Cape Town - 2026 ISMRM-ISMRT Annual Meeting and Exhibition

AI Methods Beyond Reconstruction

Oral

Acquisition & Reconstruction

Monday, 11 May 2026

Auditorium 1

08:20 - 10:10

Moderators: Esin Ozturk Isik & KyungHyun Sung

Session Number: 303-01

No CME/CE Credit

Similar Sessions

This session showcases the transformative impact of Artificial Intelligence across a variety of topics beyond reconstruction.

08:20		303-01-001. In-depth Radiological and Explainability Analysis of Deep Learning Predictions for IPMN Risk Stratification on MRI Halil Ertugrul Aktas, Andrea Bejar, Ziliang Hong, Rutger Hendrix, Elif Keles, Alpay Medetalibeyoglu, Sukru Mehmet Erturk, Rajesh Keswani, Frank Miller, Michael Wallace, Gorkem Durak, Ulas Bagci Northwestern University Feinberg School of Medicine, Chicago, United States of America Impact: This study demonstrates that explainability and uncertainty frameworks enhance the transparency and reliability of MRI-based deep learning (DL) models for Intraductal papillary mucinous neoplasms (IPMNs) malignancy risk assessment by revealing imaging-derived factors associated with variable prediction performance and confidence.
08:31		303-01-002. Initial Clinical Evaluation of DeepAcq: Ultra-Fast Multi-Contrast MRI and T₁/T₂ Mapping in Multiple Sclerosis Beril Alyuz, Shihan Qiu, Hsu-Lei Lee, Chang Gao, Sreekanth Madhusoodhanan , Dan Ruan, Nancy Sicotte, Pascal Sati, Yibin Xie, Debiao Li Cedars-Sinai Medical Center, Los Angeles, United States of America Impact: DeepAcq enables high-resolution multi-contrast MRI and quantitative mapping in under 3 minutes, with performance comparable to reference acquisitions. With clinical validation, it holds promise for increasing scanner efficiency, improving patient comfort, and facilitating longitudinal monitoring of disease progression in MS.
08:42		303-01-003. Patch2Voxel: Denoising perfusion MRI with Self-Supervised Learning Puneet Kumar, Natenael Semmineh, Indranil Guha, C. Chad Quarles The University of Texas MD Anderson Cancer Center, Houston, United States of America Impact: Patch2Voxel (P2V) enables effective denoising of DSC-MRI time-series, enabling more reliable estimation of hemodynamic biomarkers. It holds potential to improve the application of DSC-MRI for brain tumor diagnosis, treatment planning, and monitoring.
08:53		303-01-004. Enhancement Margin Detection in Glioma Patients using Non-Contrast MRI. Human versus artificial intelligence Ivar Wamelink, Aynur Azizova, Elif Kaya, João Ramos, Aziz Tan, Norman Kornemann, Frederik Barkhof, Alle Meije Wink, Vera Keil Amsterdam University Medical Center, University of Amsterdam, Amsterdam, Netherlands Impact: Human delineation of glioblastoma enhancing tumor core is feasible using GBCA-free sequences. AI-assistance lifts accuracy to levels achieved by contrast-dependent automated benchmarks. This approach facilitates contrast use reduction in routine clinical practice and guides physicians towards non-invasive alternatives.
09:04		303-01-005. AI‑Enhanced ULF MRI Achieves High-Field MRI Diagnostic Accuracy in Neonatal Neuroimaging: A Blinded Multi‑Reader Study Austin Tapp, Anne Groteklaes, Sebiha Demir, Zeynep Bendella, Paul Cawley, Ralf Claugberg, Axel Heep, Ildiko Kabat, Andreas Mueller, Natasha Lepore, Hemmen Sabir, Marius Linguraru Children's National Hospital, Washington DC, United States of America Impact: AI image enhancement of ULF (0.064T) neonatal brain MRI resulted in HF‑level diagnostic accuracy by neuroradiologists while preserving bedside workflows. This could expand safe, sedation‑sparing neuroimaging access in NICUs and shift neonatal brain MRI acquisition toward point‑of‑care practices.
09:15		303-01-006. Towards Clinically Feasible Multi-Parametric MRI Reconstruction using Transfer Learning: From Hours to Minutes Natalia Pato Montemayor, Yuting Chen, Yohan Jun, Jocelyn Philippe, Marcel Dominik Nickel, Patrick Liebig, Robin Heidemann, Jean-Philippe Thiran, Tom Hilbert, Gian Franco Piredda, Berkin Bilgic, Thomas Yu Siemens Healthineers International AG, Lausanne, Switzerland Impact: Using transfer learning and multi-contrast fine-tuning, the proposed reconstruction method reduces MIMOSA reconstruction time to 10 minutes while maintaining quantitative accuracy, bridging the gap for practical use in high-resolution comprehensive brain tissue assessment including $T_{1},T_{2},T_{2}^*$ and susceptibility mapping.
09:26		303-01-007. Magnetic resonance fingerprinting reconstruction using spatiotemporal maps and implicit neural representation Rodrigo Lobos, Christopher Keen, Sydney Kaplan, Hongze Yu, Yun Jiang, Nicole Seiberlich, Jon-Fredrik Nielsen, Jeffrey Fessler, Douglas Noll University of Michigan, Ann Arbor, United States of America Impact: The proposed reconstruction framework utilizes a signal model with enhanced representation capabilities that reduces the number of unknowns (i.e., coefficient images) compared to using a low-rank model learned from a dictionary, facilitating reconstruction.
09:37		303-01-008. Transfer learning–based segmentation of paediatric optic pathway gliomas with incomplete MRI Xinyu Tian, Enrico De Vita, Kshitij Mankad, Emily Drabek-Maunder, James Ruffle, Chris Clark University College London, London, United Kingdom Impact: The framework may facilitate more consistent and practical volumetric assessment of paediatric optic pathway gliomas, supporting longitudinal follow-up and potentially improving tumour monitoring in routine clinical practice.
09:48		303-01-009. Deep Learning driven Inversion framework for shear Modulus estimation in magnetic resonance Elastography (DIME) Hassan Iftikhar, Rizwan Ahmad, Arunark Kolipaka Impact: Deep Learning (DL) driven inversion framework for Magnetic Resonance Elastography (MRE) is proposed which provides robust inversion framework that uses Finite Element Modelling based dataset for training of a convolutional neural network (CNN).
09:59		303-01-010. Guided Discussion KyungHyun Sung University of California Los Angeles, Los Angeles, United States of America