FgC2F-UDiff: Frequency-guided and Coarse-to-fine Unified Diffusion Model for Multi-modality Missing MRI Synthesis

TL;DR

A novel unified synthesis model, the Frequency-guided and Coarse-to-fine Unified Diffusion Model (FgC2F-UDiff), designed for multiple inputs and outputs, and integrates specific mechanisms to accelerate the diffusion model and enhance the feasibility of many-to-many synthesis.

Abstract

Multi-modality magnetic resonance imaging (MRI) is essential for the diagnosis and treatment of brain tumors. However, missing modalities are commonly observed due to limitations in scan time, scan corruption, artifacts, motion, and contrast agent intolerance. Synthesis of missing MRI has been a means to address the limitations of modality insufficiency in clinical practice and research. However, there are still some challenges, such as poor generalization, inaccurate non-linear mapping, and slow processing speeds. To address the aforementioned issues, we propose a novel unified synthesis model, the Frequency-guided and Coarse-to-fine Unified Diffusion Model (FgC2F-UDiff), designed for multiple inputs and outputs. Specifically, the Coarse-to-fine Unified Network (CUN) fully exploits the iterative denoising properties of diffusion models, from global to detail, by dividing the denoising process into two stages, coarse and fine, to enhance the fidelity of synthesized images. Secondly, the Frequency-guided Collaborative Strategy (FCS) harnesses appropriate frequency information as prior knowledge to guide the learning of a unified, highly non-linear mapping. Thirdly, the Specific-acceleration Hybrid Mechanism (SHM) integrates specific mechanisms to accelerate the diffusion model and enhance the feasibility of many-to-many synthesis. Extensive experimental evaluations have demonstrated that our proposed FgC2F-UDiff model achieves superior performance on two datasets, validated through a comprehensive assessment that includes both qualitative observations and quantitative metrics, such as PSNR SSIM, LPIPS, and FID.

Figures (9)

• Figure 1: Different image modalities provide different types of tissue contrast views and spatial resolution, which has variety in the histogram.
• Figure 2: Overview of the proposed FgC2F-UDiff, which cross-modality synthesizes missing modalities from multiple inputs and outputs. It includes forward diffusion progress and coarse-to-fine reverse denoising progress. The FgC2F-UDiff decomposes the frequency domain into low-high-frequency as guidance information in coarse-to-fine stages based on the iterative denoising properties of diffusion models.
• Figure 3: Visualize analyzing and visualizing the denoising synthesis images, which shows the significant properties of iterative denoising. Specifically, (a) shows denoised synthesis images corresponding to different time steps $T$. (b) shows the low frequency. (c) shows the high frequency.
• Figure 4: The image can usually be decomposed into a high-frequency sub-band with edges and details and a low-frequency sub-band with anatomical structure.
• Figure 5: Illustrative instances of synthetic images generated by our FgC2F-UDiff on the BraTS Dataset. Each row shows composite diagrams portraying distinct modes and error maps juxtaposed with the corresponding ground truth. The enlarged orange squares represent selected regions with notable disparities, providing enhanced insights into texture, edge enhancement, and shape characteristics.