Abstract:Diffusion model-based approaches recently achieved re-markable success in MRI reconstruction, but integration into clinical routine remains challenging due to its time-consuming convergence. This phenomenon is partic-ularly notable when directly apply conventional diffusion process to k-space data without considering the inherent properties of k-space sampling, limiting k-space learning efficiency and image reconstruction quality. To tackle these challenges, we introduce subspace diffusion model with orthogonal decomposition, a method (referred to as Sub-DM) that restrict the diffusion process via projections onto subspace as the k-space data distribution evolves toward noise. Particularly, the subspace diffusion model circumvents the inference challenges posed by the com-plex and high-dimensional characteristics of k-space data, so the highly compact subspace ensures that diffusion process requires only a few simple iterations to produce accurate prior information. Furthermore, the orthogonal decomposition strategy based on wavelet transform hin-ders the information loss during the migration of the vanilla diffusion process to the subspace. Considering the strate-gy is approximately reversible, such that the entire pro-cess can be reversed. As a result, it allows the diffusion processes in different spaces to refine models through a mutual feedback mechanism, enabling the learning of ac-curate prior even when dealing with complex k-space data. Comprehensive experiments on different datasets clearly demonstrate that the superiority of Sub-DM against state of-the-art methods in terms of reconstruction speed and quality.

What problem does this paper attempt to address?

The paper attempts to address the issue in Magnetic Resonance Imaging (MRI) reconstruction where Diffusion Models (DMs), despite achieving significant success in image reconstruction quality, face a major obstacle in clinical practice due to slow convergence speed. This is especially true when directly applied to k-space data, as the inherent characteristics of k-space sampling are not considered, leading to inefficient k-space learning and decreased image reconstruction quality. Specifically, the paper proposes a method based on subspace diffusion models and orthogonal decomposition (referred to as Sub-DM), aiming to accelerate MRI reconstruction and improve reconstruction quality by confining the diffusion process within a subspace. This method effectively addresses the complexity of high-dimensional k-space data, reduces the number of iterations, and thus achieves fast and accurate image reconstruction. ### Main Problem Summary: 1. **Slow convergence speed of diffusion models in MRI reconstruction**: Existing diffusion models require a large number of reverse steps to generate high-quality images, resulting in heavy computational burden. 2. **Complexity and high-dimensional characteristics of k-space data**: Directly performing the diffusion process on high-dimensional k-space data leads to information loss and inefficient learning. 3. **Limitations of existing optimization methods**: Current optimization methods mainly focus on the optimization of the model itself, neglecting the intrinsic characteristics of k-space data, making it difficult to achieve fast convergence. ### Solution: - **Subspace Diffusion Model (Sub-DM)**: By confining the diffusion process within a subspace, the complexity of high-dimensional data is reduced, thereby accelerating convergence speed. - **Orthogonal Decomposition**: Utilizing wavelet transform for orthogonal decomposition ensures no information loss during dimensionality reduction, and the entire process is reversible, allowing mutual feedback in different spaces to enhance the model's learning capability. - **Experimental Validation**: Comprehensive experiments on different datasets demonstrate that Sub-DM outperforms existing methods in terms of reconstruction speed and quality, especially under high undersampling rates. ### Main Contributions of the Paper: 1. **Dimensionality reduction through orthogonal decomposition**: Reducing the dimensionality of high-dimensional k-space data to improve the convergence speed and reconstruction performance of diffusion models. 2. **Utilizing wavelet transform for orthogonal decomposition**: Extracting relevant features to optimize the solution space of the reverse diffusion process. 3. **Experimental results**: Demonstrating the fast convergence speed and high reconstruction accuracy of Sub-DM under high undersampling rates on different datasets.

Sub-DM:Subspace Diffusion Model with Orthogonal Decomposition for MRI Reconstruction

Reconstructing High-Quality Diffusion MRI Data from Orthogonal Slice-Undersampled Data Using Graph Convolutional Neural Networks

High-Frequency Space Diffusion Models for Accelerated MRI

High-Frequency Space Diffusion Model for Accelerated MRI

Rethinking Diffusion Model for Multi-Contrast MRI Super-Resolution

Diffusion Modeling with Domain-conditioned Prior Guidance for Accelerated MRI and qMRI Reconstruction

Multifold Acceleration of Diffusion MRI Via Deep Learning Reconstruction from Slice-Undersampled Data.

Correlated and Multi-frequency Diffusion Modeling for Highly Under-sampled MRI Reconstruction

K-space Cold Diffusion: Learning to Reconstruct Accelerated MRI without Noise

CDiffMR: Can We Replace the Gaussian Noise with K-Space Undersampling for Fast MRI?

Diffusion model based on generalized map for accelerated MRI

SPIRiT-Diffusion: Self-Consistency Driven Diffusion Model for Accelerated MRI

DP-MDM: Detail-Preserving MR Reconstruction via Multiple Diffusion Models

Discrete residual diffusion model for high-resolution prostate MRI synthesis

Self-Learned Kernel Low Rank Approach TO Accelerated High Resolution 3D Diffusion MRI

Learning to reconstruct accelerated MRI through K-space cold diffusion without noise

Physics-Informed DeepMRI: k-Space Interpolation Meets Heat Diffusion

Zero-shot Dynamic MRI Reconstruction with Global-to-local Diffusion Model

Simultaneous q-Space Sampling Optimization and Reconstruction for Fast and High-fidelity Diffusion Magnetic Resonance Imaging

Submillimeter diffusion MRI using an in-plane segmented 3D multi-slab acquisition and denoiser-regularized reconstruction

DisC-Diff: Disentangled Conditional Diffusion Model for Multi-Contrast MRI Super-Resolution