Cape Town - 2026 ISMRM-ISMRT Annual Meeting and Exhibition

Image Reconstruction II

Digital Poster

Acquisition & Reconstruction

Wednesday, 13 May 2026

Digital Posters Row I

16:55 - 17:50

Session Number: 568-06

No CME/CE Credit

Similar Sessions

Poster section about image reconstruction, mostly non-AI methods.

Skill Level: Advanced

		568-06-001. B0 Map Correction using Autofocusing for Single-Shot Spiral MRI Mayuri Sothynathan, Anthony Chu, Corey Baron Robarts Research Institute - Western University, Canada Impact: We introduce a B₀ map correction algorithm that estimates B₀ offsets to enhance spiral image reconstruction. This may be valuable to reduce image blurring in spiral and other non-Cartesian MRI acquisitions.
		568-06-002. Image Derived Input Function Estimation using PET Image Reconstruction with MR Priors Mehdi Khalighi, Farshad Moradi, Greg Zaharchuk, Fernando Boada Stanford University, Stanford, United States of America Impact: Estimating AIF without the need for additional scans during PET/MR exams enables accurate quantification of PET images, which may lead to improved diagnostic capabilities and a better assessment of physiological processes and disease states.
		568-06-003. Reference-Superimposed Reconstruction (RS-Recon) for Arterial Spin Labeling Imaging Jia Guo University of California Riverside, Riverside, United States of America Impact: The RS-Recon method provides a robust ASL signal detection even when the MR signal is extremely weak. This allows strong background suppression to be applied in ASL for improved SNR. It may be useful in many other MRI reconstruction applications.
		568-06-004. Self-Supervised Physics-Guided Reconstruction for 3D Automatic Landmarking Hanrui Shi, Xin Tang, Hongyi Gu, Feng Fang, Jian Xu, Qi Liu, Hongyu Li University of Washington, Seattle, United States of America Impact: This work enables rapid 3D Automatic Landmarking MRI for anatomical localization through self-supervised physics-guided learning. The framework achieves over 10× acceleration and reconstructs a full 3D volume within 5 seconds, offering a practical approach for streamlined clinical MRI positioning.
		568-06-005. SNR-Guided Compressed Sensing Reconstruction for Ultra-High Field Non-Cartesian MRI Hongyi Gu, Hongyu Li, Zhibo Zhu, Jue Hou, Zheng Zhong, zihao zhu, Qi Liu United Imaging Healthcare North America, Houston, United States of America Impact: Proposed SNR-guided non-Cartesian Compressed Sensing reconstruction enhances image quality in ultra-high-field MRI by integrating spatial SNR information into regularization. Improved reconstruction quality on 5T data gives potential for better diagnostic confidence and inspires further investigations on adaptive reconstruction at ultra-high-fields.
		568-06-006. GROG-facilitated Diffusion Model Reconstruction with 3D Data Consistency for Accelerated Non-Cartesian MRI Zihao Chen, Guangyu Dan, Vips Patel, Qi Liu, Hongyu Li United Imaging Healthcare North America, Houston, United States of America Impact: This work enables accelerated 3D non-Cartesian MRI by integrating self-calibrating GROG, memory-efficient chunk-wise diffusion inference, and a unified 3D data-consistency, leading to improved image quality under high acceleration while substantially reducing GPU memory usage.
		568-06-007. An Alternative Framework for Sparsity-Promoting Dynamic MRI Reconstruction with Reduced Hyper-Parameter Sensitivity Aidan Mason-Mackay, Daniela Calvetti, Erkki Somersalo, Mikko Kettunen, Antti Aarnio, Ekaterina Paasonen, Olli Gröhn, Ville Kolehmainen University of Eastern Finland, Kuopio, Finland Impact: High-temporal resolution MRI often uses k-space undersampling and requires advanced reconstruction methods with Compressed Sensing (CS) or AI. We demonstrate cases where the Iterative Alternating Sequential Algorithm (IAS) has reduced dependence on expert parameter tuning compared to CS.
		568-06-008. Power spectral density in ESPIRiT for faster coil sensitivity estimations in MRI Ilseok Lee, Jörn Huber, Laurent Ruck, Armin NAGEL, Simon Konstandin Fraunhofer Institute for Digital Medicine MEVIS, Bremen, Germany Impact: Various power spectral density estimation methods can be applied to coil sensitivity estimation, resulting in different sensitivity estimation methods. Overall, these approaches can reduce reconstruction times for larger MRI images, enabling more detailed imaging in a shorter period.
