Title

A Systematic Review on Impact of Diffusion Models in Health Informatics: The Role of Explainable AI and Clinician Involvement

Abstract

Diffusion models, as the advanced evolvement of generative models, have been adopted across various domains and demonstrate remarkable performance in a wide range of applications. Particularly, these models have revolutionized the computer vision field, outperforming state-of-the-art models in generating high-resolution images, reconstructing images, and more. These advancements have started to capitalize on the health informatics domain, especially in medical imaging. However, researchers often struggle due to the lack of a structured discussion on diffusion model applications in this field. This review provides a comprehensive overview of diffusion models, including foundational knowledge and their application in medical imaging. We conducted a literature analysis of 68 papers, categorizing their diffusion model applications into eight domains: Anomaly Detection, Classification, Denoising, Generation, Reconstruction, Segmentation, Super Resolution, and Translation. The review covers various medical image modalities, demonstrating the diversity of applications. Additionally, clinician involvement and explainability were investigated, ensuring model outcome transparency. Our analysis revealed that explainable AI (XAI) is present in 22.06% of the papers, clinician involvement in 57.35%, and realtime implementation in 10.30%. These findings highlight the need for future research to enhance XAI, improve clinician integration, and advance real-time functionalities in diffusion models. As a key contribution, we propose ten future research questions to guide advancements in these areas. This review systematically contextualizes diffusion model applications in medical imaging, emphasizing key areas and use cases and offering valuable insights for future practitioners.

Authors

Mohammad Azad, Tanvir Rahman Anik