Cape Town - 2026 ISMRM-ISMRT Annual Meeting and Exhibition - Noise and Artifact Mitigation in MRI

Noise and Artifact Mitigation in MRI

Digital Poster

Analysis Methods

Thursday, 14 May 2026

Digital Posters Row B

13:40 - 14:35

Session Number: 661-03

No CME/CE Credit

Similar Sessions

The session covers image enhancement techniques with a particular focus on mitigation of noise and artifacts.

Skill Level: Basic,Intermediate,Advanced

		661-03-001. Shuffled Repetition-to-Repetition Learning (Rep2Rep-Shuffle) for Noise-Adaptive Self-Supervised Denoising in Sodium MRI Renqing Luo, Nikola Janjusevic, Haoyang Pei, Yao Wang, Guillaume Madelin, Li Feng Bernard and Irene Schwartz Center for Biomedical Imaging, Department of Radiology, New York University Grossman School of Medicine, New York, New York, United States of America Impact: Rep2Rep-Shuffle improves image SNR in low-SNR MRI without relying on supervised training references, enabling shorter scans with reduced repetitions. Its noise-adaptive feature offers a general framework for efficient denoising across different low-SNR MRI applications.
		661-03-002. Graph2Self: Fast Self-Supervised Denoising of Diffusion MRI via Graph-Based Collaborative Filtering Kamyar Rajabalifardi, Onat Dalmaz, Akshay Chaudhari Stanford University, Stanford, United States of America Impact: Graph2Self enables fast, training-free denoising of 4D diffusion MRI by learning a q-space graph and applying slice-wise collaborative filtering. It preserves anatomy and downstream metrics while drastically reducing runtime, facilitating robust, reproducible preprocessing for clinical protocols, large-scale studies, and research.
		661-03-003. Self-Supervised Denoising Reconstruction of 7T ASL Perfusion-Weighted Images from Fewer Frames Jian Hu, Caohui Duan, Zhixuan Li, Junfeng Zhang, Xiaojun Yu, Cong Zhang, Jianxun Qu, Jinhao Lyu, Xin Lou The First Medical Center, Chinese PLA General Hospital, Beijing, China Impact: The proposed denoising model enables the reconstruction of high-SNR PWIs from fewer repeats, potentially making 7T ASL more practical by cutting scan time and mitigating motion artifacts, albeit with reduced CNR.
		661-03-004. Denoising of High-resolution 3D UTE-MR Angiogram Data using Lightweight and Efficient Convolutional Neural Networks Abel Tessema, Dagnachew Ambaye, HyungJoon Cho Ulsan National Institute of Science and Technology, Ulsan, Korea, Republic of Impact: We propose an efficient CNN-based denoising frameworks for high-resolution 3D UTE-MR angiograms that enhances vessel clarity and diagnostic reliability while reducing noise and computational cost, enabling practical, high-quality, radiation-free vascular imaging.
		661-03-005. KLEAN: A Generalized Acquisition-agnostic LLR k-space Denoising Method for High-dimensional Imaging Ludwig Sichen Zhao, Manuel Taso, John Detre, M. Dylan Tisdall University of Pennsylvania, Philadelphia, United States of America Impact: KLEAN is a robust and versatile k-space denoising method applicable to all high-dimensional MRI acquisitions. It improves image quality and acquisition efficiency, allowing higher spatiotemporal resolution or, alternatively, shorter acquisition times across diverse imaging applications.
		661-03-006. PCMRI: Leveraging Latent Diffusion Models for Prompt-Controlled Text-to-Image MRI Synthesis Junjie WU, Yanlin Wu, Xingjian TANG, Kai TONG, Fujin Ai, Jingwei GUAN Shenzhen Technology University, Shenzhen, China Impact: The proposed Prompt-Controlled MRI synthesis method (PCMRI) can generate multi-slice high-quality MR images. It provides robust data support for downstream reconstruction tasks, and is promising to enhance both diagnostic efficiency and accuracy.
		661-03-007. Deep Residual Learning for Artifact Suppression in Simultaneous Sodium MRI–EEG Acquisition Nikolaos Prasinos, Ying-Chia Lin, Sara Hejazi, Kamri Clarke, Justin Quimbo, Malika Kumbella, Kennedy Watson, Simon Henin, Anli Liu, Arjun Masurkar, Yulin Ge, Yvonne Lui, Yongxian Qian New York University, New York, United States of America Impact: This work evaluates EEG artifact removal strategies during sodium MRI. With sufficient training data, supervised learning demonstrates superior performance in mitigating MRI-induced interference compared with conventional ICA, enabling cleaner EEG signals for simultaneous sodium MRI–EEG acquisitions.