		568-06-009. Wavelet-Regularized Subspace Reconstruction for Highly Accelerated Multi-Echo ASL MRI Ershad Hassanpour Golagani, Yiran Li, Bo Li, Xiao Liang, Yulin Chang, Manuel Taso, John Detre, Min Wu, Ze Wang University of maryland Baltimore, Baltimore, United States of America Impact: This approach supports high-quality multi-echo ASL under strong acceleration, facilitating reliable quantification of BBB transport and other multi-parametric ASL biomarkers without increasing acquisition time, enabling broader clinical feasibility of advanced ASL protocols.
		568-06-010. Comparison of SENSE and GRAPPA Reconstruction on GE-SE EPIK Data while incorporating different sensitivity map computations Fabian Küppers, Omar Omari, Felix Landmeyer, Markus Zimmermann, N. Jon Shah Forschungszentrum Juelich, Juelich, Germany Impact: Incorporating GRAPPA and SENSE reconstructions with coil sensitivity maps from ESPIRiT (BART and JuART) shows comparable performance for Cartesian acquisition. JuART demonstrates practical advantages through OS-independence and faster computation, supporting efficient sensitivity map estimation in research and clinical applications.
		568-06-011. A mixed precision FFT with applications in MRI Nikhil Deveshwar, Abhejit Rajagopal, Peder Larson University of California San Francisco, San Francisco, United States of America Impact: This work enables memory-efficient MRI reconstruction on portable/edge devices while maintaining image quality. MX-scaled FP8 FFT could accelerate iterative reconstruction algorithms, reduce scanner costs, and improve MRI accessibility in resource-limited settings, with future potential in real-time low-precision imaging pipelines.
		568-06-012. Motion Correction in Dental-dedicated MRI Zihan Ning, Yannick Brackenier, Rubens Spin-Neto, Saoirse O'Toole, Kathleen Colford, Sarah McElroy, Francesco Padormo, Lucilio Cordero-Grande, Jonathon Davies, Owen Addison, Jo Hajnal King's College London, London, United Kingdom Impact: This study presents a region-specific motion correction method for dental-dedicated MRI (ddMRI), enabling independent motion-correction for upper- and lower-jaw. It significantly enhances image quality in paediatric scans and demonstrates potential for more robust ddMRI in clinical settings.
		568-06-013. Accelerated ML-DIP: Where ML-DIP Meets DeepSpeed Accelerator Jianli Wei, Orlando Simonetti, Rizwan Ahmad, Samuel Ting The Ohio State University, Columbus, United States of America Impact: Accelerated ML-DIP reduces 3D cardiac cine MRI reconstruction time from ~8.5 hours to ~1 hour by leveraging DeepSpeed-based parallel training across 16 GPUs on the Ohio Supercomputing Center (OSC) cloud platform, enabling efficient large-scale model training with preserved image quality.
		568-06-014. Gaussian representation-based dynamic reconstruction and motion estimation framework for time-resolved volumetric MR imaging Jiacheng Xie, Hua-Chieh Shao, Can Wu, Ricardo Otazo, Mu-Han Lin, Jie Deng, You Zhang University of Texas Southwestern Medical Center, Dallas, United States of America Impact: Leveraging the strong representation power of Gaussians, DREME-GSMR enables ‘one-shot’ dynamic MRI reconstruction and subsequent real-time MRI inference from limited k-space data, eliminating the need for prior anatomical or motion models, enhancing the applicability of dynamic/real-time MRI towards motion-adapted radiotherapy.
		568-06-015. From Simulations to Actual Data - Generalizability and Robustness of Learned Image Reconstruction for Portable Low-Field MRI David Schote, Helge Herthum, Christoph Kolbitsch, Luca Calatroni, Kostas Papafitsoros, Andreas Kofler Physikalisch-Technische Bundesanstalt (PTB), Braunschweig and Berlin, Berlin, Germany Impact: Preliminary experiments suggest that hybrid physics-informed learned methods relying on the combination of hand-crafted priors with learned spatially varying regularization strengths might be less susceptible to data distribution shifts compared to methods whose regularizer is entirely learned from data.