Cape Town - 2026 ISMRM-ISMRT Annual Meeting and Exhibition - Generative Models for Image Synthesis

Generative Models for Image Synthesis

Digital Poster

Analysis Methods

Wednesday, 13 May 2026

Digital Posters Row B

08:20 - 09:15

Session Number: 561-01

No CME/CE Credit

Similar Sessions

This session will focus on abstracts using generative tools for the task of image synthesis within and across modalities.

		561-01-001. Rapid High-Fidelity Abdominal sCT Generation for MR-Only Radiotherapy on an MR-Linac with a 2.5D Hybrid VAE-GAN Sukhraj Virdee, Anna Briskina, Amir Moslemi, Angus Lau, Brige Chugh Sunnybrook Health Sciences Centre, University of Toronto, Toronto, Canada Impact: This work validates a fast, high-fidelity synthetic CT generation method for MR-Linac systems, paving the way for efficient MR-only workflows that eliminate resource-intensive CT simulation, enabling accurate dose calculations and robust adaptive radiotherapy in the challenging abdominal region.
		561-01-002. Automated FLAIR Synthesis from T1 and T2 Brain MRI at 3T Yaozhong Huang, Patricia Johnson, Narges Razavian, Riccardo Lattanzi New York University Grossman School of Medicine, New York, United States of America Impact: Eliminating the FLAIR acquisition through synthesis from T1 and T2 sequences shortens brain MRI protocols by ~5 minutes (20% reduction). This decreases motion artifacts, particularly important for pediatric and cognitively impaired patients, while improving patient comfort and increasing scanner capacity
		561-01-003. OurGAN: A Deep Learning Approach for Synthesizing High-Quality T2 FLAIR Images from T2 Mapping Data Xueao Li, Wenqian Zhao, Che Wang, Yao Liu, Xiaonan Zhang, Chunyan Zhang, Yuchuan Zhuang, Andrey Tulupov, Jing Li, Liufei Yang, Tong Wang, Yongping Wu, Rongning Zhuang, Fengshou Zhang, Jianfeng Bao Henan University of Science and Technology, Luoyang, China Impact: In this study, we propose a deep learning–based approach for synthesizing T2FLAIR images directly from T2mapping, representing a pioneering attempt in this field. This method provides a feasible pathway to obtain diagnostically valuable FLAIR images while substantially reducing acquisition time.
		561-01-004. 3D Conditional VAE with ViT-UNETR for Multi-Sequence Brain MRI Generation Tayyaba Arshad, Endre Grovik Helse Møre and Romsdal Hospital Trust, Alesund, Norway Impact: The proposed 3D conditional VAE offers a scalable approach for generating realistic multi-sequence MRIs, expanding data availability for AI research. It enables the creation of synthetic neuroimaging datasets to enhance model generalization and facilitate reproducible, data-driven brain analysis.
		561-01-005. Accelerating Prostate Microstructure Mapping: A CycleGAN-Based Framework for HM-MRI Synthesis Dianning He, Guantian Huang, Aritrick Chatterjee, Abel Campos, Wei Qian, Shouliang Qi, Gregory Karczmar, Aytekin Oto School of Health Management, China Medical University, China Impact: This study enables clinically feasible implementation of prostate microstructure imaging using Hybrid Multidimensional MRI by reducing scan time by over half while preserving diagnostic fidelity, facilitating routine deployment in prostate cancer assessment.
		561-01-006. Deep learning-based synthetic CT from black-bone MRI with limited retrospective clinical data for MR-only treatment planning Merlin Owens, N. Jane Taylor, Andrew King School of Biomedical Engineering and Imaging Sciences, King's College London, London, United Kingdom Impact: Synthetic CTs can reduce radiation dose to patients, time spent in hospital, and imaging workload. Demonstrating a clinically available black-bone MRI sequence as an input to sCT models, and performing comparative dose calculations, expands feasibility of MR-only treatment planning.
		561-01-007. FiLM-cGAN based synthesis of individualized cerebral blood flow (CBF) maps from T1-weighted MRI Wiebke Entelmann, Thomas Lindner, Matthias Günther University Hospital Hamburg Eppendorf, Germany Impact: This study demontrated the feasibility of generating sythetic cerebral blood flow (CBF) maps from structural T1-weighted MRI and metadata using a FiLM-cGAN architecture.
		561-01-008. PSF-EPI-DWI to Multi-Contrast MRI Translation using Deep Learning Yuhang He, Juanhua Zhang, Zhe Zhang, Yuan Lian, Yibei Yu, Wen Zhong, Xiaoguang Yang, Jing Jing, Zhuozhao Zheng, Hao Huang, Hua Guo Tsinghua University, Beijing, China Impact: This AI-powered method synthesizes high-fidelity T2w, T2-FLAIR and T1-FLAIR from a single PSF-EPI DWI scan. It paves the way for radical MRI acceleration, potentially cutting protocol times while preserving full diagnostic content.
		561-01-009. Unsupervised Image Harmonization of Multi-Parametric Maps of the Brain Ann Laube, Christian Stehning, Joachim Weber, Matthias Endres, Katharina Schönrath, Ira Rohrpasser-Napierkowski, Tobias Leutritz, Nikolaus Weiskopf, Jeanette Schulz-Menger, Kersten Villringer, Anja Hennemuth Deutsches Herzzentrum der Charité (DHZC), Berlin, Germany Impact: We present an unsupervised deep learning method that harmonizes quantitative multi-parametric mapping across clinical scanners without needing matched travelling-subject data, enabling more consistent measurement of brain microstructure and improving the reliability of multi-center neuroimaging studies.
