Radiomics-Driven Diffusion Model and Monte Carlo Compression Sampling for Reliable Medical Image Synthesis
IEEE Journal of Biomedical and Health Informatics
|August 25, 2025
Summary
This study introduces a new method for reliable medical image synthesis using radiomics prompts and Monte Carlo Compression Sampling (MCCS) for better uncertainty estimation. The approach enhances image quality and confidence in clinical applications.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computational Medicine
Background:
- Reliable medical image synthesis is vital for clinical tasks, demanding high anatomical accuracy and confidence.
- Current methods struggle with data scarcity of medical text prompts and insufficient uncertainty estimation.
Purpose of the Study:
- To develop a novel approach for dependable medical image synthesis using radiomics prompts.
- To enhance uncertainty quantification for improved reliability in generated medical images.
Main Methods:
- Utilized clinically focused radiomics prompts to guide image generation.
- Implemented Monte Carlo Compression Sampling (MCCS) within denoising diffusion implicit models (DDIM) for uncertainty quantification.
- Introduced a MambaTrans architecture to model long-range dependencies and incorporate prior conditions.
Main Results:
- The proposed method significantly improved medical image quality and reliability.
- Demonstrated superior performance over state-of-the-art (SoTA) methods in qualitative and quantitative evaluations.
- Successfully leveraged radiomics prompts for conditional medical image synthesis.
Conclusions:
- The novel approach effectively addresses challenges in medical image synthesis, particularly data scarcity and uncertainty estimation.
- Radiomics prompts and MCCS offer a promising direction for generating reliable and high-quality medical images.
- The MambaTrans architecture enhances the modeling of complex dependencies in medical imaging data.
