MRI Data Multi-Site Harmonization
Weekday Course
ORGANIZERS: Berkin Bilgic, Yogesh Rathi, Gary Zhang
Wednesday, 14 May 2025
313A
08:15 -
10:15
Moderators: Suheyla Cetin-Karayumak & Eleftherios Garyfallidis
Skill Level: Basic to Advanced
Session Number: W-01
No CME/CE Credit
Session Number: W-01
Overview
This tutorial will start with covering the overall problem of reproducibility and how it impacts reserach studies, with specific emphasis on inter-scanner and inter-site variability. Three different ways in which the field has addressed the inter-scanner variability will be covered, beginning with the use of statistical methods on post-processed data and the use of AI algorithms to remove variability from the data itself. Finally, the emerging field of vendor neutral pulse sequence programming to acquire data consistenly across scanners will be covered.
Target Audience
Clinicians who want to learn about inter-scanner variability. Students, research fellows and investigators who want to know about different ways to address the problem of inter-scanner variability and how harmonization techniques help reduce bias and scanner-specific effects in your data.
Educational Objectives
As a result of attending this course, participants should be able to:
• Describe the challenges to reproducibility due to inter-scanner effects;
• Describe harmonization techniques for data already acquired; and
• Summarize harmonization methods for prospective MRI studies.
| 08:15 | | Retrospective Harmonization: Statistical Methods Chantal Tax |
| 08:45 | | Prospective Harmonized Acquisition & Reconstruction Shohei Fujita Keywords: Image acquisition: Sequences, Image acquisition: Reconstruction, Transferable skills: Reproducible research This talk introduces a prospective approach to harmonize MRI data acquisition and reconstruction across multiple sites and scanners. Focusing primarily on the open-source Pulseq framework, I explore how similar pulse sequences can be implemented on various platforms, promoting consistent acquisition protocols. Such vendor-harmonized imaging significantly reduces cross-vendor variability in MRI metrics compared to closed vendor-specific sequences. This harmonization, combined with unified image reconstruction, facilitates reliable data pooling across centers. Alongside the advantages, I address the limitations and practical considerations of using Pulseq. Insights shared will benefit imaging physicists, MRI technologists, and radiologists looking to minimize variability in multi-site MRI research. |
| 09:15 | | Retrospective Harmonization: AI Methods Daniel Moyer |
| 09:45 | | Challenges to Reproducibility in Multi-Site MRI Patricia Grant |