		561-01-010. Synthetic STIR Sequences Offer Same Edema Contrast as Standard-of-Care Acquisitions Thomas Arnold, Long Wang, Ajit Shankaranarayanan, Lawrence Tanenbaum Subtle Medical Inc, Menlo Park, United States of America Impact: This work demonstrates that deep-learning-based synthesized STIR sequences provide equivalent edema contrast to standard, slow acquisitions. This allows for significantly accelerated spine MRI protocols and can increase patient throughput without compromising the detection of critical spinal pathology like edema.
		561-01-011. Integrating DeepOxyMap and Residual GAN–Based Mapping for Differentiating Ischemic,Non-Ischemic, and Amyloid Cardiomyopathies faezeh Lotfikazemi, Matthias Fridreich, mitchel benovoy, moezedinjavad rafiee, leila haririsanati, michael chetrit McGill University Health Centre, Montreal, Canada Impact: Our DeepOxyMap + R-GAN framework generates contrast-free T1 maps from OS-CMR, enabling accurate differentiation of amyloid, ischemic, and non-ischemic cardiomyopathies—overcoming limitations of motion, scan time, and contrast injection with fast, one-minute fibrosis mapping.
		561-01-012. Lung Template-Based Quantitative Assessment of Pulmonary Ventilation Function Yuan Fang, Haidong Li, Hongchuang Li, Ming Zhang, Ming Luo, Xiuchao Zhao, Lei Shi, Yeqing Han, Xin Zhou State Key Laboratory of Magnetic Resonance Spectroscopy and Imaging, National Center for Magnetic Resonance in Wuhan, Wuhan Institute of Physics and Mathematics, Innovation Academy for Precision Measurement Science and Technology, Chinese Academy of Sciences, Wuhan, China Impact: This study developed and established a healthy lung template, enabling sensitive, segmental-level ventilation quantification without thoracic CT. This standardized tool facilitates detection of early, localized dysfunction and guides personalized treatment planning.
		561-01-013. An Optimized Synthetic MRI Framework for Enhanced Tumor-to-Background Contrast Mavidu Iddagoda, Kathleen Earhart, Evan Noch, Salvador Pena, Janaka Wansapura University of Colombo, Colombo, Sri Lanka Impact: The proposed framework facilitates enhanced tumor detection by significantly increasing the contrast between tumor and surrounding tissue. Furthermore, the synthetic MRI methodology is designed to be generalisable, demonstrating flexibility for application across diverse anatomical regions and tissue types
		561-01-014. Accurate physics-guided synthesis and harmonisation of T1-Weighted gray–white matter ratio Hongyan Liu, Somtochukwu Ibeme, Nikos Priovoulos, William Clarke, Aaron Hess, Karla Miller University of Oxford, Oxford, United Kingdom Impact: This physics-guided synthesis framework enables accurate prediction and harmonisation of sequence-dependent Gray-white matter ratio variability, improving cross-protocol comparability of T₁-weighted MRI biomarkers. It provides a quantitative and physically interpretable tool for protocol optimisation and data harmonisation.
		561-01-015. Super-resolution synthetic MRI using deep learning reconstruction for accurate diagnosis of lumbar intervertebral disc degene Fei-fei Zeng, Yunfei Zha, Weiyin Vivian Liu, Yang Fan Renmin Hospital of Wuhan University, Wuhan, China Impact: DLR SyMRI significantly reduces scan time while providing morphological image quality comparable to conventional MRI and more precise quantitative T2 value. It offers an efficient, one-stop solution for the precise identification and quantitative assessment of early-stage lumbar disc degeneration.
		561-01-016. Neuro GPT \| A tool for scientific discovery Niall Bourke, Hajer Karoui, Kenneth Ae-Ngibise, Firehiwot Abate, Kwaku Asante, Florence AWEYO, Victor Akelo, Muriel Bruchhage, Chiara Casella, Vanessa Cavallera, Kirsten Donald, Tarun Dua, Laurel Gabard-Durnam, Bethany Freeman, Emmanuela Gakidou, Zahra Hoodbhoy, Margaret Kasaro, Sidra Kaleem, AMNA KHAN, Patricia Kitsao-Wekulo, Beena Koshy, Anne Lee, Aksel Leknes, Natasha Lepore, Marius Linguraru, Russell Macleod, Yaw Mensah, Jonathan O'Muircheartaigh, Hans-Georg Müller, Victoria Namazzi, Margaret Nampijja, Gloria Nandudu, Victoria Nankabirwa, Krysten North, Solomon Nyame, Dickens Onyango, Samuel Oppong, Salman Osmani, Harun Owuour, Sadia Parkar, Marc Seal, Emily Smith, Jamie Steinmetz, Austin Tapp, Jeffrey Tanedo, Maclean Vokhiwa, Ayo Zahra, Steven Williams, Sean Deoni, Joshua Proctor King's College London, London, United Kingdom Impact: Sites in a global network have collected paediatric ultra-low field MR images, generating derived volume estimates. This new AI-assisted tool will facilitate answering of locally relevant clinical questions on factors affecting neurodevelopment, such as maternal anaemia, HIV exposure, malnutrition etc.