Cape Town - 2026 ISMRM-ISMRT Annual Meeting and Exhibition - Next-Generation MRI Image Enhancement

Next-Generation MRI Image Enhancement

Power Pitch

Analysis Methods

Tuesday, 12 May 2026

Power Pitch Theatre 2

16:00 - 17:36

Moderators: Mariya Doneva & Merlin Fair

Session Number: 452-03

No CME/CE Credit

Similar Sessions

This session focuses on AI-driven MRI image formation and enhancement, spanning the major quality-limiting factors encountered in practice: noise, limited spatial/temporal resolution, artifacts (e.g., Gibbs ringing), and protocol/field-strength variability. The talks highlight modern approaches for denoising and super-resolution, contrast/field-strength transfer for neonatal and infant imaging, and robust, acquisition-agnostic methods operating in both image space and k-space. Multiple presentations emphasize clinical feasibility and validation, including accelerated musculoskeletal MRI, improved UTE lung imaging for nodule evaluation, and high-fidelity diffusion and flow imaging (4D Flow, ASL angiography).

Skill Level: Advanced

16:00		452-03-001. A Universal Deep Model for Gibbs Artifacts Removal in MRI Yu Zhou, Jia Ning, Hongyu Guo, Wei Xi, Hongbin Wang, Weinan Tang Midea Corporate Research Center, Shanghai, China Impact: The proposed Gibbs artifacts removal method demonstrates robust generalization across diverse image contrasts, anatomical regions, and sequences, without requiring retraining. This underscores its significant potential for clinical translation and broad applicability in real-world settings.
16:02		452-03-002. Diffusion MRI-guided Infant Brain MRI Tissue Contrast Enhancement through Image Translation Yingqi Hao, Haoxiang Jiang, Anqi Qiu The Hong Kong Polytechnic University, Hong Kong, Hong Kong Impact: This study introduces the first deep learning–based framework for enhancing infant brain MRI contrast using diffusion-derived information, significantly improving WM–GM delineation and diagnostic quality across developmental stages while maintaining anatomical fidelity.
16:04		452-03-003. Improving Ultra-low-field Cardiac Cine MRI through Transfer Learning Vick Lau, Ye Ding, Shi Su, Jiahao Hu, Junhao Zhang, Alex T. L. Leong, Yujiao Zhao, Ed X Wu The University of Hong Kong, Hong Kong, China Impact: This work demonstrates that 0.05 Tesla free-breathing 3D cardiac cine images can be enhanced by a transformer-based 4D deep learning model with transfer learning. The improvement in image fidelity will facilitate assessment of cardiac function and morphology at 0.05 Tesla.
16:06		452-03-004. Deep 4D Spatiotemporal U-Net for Joint Denoising and Superresolution of Intracranial Aneurysm 4DFlow-MRI Aryan Ghazipour, Amirkhosro Kazemi, Laurel Marsh, MJ Negahdar, Isaac Josh Abecassis, Juan Cebral, Amir Amini University of Louisville, Louisville, United States of America Impact: This work introduces a physics-informed 4D U-Net that integrates spatial and temporal flow constraints to denoise and superresolve 4DFlow-MRI. By leveraging diverse aneurysmal CFD training data, the model achieves improved physical consistency and generalization across complex intracranial aneurysm geometries.
16:08		452-03-005. Multi-input Generalised-Hilbert Mamba for Super-resolution of Ultra-Low-Field MRI Levente Baljer, Niall Bourke, Kirsten Donald, Layla Bradford, Simone Williams, Sean Deoni, Steven Williams, Rosalyn Moran, Emma Robinson, František Váša King's College London, London, United Kingdom Impact: Hyperfine ultra-low-field MRI scanners allow for the study of neurodevelopment in low-income settings, however their diminished signal-to-noise ratio limits their applicability for research and clinical use. Deep-learning approaches can enhance the quality of scans at no additional cost.
16:10		452-03-006. Feasibility of 64mT-3T contrast transfer for T2-weighted neonatal MRI Alena Uus, Paul Cawley, Kathleen Colford, Francesco Padormo, Rui Pedro Teixeira, Ines Tomazinho, Jana Hutter, Jonathan O'Muircheartaigh, David Edwards, Tomoki Arichi, Jo Hajnal, MARY RUTHERFORD King's College London, London, United Kingdom Impact: This work demonstrates that contrast transfer can enhance interpretability and analysis of 64mT neonatal MRI, supporting bedside assessment in the NICU and expanding access to actionable neuroimaging in clinical settings and low-resource environments where conventional 1.5/3T systems are unavailable.
16:12		452-03-007. A pan-contrast, pan-resolution super-resolution model for brain MRI across the human lifespan Johnathan Leung, Xiaoyang Chen, Chengche Tsai, Khoi Huynh, Kim-Han Thung, Rhea Adams, Sahar Ahmad, Dan Ma, Pew-Thian Yap University of North Carolina at Chapel Hill, Chapel Hill, United States of America Impact: Our model demonstrates a superior quality of super-resolution of brain MRI across contrast types and different resolutions across the lifespan of brain development, broadening access to high-resolution imaging and downstream analysis without additional costs.
16:14		452-03-008. Double Diffusion: A Denoising Diffusion Model for Accurate Single-Acquisition Breast DWI Rodrigo Costa, Yasna Forghani, Nuno Loução, Tiago Marques, Pedro Gouveia, Celeste Alves, Mário Figueiredo, Joao Santinha Digital Surgery Lab, Breast Cancer Research Program, Champalimaud Foundation, Lisboa, Portugal Impact: Our diffusion model enables single-acquisition (NSA/NEX) breast DWI, dramatically cutting scan times and costs. This approach maintains high diagnostic accuracy while reducing image noise, enhancing patient comfort, and accessibility for vital breast cancer diagnosis.
