Cape Town - 2026 ISMRM-ISMRT Annual Meeting and Exhibition - Data-Driven Reconstruction: From Networks to Clinical Practice

Data-Driven Reconstruction: From Networks to Clinical Practice

Digital Poster

Acquisition & Reconstruction

Tuesday, 12 May 2026

Digital Posters Row D

08:20 - 09:15

Session Number: 463-01

No CME/CE Credit

Similar Sessions

This session focuses on AI-powered image reconstruction methods.

		463-01-001. Motion-Corrected Deep-Learning Reconstruction Framework for 3D Whole Heart Joint T1/T2 mapping at 0.55T Matias Paredes, Dongyue Si, Andrew Phair, Rene Botnar, Claudia Prieto Millennium Institute for Intelligent Healthcare Engineering - iHEALTH, Santiago, Chile Impact: This investigation shows that it is possible to incorporate motion estimation and multi-contrast dual-echo reconstruction into a deep learning framework, enabling 3D whole-heart joint T1/T2 mapping within 30 seconds, which significantly reduces reconstruction time and facilitates potential clinical adoption.
		463-01-002. MoE-Unet: Using Conditional Sparse Activation to Improve the Capacity of Multi-task End-to-End Reconstruction Yuyang Li, Yipin Deng, Wenlei Shang, Zijian Zhou, Peng Hu School of Biomedical Engineering, ShanghaiTech University, Shanghai, China Impact: Data-driven AI reconstruction methods are often sensitive to dataset heterogeneity and distribution shifts between training and test sets. Mixture-of-Experts (MoE) models offer promising solutions to these multi-domain learning challenges.
		463-01-003. Comparative Evaluation of Retrospective Undersampling Strategies for Active Sampling in low-field 3D Cartesian MRI Alexis Cogne, Guillaume Daval-Frérot, Zineb Belkacemi, Nina Gidel-Dissler, Romain Couvreur, Dimitri Labat Chipiron, Paris, France Impact: There is no clear evidence that customizing Cartesian undersampling patterns for individual subjects over the traditional patient-blind approach brings any substantial improvements in image quality. Active sampling may still be relevant using non-Cartesian patterns or more advanced reconstruction methods.
		463-01-004. Assessment of Age-Related PVS Enlargement Using DLS-Accelerated Cranial 3D T1WI Imaging yanhong Zhao, Guangxu Han, xiaowen Zhang, Zhu Rongrong, Dan Rettmann Medical Imaging Center, People's Hospital of Ningxia Hui Autonomous Region, Yinchuan, China Impact: This study validates the reliability of DLS-accelerated 3D T1WI for PVS quantification, offering a practical reference for clinicians and researchers seeking efficient, high-quality neuroimaging in aging and cerebrovascular studies.
		463-01-005. Mamba-MRF:A Deep Mamba Network for Highly Accelerated Magnetic Resonance Fingerprinting Reconstruction Tianyi Ding, Hongli Chen, Yang Gao, Peng Wu, Zhuang Xiong, Shanshan Shan, Feng Liu, Martijn Cloos, Hongfu Sun The University of Queensland, Brisbane, Australia Impact: This study demonstrates that modern deep learning architectures, particularly state-space–based models, can substantially improve the accuracy and efficiency of quantitative MRF reconstruction, highlighting their potential to accelerate and enhance future clinical qMRI applications.
		463-01-006. 3D T2-Weighted CUBE with AIR Recon DL: Superior Facial Nerve Visualization Compared to 3D-FIESTA Taiki Koshiishi, Yuka Ishimoto, Satoru Ide, Keita Watanabe, Kazuhiko Oyu, Tomohiro Shintaku, Sera Kasai, Miho Sasaki, Jusei Kudo, Kana Saito, Amo Ozawa, Mizuki Imura, Atsushi Nozaki, Tetsuya Wakayama, Shingo Kakeda Hirosaki University School of Medicine, Hirosaki, Japan Impact: CUBE with ARDL enables superior facial nerve visualization compared to current gold standard 3D-FIESTA, potentially improving preoperative identification of neurovascular compression sites in hemifacial spasm patients. This may lead to establishing a new imaging standard for neurovascular compression syndromes.
		463-01-007. Highly accelerated 3D water/fat LGE imaging with deep-learning motion estimation and motion corrected reconstruction Dongyue Si, Andrew Phair, Alina Hua, Simon Littlewood, Michael Crabb, Haikun Qi, Tevfik Ismail, Claudia Prieto, Rene Botnar King's College London, London, United Kingdom Impact: A MoCo-MoDL reconstruction framework is proposed to accelerate 3D isotropic LGE imaging, allowing 7-fold undersampling and ~3+1 min acquisition-reconstruction time and good image quality and scar assessment, which is promising to promote the application of 3D LGE in clinics.