		661-03-008. Pre-Scan Noise Map Guided Deep Learning for Denoising and Rician Bias Correction in Prostate DWI Mustafa Abbas, Fredrik Langkilde, Stephan Maier, Stefan Kuczera Sahlgrenska Academy, University of Gothenburg, Gothenburg, Sweden Impact: This study shows that noise-aware neural networks enable faster model fitting and denoising of multi-b diffusion signal than conventional algorithms. Incorporating scanner-derived noise maps makes the approach more generalizable, with potential for clinical translation, particularly in prostate cancer screening.
		661-03-009. Noise-Aware Fast Magnetic Resonance Spectroscopy Reconstruction with Complex U-Net Huaizhi Liu, Christopher Wu, Lawrence Kegeles, Jia Guo Columbia University, New York, United States of America Impact: With very few in vivo scans needed for finetuning, our ML denoiser can potentially reduce MRS scan time by more than 10 times (vs. 160-transient/voxel baseline), enabling real-time QA and real-time quantification, making functional MRS practical, and increasing per-scanner throughput.
		661-03-010. SACRED: Susceptibility Artifact Correction without Reverse phase-encoding for EPI using Deep learning Wooseung Kim, Sung-Hong Park Korea Advanced Institute of Science & Technology, Daejeon, Korea, Republic of Impact: SACRED offers a fast, accurate, and reverse phase-encoding-free distortion correction solution, enhancing EPI data interpretability and enabling broader use of existing datasets. This advances neuroimaging research by providing a robust tool for improved spatial fidelity.
		661-03-011. MDPMM: Towards MRI Motion artifact modeling via Multi-modal Controllable Generative Diffusion Prior Jiawei YAO, Zihan CHEN, Kai TONG, Junjie WU, Chenchen GE, Jingwei GUAN Shenzhen Technology University, Shenzhen, China Impact: Multi-modal Diffusion Priors for MRI Motion Modelling (MDPMM) is focused on motion modeling, providing abundant high-quality data for downstream tasks like MRI motion/hybrid distortion reconstruction. Moreover, MDPMM is promising for enhanced clinical efficiency and accuracy in practical MRI applications.
		661-03-012. Synthetic data-driven MR image quality enhancement Keerthi Sravan Ravi, Ashok Vardhan Addala, Ravi Soni, Gopal Avinash GE HealthCare, San Ramon, United States of America Impact: We demonstrate that high-performance deep learning models can be trained for MRI artifact removal using synthetic data, overcoming the traditional reliance on limited and sensitive real-world datasets, prompting further investigation into synthetic data for cross-sequence and anatomy generalization.
		661-03-013. Deep learning based correction of FID artifacts in SPACE imaging Hanna Wichtel, Hans-Peter Fautz, Dominik Paul, Florian Putz, Jana Hutter Friedrich-Alexander Universität Erlangen-Nürnberg (FAU), Erlangen, Germany Impact: The proposed deep learning approach corrects FID artifacts in SPACE MRI by learning from single- and two-average acquisitions. It enables artifact-reduced images without a second acquisition, halving scan time and improving imaging efficiency for clinical and research applications.
		661-03-014. Denoising 3D Motion Corrected Image with a Pretrained 2D U-Net Model Yajun Li, Cheng-Chieh Cheng, Jayant Dubey, Jeffrey Guenette, Bruno Madore, Lei Qin Dana Farber Cancer Institute, Boston, United States of America Impact: We aimed to integrate a pretrained 2D U-Net model into a 3D motion compensation pipeline to enhance image quality. The method achieved improved image SNR and reduced artifact levels, in both volunteer and patient scans.
		661-03-015. Self-Consistent Physics-Informed Distortion Correction for DW-PROPELLER-EPI Using Differentiable Field-Map Estimation HAILIN XIONG, Yi Li, Chenglang Yuan, Liyuan Liang, Shihui Chen, Tianbaige Liu, Teng-Yi Huang, Qi DOU, Hing-Chiu Chang The Chinese University of Hong Kong, Shatin, Hong Kong Impact: This physics-informed, calibration-free framework enables high-fidelity, distortion-free DW-PROPELLER-EPI reconstruction without additional calibration, such as paired blades with reversed phase-encoding or separate field-mapping scans, supporting faster and more robust clinical brain DWI.
		661-03-016. Accelerated Eddy-corrected Multi-average Diffusion Plug-and-play Reconstruction in DWI Yitong Yang, Shihan Qiu, Yahang Li, Wei Liu, Mahmoud Mostapha, Radu Miron, Thorsten Feiweier, David Grodzki, Rainer Schneider, Mariappan Nadar Biomedical Engineering, Georgia Institute of Technology and Emory University, Atlanta, United States of America Impact: The proposed multi-average joint diffusion plug-and-play framework substantially reduces the computational time required for the generative diffusion-based DWI reconstruction. It achieves higher SNR, improves image sharpness, and enables clearer delineation of anatomical details even with fewer averages employed.