Cape Town - 2026 ISMRM-ISMRT Annual Meeting and Exhibition

AI Methods

Power Pitch

Acquisition & Reconstruction

Monday, 11 May 2026

Power Pitch Theatre 1

16:10 - 17:46

Moderators: Ilias Giannakopoulos & Mariya Doneva

Session Number: 351-03

No CME/CE Credit

Similar Sessions

This session gives an overview of the use of AI methods in MR acquisition, calibration and analysis - anything but reconstruction!

Skill Level: Advanced

16:10		351-03-001. Auto-calibrating delay correction for radial MRI Sebastian Flassbeck, Ivo Maatman, Jakob Assländer NYU Grossman School of Medicine, New York, United States of America Impact: We introduce a robust auto-calibrating method to estimate and correct gradient delays in 2D and 3D radial MRI, improving image quality and mitigating artifacts.
16:12		351-03-002. Deep Learning Enhanced Weighted Average CSI for High-Resolution Deuterium Metabolic Imaging Hongkang Chu, Gang Chen, Xinjie Liu, Hong Shi, Chaoyang Liu, Maili Liu, Xin Zhou, Elton Montrazi, Lucio Frydman, Qingjia Bao Innovation Academy for Precision Measurement Science and Technology,CAS, Wuhan, China Impact: This work overcomes the SNR–resolution trade-off in deuterium metabolic imaging by combining weighted-average acquisition with deep learning fusion of anatomical and spectral priors, producing sharper, higher-SNR metabolite maps and reliable tumor dynamics for advanced metabolic imaging.
16:14		351-03-003. PINN-Based Electrical properties Tomography Using Tensor Diffusion Regularization Chunyou Ye, Yaqing Jia, Yikun Hong, Yunyu Gao, Yanming Wang, BenSheng Qiu, Xiang Nan, Jijun Han School of Biomedical Engineering, Anhui Medical University, Hefei, China Impact: This study enhances the stability and adaptability of conductivity reconstruction through a physics-informed tensor diffusion framework, showing promise for practical EPT implementation.
16:16		351-03-004. Reference-free and spatial-resolved image quality assessment in brain MRI using contrastive representation learning Veronika Ecker, Elisa Marchetto, Hannah Eichhorn, Melanie Ganz, Sergios Gatidis, Bin Yang, Thomas Küstner University Hospital of Tuebingen, Tuebingen, Germany Impact: The proposed reference-free method allows for automated detection of motion artifacts under realistic clinical conditions in comparison to traditional quality metrics, thus enabling more consistent, quality-controlled MRI acquisition.
16:18		351-03-005. Learning Beyond Interpolation: Zero-shot Resolution Enhancement for Low-Field MRI Ajay Sharma, Sairam Geethanath Johns Hopkins University School of Medicine, Baltimore, United States of America Impact: Zero-shot self-supervised learning can improve edge strength while simultaneously reducing the risk of hallucinations. The ZSSR method outperforms interpolation in improving resolution by utilizing a small, image-specific CNN.
16:20		351-03-006. Accelerated 3D dual-echo MRI with Cross-Attention-Guided Joint Optimization of Sampling and Reconstruction Aiqi Sun, Cagan Alkan, Marcus Alley, Yan Wu, Ali Syed, John Pauly, Daniel Ennis, Shreyas Vasanawala Stanford University, Stanford, United States of America Impact: This work demonstrates that incorporating an inter-echo information exchange mechanism into joint sampling and reconstruction optimization for accelerated 3D dual-echo MRI improves both training efficiency and reconstruction accuracy, providing a promising direction for fast and high-quality multi-echo MRI applications.
16:22		351-03-007. Learnable SENSE MRI Inversion Operator with Embedded Image Priors Junzhou Chen, Anthony Christodoulou, Zhaoyang Fan, Debiao Li Cedars-Sinai Medical Center, Los Angeles, United States of America Impact: This novel MRI reconstruction algorithm enables greatly accelerated MR reconstruction. The method opens new research avenues in fast dynamic imaging and DL-based reconstruction algorithms.
16:24		351-03-008. Hybrid Deep Denoising for In Vivo Dynamic DMI Hauke Fischer, Fabian Niess, Anna Duguid, Stanislav Motyka, Bernhard Strasser, Aaron Osburg, Viola Bader, Lukas Hingerl, Sabina Frese, Wolfgang Bogner Medical University of Vienna, Vienna, Austria Impact: The proposed denoising approach mitigates low-SNR limitations in DMI, enabling more reliable quantification of metabolites at high spatiotemporal resolution. This facilitates visualization of altered tumor metabolism, which may benefit future studies on metabolic heterogeneity and treatment response.
16:26		351-03-009. Deep learning-enhanced biparametric prostate MRI for optimized clinical workflows José de Arcos, María de la Luz Jurado Gómez, Patricia Lan, Xinzeng Wang, Michael Carl, Dan Rettmann, Ajeetkumar Gaddipati, Pablo Garcia-Polo, Arnaud Guidon, Polina Rudenko, Claudia Fontenla Martínez, Asunción Torregrosa, Luis Martí-Bonmatí Impact: Novel deep learning approaches enable rapid, high-resolution prostate MRI, reducing scan time by up to 70% without compromising diagnostic confidence. This approach may transform clinical workflows, improve patient comfort and increase clinical throughput.
