[Paper Review] DP-MDM: Detail-Preserving MR Reconstruction via Multiple Diffusion Models

Mengxiao Geng, Jiahao Zhu+4 more|arXiv (Cornell University)|May 9, 2024

Advanced MRI Techniques and ApplicationsMedicine3 citations

TL;DR

This paper proposes DP-MDM, a novel MRI reconstruction method using multiple diffusion models in the k-space domain to preserve fine anatomical details. By leveraging virtual binary masks, an inverted pyramid structure, and cascade training, it achieves superior detail recovery and reconstruction quality over state-of-the-art methods on clinical and public datasets.

ABSTRACT

Detail features of magnetic resonance images play a cru-cial role in accurate medical diagnosis and treatment, as they capture subtle changes that pose challenges for doc-tors when performing precise judgments. However, the widely utilized naive diffusion model has limitations, as it fails to accurately capture more intricate details. To en-hance the quality of MRI reconstruction, we propose a comprehensive detail-preserving reconstruction method using multiple diffusion models to extract structure and detail features in k-space domain instead of image do-main. Moreover, virtual binary modal masks are utilized to refine the range of values in k-space data through highly adaptive center windows, which allows the model to focus its attention more efficiently. Last but not least, an inverted pyramid structure is employed, where the top-down image information gradually decreases, ena-bling a cascade representation. The framework effective-ly represents multi-scale sampled data, taking into ac-count the sparsity of the inverted pyramid architecture, and utilizes cascade training data distribution to repre-sent multi-scale data. Through a step-by-step refinement approach, the method refines the approximation of de-tails. Finally, the proposed method was evaluated by con-ducting experiments on clinical and public datasets. The results demonstrate that the proposed method outper-forms other methods.

Motivation & Objective

• To address the limitation of naive diffusion models in capturing subtle, high-frequency anatomical details in MRI.
• To improve reconstruction quality by modeling structure and detail features directly in the k-space domain.
• To enhance feature representation through an inverted pyramid architecture with cascade training.
• To refine k-space data using adaptive center windows via virtual binary modal masks.
• To achieve state-of-the-art performance in detail-preserving MRI reconstruction.

Proposed method

• Multiple diffusion models are trained in the k-space domain to extract both structural and fine-grained detail features.
• Virtual binary modal masks are introduced to dynamically adjust the center window of k-space data, enabling adaptive value range refinement.
• An inverted pyramid structure is employed to progressively reduce image resolution from top to bottom, enabling multi-scale representation.
• The framework uses cascade training to model multi-scale data distributions, improving feature learning across scales.
• A step-by-step refinement process gradually enhances detail approximation in the reconstruction pipeline.
• The method operates entirely in k-space, avoiding the limitations of image-domain diffusion modeling for fine details.

Experimental results

Research questions

• RQ1 Can multiple diffusion models in k-space outperform single-model or image-domain diffusion approaches in preserving MRI details?
• RQ2 How effective are virtual binary modal masks in refining k-space data for improved detail representation?
• RQ3 To what extent does the inverted pyramid architecture enhance multi-scale feature learning in MRI reconstruction?
• RQ4 Does cascade training improve the model's ability to reconstruct fine anatomical structures?
• RQ5 How does DP-MDM compare to existing state-of-the-art methods in terms of quantitative and qualitative detail preservation?

Key findings

• The proposed DP-MDM method achieves superior reconstruction quality compared to existing state-of-the-art methods on both clinical and public MRI datasets.
• The use of multiple diffusion models in k-space significantly improves the recovery of fine anatomical details compared to naive diffusion models.
• Virtual binary modal masks effectively adapt the k-space center window, enhancing model focus on relevant frequency components.
• The inverted pyramid structure enables effective multi-scale feature representation, contributing to better detail recovery.
• Cascade training improves the model's ability to learn and reconstruct complex, sparse, high-frequency patterns in MRI data.
• Quantitative results demonstrate that DP-MDM outperforms baseline methods in metrics such as PSNR, SSIM, and LPIPS, indicating improved structural and perceptual fidelity.