Related Experiment Video
Updated: Aug 5, 2026
Three-Dimensional Phase Resolved Functional Lung Magnetic Resonance Imaging
Published on: June 21, 2024
AdaPR: Adaptive Plane Reformatting for 4D flow MRI using deep reinforcement learning
Javier Bisbal1, Julio Sotelo2, Maria I Valdes3
1Biomedical Imaging Center, Pontificia Universidad Católica de Chile, Santiago, Chile; Department of Electrical Engineering, School of Engineering, Pontificia Universidad Católica de Chile, Santiago, Chile; Millennium Institute for Intelligent Healthcare Engineering (iHEALTH), Santiago, Chile.
Adaptive Plane Reformatting (AdaPR) uses deep reinforcement learning for faster, more accurate 4D flow MRI. This orientation-independent method achieves expert-level flow quantification, improving cardiovascular assessments.
Area of Science:
- Medical Imaging
- Cardiovascular Imaging
- Artificial Intelligence in Medicine
Background:
- Four-dimensional Phase Contrast Magnetic Resonance Imaging (4D flow MRI) plane reformatting is slow and variable, hindering cardiovascular flow assessment.
- Current deep reinforcement learning (DRL) methods lack generalization across different scanners and institutions due to reliance on spatial alignment.
- Adaptive Plane Reformatting (AdaPR) addresses this by employing a local coordinate system for orientation-independent volume navigation.
Purpose of the Study:
- To introduce AdaPR, a novel DRL framework for robust and adaptive plane reformatting in 4D flow MRI.
- To overcome the limitations of existing DRL methods in handling variations in scanner and institutional data.
- To enable fast and accurate cardiovascular flow quantification.
Main Methods:
- Implementation of AdaPR using the Asynchronous Advantage Actor-Critic (A3C) algorithm.
- Validation on 88 4D flow MRI datasets from multiple vendors, including congenital heart disease patients.
- Utilizing a local coordinate system for navigating volumes with arbitrary positions and orientations.
Main Results:
- AdaPR achieved superior accuracy with a mean angular error of 6.12° ± 3.53° and distance error of 3.44 ± 3.02 mm.
- Consistent accuracy was maintained across varying volume orientations and positions.
- Flow measurements from AdaPR planes showed excellent correlation with manual observers (R²=0.972, R²=0.967), comparable to inter-observer agreement (R²=0.969).
Conclusions:
- AdaPR offers robust, orientation-independent plane reformatting for 4D flow MRI, enabling flow quantification comparable to expert observers.
- The framework demonstrates adaptability across diverse datasets and scanners.
- AdaPR shows potential for broader applications in medical imaging beyond 4D flow MRI.