16:28		351-03-010. K-space parallel imaging reconstruction using complex-valued deep Koopman autoencoders Wassim Ben Salah, Jon Cleary, Sarah McElroy, Antoine Naegel, Kasim Ali Mohamed, Saranya BALESWARAN, Sebastien Ourselin, Jonathan Shapey, Christos Bergeles, Radhouene Neji King's College London, London, United Kingdom Impact: This work introduces an interpretable neural network for k-space interpolation, enabling good reconstruction quality.
16:30		351-03-011. Reliability of Liver-Fat-Quantification in Deep Learning-Accelerated Image Reconstructions of VIBE Dixon Sequences Anna Fink, Maximilian Frederik Russe, Ralph Strecker, Marcel Dominik Nickel, Lea Jigme Michel, Vlad Sacalean, Kai Falko Kästingschäfer, David Klemm, Alexander Rau, Fabian Bamberg, Jakob Weiß, Stephan Rau University Medical Center Freiburg, Freiburg, Germany Impact: Shortened breath-holds with deep learning-accelerated Dixon MRI improve feasibility in patients with limited compliance, thereby expanding the usability of liver-fat quantification for screening and longitudinal monitoring of non-alcoholic fatty liver disease.
16:32		351-03-012. Automated Sequence Design using Neural Architecture Search: In-depth Exploration on Simulation Fidelity Rokgi Hong, Hongjun An, Sooyeon Ji, Taechang Kim, Jongho Lee Seoul National University, Seoul, Korea, Republic of Impact: Our previous work, Sequence Search, has been further explored with a more realistic simulation environment. Our method successfully designed robust pulse sequences, paving the path toward future high-fidelity MR design that potentially surpasses human intuition.
16:34		351-03-013. Spatially-Aware Neural Controlled Differential Equations for IVIM MRI Parameter Estimation in Esophageal Cancer Patients Daan Kuppens, Roman Oort, Stella Mook, Gert Meijer, Oliver Gurney-Champion Amsterdam University Medical Center, University of Amsterdam, Amsterdam, Netherlands Impact: Spatially-aware NCDEs enable accurate, noise-robust IVIM MRI parameter estimation across acquisition protocols without retraining, potentially improving tumor assessment and therapy response monitoring. This advancement allows exploration of spatial context in quantitative imaging, previously limited by voxelwise or acquisition-specific methods.
16:36		351-03-014. Physics-guided self-supervised deep learning of non-selective scalable 7T RF pulses with low-rank subject-specific adaptation Albert Jang, Manoe Meunier, Xingxin He, Fang Liu, Bastien Guerin Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital and Harvard Medical School, Charlestown, United States of America Impact: Our physics-guided, self-supervised deep learning framework with online adaptation (GPS) enables rapid subject-specific RF pulse design that adjusts for subject-specific B₀/B₁⁺ inhomogeneity, improving flip-angle uniformity and image contrast at ultra-high-field MRI.
16:38		351-03-015. Anatomically Guided Source Localization of Uterine Peristalsis Using T1-Weighted MRI and Electrohysterography Maria Bustos-Vivas, Smiti Tripathy, Nyvenn da Mota Alves de Castro , Milauni Desai, Michael Uder, Jana Hutter University Hospital Erlangen, Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU), Erlangen, Germany Impact: Combining electrohysterography and MRI data symbiotically allows tracing the source of uterine peristaltic signals. Spatial localization enables the disentanglement from other signal sources and thus contributes to a better understanding of uterine physiology.
16:40		351-03-016. Automated Susceptibility-Informed Gold Seed Detection for MR-Guided Prostate Radiation Therapy using Routine mDixon Scans Siddhant Sahay, Viswanath Pamulakanty Sudarshan, Gipson Anto, Suthambhara Nagaraj Philips Healthcare, Bengaluru, India Impact: Automated gold fiducials localization in MR-guided radiation therapy (MRRT) reduces artifacts and improves accuracy in estimated pseudo-CT maps. Accurate pseudo CT maps with localization of fiducials can potentially enhance precise dose planning and delivery, and encourage wider adoption of MR-RT.
16:42		351-03-017. PINN-Based Shear Modulus Estimation for Brain MR Elastography with Forward Wave Decomposition Niloufar Seyfi, Xi Peng University of Iowa, Iowa City, United States of America Impact: Accurate shear modulus estimation is critical for MRE applications in both clinical and research settings. This work presents a novel physics-informed network approach that enables reliable stiffness quantification from non-ideal measurements affected by noise and nuisance wave fields.
16:44		351-03-018. A Fraction of the Cost: Portable, Low-Cost and Energy-Efficient AI-Driven MRI analysis on Raspberry Pi and NVIDIA Jetson Nano Minh Nhat Trinh, Teresa Correia Quantitative Bio-Imaging Lab, CCMAR, Faro, Portugal, Portugal Impact: Our proposed lightweight AI-driven MRI analysis has great performance on portable, low-cost, energy-efficient devices, enabling non-specialist clinicians and healthcare providers in under-resourced settings to generate accurate quantitative cardiac reports, expanding global diagnostic access, and ultimately improving patient outcomes.