		463-01-008. Quantitative assessment of a temporal enhancement deep learning algorithm for accelerated cardiac cine MRI Tzu Cheng Chao, Enas Ahmed, Spencer Waddle, Dinghui Wang, Jacinta Browne, Tim Leiner Mayo Clinic, Rochester, United States of America Impact: A deep learning based temporal enhancement can improve scan efficiency. This study quantitatively evaluates its impact on the accuracy of the functional parameters derived from low temporal resolution scans.
		463-01-009. Learning Patient-Adaptive Undersampling Patterns for Cardiac MRI Using Nearest Neighbor Search Siddhant Gautam, Angqi Li, Jeffrey Fessler, Nicole Seiberlich, Saiprasad Ravishankar Michigan State Universtiy, East Lansing, United States of America Impact: Our approach learns patient-adaptive Cartesian sampling patterns that improve reconstruction quality and reduce acquisition time, enabling faster, personalized cardiac MRI and potentially lowering motion artifacts and patient discomfort.
		463-01-010. Artificial Intelligence methods for low field MRI enhancement comparing brain volumes and neurodevelopment scores for small v Shayan Khakwani, Ishrat Fatima, Levente Baljer, František Váša, Kiran Hilal, Sidra Kaleem, Rosalyn Moran, Kirsten Donald, Zahra Hoodbhoy Aga Khan University Hospital, Pakistan, Pakistan Impact: This study reports high correlation between super resolved ultra-low field and high field (3T) brain volumes in children aged 3-6 years. These findings validate the use of Hyperfine devices for neuroimaging in low- and middle-income countries.
		463-01-011. Hybrid 2D CNN-Transformer Achieves Fast and Robust Ktrans Mapping of FUS-Induced BBB Opening Akito Yamauchi, Zongyu Li, Zhonghui Qie, Jia Guo Columbia University, New York, United States of America Impact: We introduce 2D ConvFormer, a state-of-the-art model for K^transmapping of FUS-induced BBB opening achieving a >50x acceleration over the Tofts model. It enables robust, rapid quantification from both full and low-dose DCE-MRI, demonstrating superior resilience to noise and motion.
		463-01-012. Deep Learning and Compressed Sensing for Fast Sampling: A Comparative Study in Rabbit Knee MRI Yang Qu, Xiance Zhao, Yuehua Li Shanghai Sixth People’s Hospital Affiliated to Shanghai Jiao Tong University School of Medicine, Shanghai, China Impact: Deep learning (DL) outperforms compressed sensing (CS) in rabbit knee MRI scanning and reconstruction, especially at higher acceleration factors, offering a way to balance imaging speed and quality for preclinical musculoskeletal studies.
		463-01-013. 3-Dimensional segmentation and radiomic feature detection of lymphedema treatment changes based on MR imaging Charissa Kim, Qianhui Dou, Christopher Bridge, Nathaniel Mercado, Mahmoud Odeh, Madeleine Givant, Katja De Paepe, Angie Sohn, Dhruv Singhal, Leo Tsai Beth Israel Deaconess Medical Center, Boston, United States of America Impact: In this study, we developed a deep-learning based 3-dimensional segmentation method and extracted radiomic features from upper extremity lymphedema at different surgical treatment time points. These features can potentially be used to predict surgical treatment responses and optimize lymphedema treatments.
		463-01-014. Uncertainty-Aware Cross-Modal MRI Reconstruction via Evidential Beta-Gated Attention Bingbing Chen, Congcong Liu, Yihang Zhou, Zhuoxu Cui, Dong Liang School of Biomedical Engineering, ShanghaiTech University, Shanghai, China Impact: EBGA provides a principled, uncertainty-aware attention fusion strategy for text-guided MRI reconstruction, improving reliability, controllability, and interpretability of cross-modal reconstructions in undersampled settings.
		463-01-015. Reducing Breath-Hold Time in Liver MRI: Clinical Performance of Deep Learning-Accelerated Post-Contrast T1 Dixon VIBE Anna Fink, Maximilian Frederik Russe, Vlad Sacalean, Kai Falko Kästingschäfer, Ralph Strecker, Marcel Dominik Nickel, Alexander Rau, Fabian Bamberg, Jakob Weiß, Stephan Rau University Medical Center Freiburg, Freiburg, Germany Impact: Deep learning-accelerated post-contrast Dixon MRI enables diagnostic liver imaging in patients with limited breath-hold capacity. Ultrafast acquisitions may reduce motion artifacts and allow multiphase dynamic imaging within a single breath-hold, improving temporal resolution and diagnostic confidence in hepatic lesion characterization.