Cape Town - 2026 ISMRM-ISMRT Annual Meeting and Exhibition - Machine Learning for Image Reconstruction

Machine Learning for Image Reconstruction

Digital Poster

Acquisition & Reconstruction

Monday, 11 May 2026

Digital Posters Row A

09:15 - 10:10

Session Number: 360-02

No CME/CE Credit

Similar Sessions

This session focuses on AI-powered image reconstruction methods.

		360-02-001. MRE-FIn: Open-Source Finite Element Framework for Inversion in Magnetic Resonance Elastography Henrik Palme, Rodrigo Moreno KTH Royal Institute of Technology, Stockholm, Sweden Impact: By creating an open-source framework for non-linear inversion (NLI) in Magnetic Resonance Elastography (MRE) of the brain, the threshold for MRE research can be reduced and the field can expand with standardized methods.
		360-02-002. Optimizing EPI acquisition for super-resolution T2-weighted imaging of the fetus Beata Bachrata, Sarah Peck, Simon Robinson, Guenther Grabner, Thomas Kau Carinthia University of Applied Sciences, Klagenfurt, Austria Impact:* This study optimizes an EPI acquisition protocol for super-resolution reconstruction and shows that EPI can enable motion-robust, high-resolution, high-SNR, T₂^*-weighted imaging of the fetus.
		360-02-003. Principled hyperparameter tuning for diffusion-based MRI solvers Irmak Sivgin, Julio Oscanoa, Cagan Alkan, Daniel Ennis, John Pauly, Mert Pilanci, Shreyas Vasanawala Stanford University, Stanford, United States of America Impact: Our framework eliminates manual hyperparameter tuning in diffusion-based reconstruction by deriving principled, data-driven rules for regularization and noise schedules. This automated, theoretically grounded approach improves reconstruction quality by over 2 dB, advancing the practicality of diffusion priors for inverse problems.
		360-02-004. Deep Learning-Based Reconstruction of Accelerated 3D MP2RAGE on a High-Performance Head-Only System at 3T Brian Burns, Xucheng Zhu, Baolian Yang, Jeff McGovern, Dan Rettmann GE HealthCare, San Ramon, United States of America Impact: This work shows that 10X acceleration of whole-brain 1mm isotropic MP2RAGE is possible on a high-performance head only 3T system with no compromise in IQ compared to conventional parallel imaging techniques.
		360-02-005. Scan-Adaptive Deep Learning-Based Sampling with Pre-Optimized Mask Supervision Aryan Dhar, Siddhant Gautam, Saiprasad Ravishankar Michigan State University, East Lansing, United States of America Impact: We propose a deep learning based approach that learns to predict scan-specific Cartesian MRI undersampling masks directly from low-frequency k-space, trained using pre-optimized scan-adaptive masks as supervision to enable efficient and generalizable accelerated imaging.
		360-02-006. Deep Learning Reconstructed Single Isotropic PD-weighted Imaging of the Knee: Impact on Acquisition Time and Image Quality Oliver Gebler, Vlad Sacalean, Maximilian Löffler, Balazs Bogner, Ralph Strecker, Dominik Paul, Marcel Dominik Nickel, Matthias Jung, Fabian Bamberg, Maximilian Frederik Russe, Jakob Weiß, Thierno Diallo Medical Center - University of Freiburg, Freiburg, Germany Impact: This study shows that DL-accelerated isotropic PD SPACE imaging preserves diagnostic confidence while cutting knee MRI time by two-thirds. These findings may inform more efficient MRI protocols, improving patient throughput and expanding clinical access without compromising diagnostic quality.
		360-02-007. Compressed SENSE AI enhanced mDixon-Quant MR Imaging in the liver Study Linzhe Li, Peng Wu, Jun Peng The Third Xiangya Hospital of Central South University, Changsha, China Impact: This study enables faster, high-quality liver fat quantification using CS AI, improving patient comfort and clinical throughput, while identifying AF=4 as optimal for guiding future protocol optimizations.
		360-02-008. DL-Accelerated Quantitative MT Imaging for Early Osteoarthritis Detection Using a Divide-and-Conquer Framework Yidan Wang, Jiyo Athertya, Jiaji Wang, Donghyun Kim, James Lo, Yajun Ma University of California, San Diego, United States of America Impact: The proposed novel DL method significantly accelerates data acquisition while preserving MMF accuracy, offering strong potential for clinical translation of quantitative MT imaging technique to improve evaluation of knee macromolecular content and early OA diagnosis.
