Meeting Banner
Abstract #3973

Universal Denoising for MRI DICOM Images Across Diverse Clinical Conditions Through Variational Diffusion Model

Yuchen Shao1, Yingwei Qiu2, Dongsheng Li3, Lingyan Zhang4, Shiwei Lin2, Shengli Chen2, Peiqi Chen2, Xiangfei Jin2, Lingyue Du Du2, Yuning Gu5, Xiaoqian Huang5, Aowen Liu5, Jiafu Zhong5, Shu Liao5, Kaicong Sun1, and Dinggang Shen1,5,6
1School of Biomedical Engineering & State Key Laboratory of Advanced Medical Materials and Devices, ShanghaiTech University, Shanghai, China, 2Department of Radiology, Shenzhen Nanshan People's Hospital, Shenzhen University, Shenzhen, China, 3Departments of Radiology, Panyu 2nd People's Hospital, Guangzhou, China, 4Departments of Radiology, Longgang Central Hospital, Shenzhen, China, 5Shanghai United Imaging Intelligence Co., Ltd., Shanghai, China, 6Shanghai Clinical Research and Trial Center, Shanghai, China

Synopsis

Keywords: AI Diffusion Models, Image Reconstruction

Motivation: MRI DICOM images can be noisy under high acceleration factors. Denoising highly accelerated DICOM images becomes crucial for disease analysis, and allows faster MR imaging.

Goal(s): We aim to build a universal denoising model, which is applicable to diverse clinical conditions without compromising diagnostic information.

Approach: We propose a sharpness-enhanced variational diffusion model by combining an elaborately designed degradation model with a pre-trained diffusion model.

Results: Our method was trained on an in-house large-scale real-world noisy and clean DICOM image pairs from three hospitals. It outperforms the SOTA methods by 1.58% in SSIM and 0.09 in LPIPS across multi-organ, multi-contrast, and multi-vendor conditions.

Impact: This study provides a novel solution to the over-smoothness issue for diffusion models when dealing with diverse and complex real-world data. Our model shows promising denoising performance on real-world clinical images scanned with 2x or 3x acceleration factor.

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.

Keywords