16:16		452-03-009. Deep Learning–Assisted 3D Iterative Enhancement Improves 3D UTE Lung MRI Quality and Pulmonary Nodule Evaluation xiaoqing Wu, Xi Zhu, Jie Shi, Jing Ye, Wennuo Huang, Wei Xia Dalian Medical University, Dalian, China Impact: 3D-DLIE significantly enhances UTE lung MRI quality and diagnostic confidence, supporting its potential as a radiation-free alternative for evaluating pulmonary nodules.
16:18		452-03-010. On super-resolution in 4D flow MRI Luuk Jacobs, Pietro Dirix, Simone Sgorbati, Stefano Buoso, Sebastian Kozerke University and ETH Zürich, Zürich, Switzerland Impact: This work discusses super-resolution for 4D flow MRI using an information theory framework and compares current data-driven and physics-informed super-resolution methods using a k-space-based analysis derived from this framework, indicating a benefit for a combined approach.
16:20		452-03-011. SNRDiff: SNR-Aware Diffusion Networks for Detail-Enhanced MRI Denoising Xinyu Chen, Zhengyong Huang, Ning Jiang, Onur Afacan, Ali Gholipour, Simon Warfield, Yao Sui Peking University, Beijing, China Impact: We develop a lightweight diffusion-based MRI denoiser, optimized for training with limited data, allowing efficient region-adaptive noise reduction while preserving intricate anatomical details. The denoiser strikes an optimal balance between effective noise suppression and structural fidelity, delivering high-quality MRI images.
16:22		452-03-012. Enhancement of Mid-Field (0.6T) T2-Weighted Prostate Scans Using a Two-Stage Refinement Framework Donghang Lyu, Martijn Nagtegaal, Ece Ercan, Zhong Li, Mart WJ van Straten, Marius Staring, Andrew Webb, Matthias van Osch, Peter Börnert, Yiming Dong Leiden University Medical Center, Leiden, Netherlands Impact: By applying the proposed two-stage refinement framework, mid-field prostate MRI with heavy noise and low resolution can be effectively enhanced, achieving image quality comparable to higher-field MRI.
16:24		452-03-013. SIINR: Structurally Informed Implicit Neural Representation for Super-Resolution of Highly Anisotropic Clinical Diffusion MRI Tom Hendriks, Martha Shenton, William Consagra, Anna Vilanova, Maxime Chamberland, Yogesh Rathi Eindhoven University of Technology, Eindhoven, Netherlands Impact: By recovering lost anatomical detail in highly anisotropic (1.5 × 1.5 × 6 mm³) clinical diffusion MRI, SIINR enables fiber tracking and microstructural analyses previously limited to research-quality data, thus bridging the gap between clinical acquisitions and advanced diffusion modeling.
16:26		452-03-014. Improving Image Quality of Knee MRI at 0.05 Tesla with a Conditional Diffusion Model Chaoyi Xing, Vick Lau, Zihao Jin, Xuehong Lin, Ye Ding, Alex T. L. Leong, Yujiao Zhao, Ed X Wu The University of Hong Kong, Hong Kong, China Impact: Our work demonstrates 3D conditional diffusion model can substantially augment 0.05T knee MR image quality. This advancement paves the way for high quality knee imaging using affordable and shielding-free ULF MRI.
16:28		452-03-015. Latent Space Fusion with Conditional Flow Matching for slice to volume Reconstruction Yunzhi Xu, Jiaxin Zheng, Liangchen Shi, Zhenyu Zhang, Ruoge Lin, Li Zhao College of Biomedical Engineering and Instrument Science, Zhejiang University, Hangzhou, China Impact: This approach provides a robust and effective solution for generating high-quality isotropic 3D brain MR images from low-resolution orthogonal acquisitions, with potential to enhance the efficiency of clinical MRI protocols and the reliability of downstream 3D analyses.
16:30		452-03-016. S²-MoCo: A Fully Self-Supervised Subject-Specific k-Space Framework for Physically Consistent Motion Detection and Correction Siyun Jung, Kyu-Jin Jung, Giulia Debiasi, Chunlei Liu, Dong-Hyun Kim Yonsei University, Seoul, Korea, Republic of Impact: This framework establishes a reference-free paradigm for fully self-supervised, subject-specific, and physically consistent MRI reconstruction, enabling reliable recovery under diverse motion patterns, including ACS corruption and real-motion conditions, and advancing robust, individualized motion correction toward practical, patient-specific clinical imaging.
16:32		452-03-017. Continuous Noise-Adaptive Denoising (CoNAD) using a Noise-Conditioned Adversarial Network Omer Demirel, Spencer Waddle, Dinghui Wang, Tzu Cheng Chao, Jacinta Browne, Tim Leiner Philips North America Clinical Science, Rochester, United States of America Impact: This work introduces a controllable MRI denoising framework that allows users to adjust denoising strength based on image noise level. The method improves cardiac MRI image quality by preserving anatomical features while reducing noise adaptively.
16:34		452-03-018. No free lunch: how to address underdetermination in deep-learning-based image super-resolution? Cosimo Campo, Hongxiang Lin, John Ashburner, Christian Lambert, Harith Akram, Gary Zhang University College London, London, United Kingdom Impact: This work can contribute to the development of ultra-low-field MRI and the democratisation of MRI in low- and middle-income countries. While our demonstration used U-Net, our approach can be readily applied to other DL architectures, such as diffusion models.