Abstract #2639

Denoising Diffusion Probabilistic Model with Dual-domain Entropy and Stochastic Differential Equations on High Undersampling Reconstruction

Zhixin Li^1,2,3, Yishuang Yang^1,3, Jing An⁴, Chen Ling^5,6, Zhaoxia Wang^5,6, Yan Zhuo^1,3, Rong Xue^1,3,7, and Zihao Zhang^1,2,3

¹State Key Laboratory of Brain and Cognitive Science, Institute of Biophysics, Chinese Academy of Sciences, Beijing, China, ²Anhui Province Key Laboratory of Biomedical Imaging and Intelligent Processing, Institute of Artificial Intelligence, Hefei Comprehensive National Science Center, Hefei, China, ³University of Chinese Academy of Sciences, 19A Yuquan Road, Beijing, China, ⁴Siemens Shenzhen Magnetic Resonance Ltd., Shenzhen, China., Shenzhen, China, ⁵Department of Neurology, Peking University First Hospital, Beijing, China, ⁶Beijing Key Laboratory of Neurovascular Disease Discovery, Peking University First Hospital, Beijing, China, ⁷Beijing Institute for Brain Disorders, Beijing, China

Synopsis

Keywords: AI/ML Image Reconstruction, AI/ML Image Reconstruction

Motivation: Using denoising diffusion probabilistic models (DDPM) for MRI reconstruction remains challenging on high-fold under-sampling.

Goal(s): To improve the reconstruction performance of 8-fold under-sampled data by a modified DDPM.

Approach: A composite loss function based on information entropy, dual-domain difference, and uniform-static-SDE (stochastic differential equations) is proposed, called EDU-DDPM. Our algorithm has been tested on fastMRI and compared to compressed sensing (CS) TOF-MRA data acquired at 7T.

Results: Compared to previous DDPM, our method performs better in single-channel knee and multi-channel brain datasets on fastMRI with 8-fold under-sampling. Additionally, it outperforms CS reconstruction in Time-of-flight (TOF-MRA) acquired in variable density Poisson sampling pattern.

Impact: The proposed EDU-DDPM significantly improves MRI reconstruction at high subsampling factors, outperforming DDPM on fastMRI and compressive sensing on 7T TOF-MRA data. This advancement enhances fidelity of reconstruction.

How to access this content:

For one year after publication, abstracts and videos are only open to registrants of this annual meeting. Registrants should use their existing login information. Non-registrant access can be purchased via the ISMRM E-Library.

After one year, current ISMRM & ISMRT members get free access to both the abstracts and videos. Non-members and non-registrants must purchase access via the ISMRM E-Library.

After two years, the meeting proceedings (abstracts) are opened to the public and require no login information. Videos remain behind password for access by members, registrants and E-Library customers.

Click here for more information on becoming a member.