinical translation.