		360-02-009. Improved Neuromelanin Imaging (NMI) at 3T using Deep Learning Reconstruction: Neuroradiological Evaluation for Clinical Use Mohammed Elgwely, Anastasia Papadaki, Karyn Chappell, Karin Shmueli, Iulius Dragonu, Christina Triantafyllou, John Thornton, David Thomas, Tarek Yousry University College London, London, United Kingdom Impact: Improved NMI of the substantia nigra and locus coeruleus can be achieved using TSE with optimised acquisition parameters and deep learning-based reconstruction. This further enhances the feasibility of using NMI for routine clinical assessment of patients with movement disorders.
		360-02-010. Training deep learning based dynamic MR image reconstruction using synthetic fractals Anirudh Raman, Olivier Jaubert, Daniel Knight, Jennifer Steeden, Vivek Muthurangu University College London, London, United Kingdom Impact: We demonstrated that Deep Learning MR reconstruction models can be trained on entirely synthetic fractal-based data. This method can potentially be applied to any type of undersampling to acquire reconstructions without needing application-specific MR training data.
		360-02-011. Application of AI-assisted compressed sensing combined with deep learning-based reconstruction in sacroiliac joint MRI Kangyu Zhang, Zhenghua He, Jian Wang, Guobin Hong, Shilong Lu, Jing Yang Zhujiang Hospital, Southern Medical University, Guangzhou, China Impact: The ACS sequence combined with DLR technology not only shortens the scanning time compared to the PI imaging but also significantly enhances image quality. As the strength of DLR reconstruction increases, the tissue SNR improves more significantly.
		360-02-012. A Unified Approach for Maintaining MRI Reconstruction Quality and Quantifying both Aleatoric and Epistemic Uncertainty Satoshi Kuroki, Naoto Fujita, Tomoki Amemiya, Suguru Yokosawa, Toru Shirai, Yasuhiko Terada The University of Tsukuba, Tsukuba, Japan Impact: Our framework delivers high-quality reconstructions alongside informative uncertainty maps, enhancing the clinical trustworthiness of deep learning MRI. This work enables the development of safer, more robust imaging applications and improves diagnostic confidence.
		360-02-013. An Online Re-parameterization Enhanced Convolutional Recurrent Neural Network for MRI Reconstruction Sijie Zhong, Zhiyong Zhang Shanghai JiaoTong University, Shanghai, China Impact: The online structural re-parameterization enhanced convolutional recurrent neural network trains a multi-branch model for better image quality and fuses them into a single kernel at deployment, eliminating extra cost and improving efficiency, particularly valuable for interventional imaging.
		360-02-014. Deep Learning–Based Phase Correction (DLPC) based Ultra-High B-value Diffusion MRI on Whole-body Scanners Yifei Zhang, Jialu Zhang, Yuhui Xiong, Yang Fan, Lizhi Xie, Sicong Wang, Ruicheng Ba, Patricia Lan, Xinzeng Wang, Xiao-Yuan Fan, Tiebao Meng, Chuanmiao Xie, Feng Feng, Bing Wu GE Healthcare, Beijing, China Impact: DLPC enables the use of UHB-dMRI on whole-body scanner, allowing its potential clinical applications (tumor and stroke characterization; microstructure/cell-size modeling) to be further explored.
		360-02-015. FISTA-Net: Increased Accuracy for Highly Accelerated Scans Using a Proximal Layer and Learned Momentum Nicholas Dwork, Alex McManus, Stephen Becker, Erin Englund, Alex Barker University of Colorado Anschutz, Aurora, United States of America Impact: With high acceleration rates, even the most sophisticated deep learning reconstruction systems introduce errors into the reconstructed images. FISTA-Net, a novel architecture based on FISTA, incorporates a proximal operator and a learned momentum to reconstruct high-quality images from fewer data.
		360-02-016. Use of Deep Learning Acceleration to Improve Throughput on a 0.55T High-Access System in Rural Under-Resourced Settings Devasenathipathy Kandasamy, Yatin sharma, Vikas Gulani, Benjamin Schmitt, Raju Sharma All India Institute of Medical Sciences, New Delhi, India Impact: Deep learning acceleration on a 0.55T system enabled substantial throughput gains without compromising image quality, supporting scalable MRI services in rural, under-resourced regions. This approach strengthens global imaging access and builds viable clinical and economic pathway for high-access MRI deployments.