Related Experiment Video
Updated: Oct 11, 2025
Stereo-Imaging System DLT Calibration to Capture 3D In Situ Displacements of Stretched Peripheral Nerves
Published on: January 12, 2024
Direct pixel to pixel principal strain mapping from tagging MRI using end to end deep convolutional neural network
Khaled Z Abd-Elmoniem1, Inas A Yassine2, Nader S Metwalli3
1Biomedical and Metabolic Imaging Branch, National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK), National Institutes of Health, 10 Center Drive, Bldg. 10, CRC, Rm. 3-5340, Bethesda, MD, 20892, USA. abdelmoniemkz@mail.nih.gov.
A new deep learning framework using conditional generative adversarial networks (cGANs) accurately maps tissue strain from tagged MRI (tMRI). This method overcomes previous limitations, providing artifact-free strain maps for better analysis of cardiac and liver conditions.
Area of Science:
- Biomedical Engineering
- Medical Imaging
- Artificial Intelligence in Medicine
Background:
- Regional soft tissue mechanical strain is vital for understanding organ function and diagnosing disorders.
- Tagging magnetic resonance imaging (tMRI) is standard for assessing mechanical properties but struggles with artifact-free strain mapping.
- Accurate, high-resolution strain maps have been a long-standing challenge in tMRI analysis.
Purpose of the Study:
- To develop an end-to-end deep learning framework for direct pixel-to-pixel mapping of Eulerian principal strains from tMRI.
- To utilize convolutional neural networks (CNNs), specifically conditional generative adversarial networks (cGANs), for this task.
- To validate the framework against simulations and in-vivo data, comparing it to the conventional Harmonic Phase (HARP) method.
Main Methods:
- An end-to-end deep learning framework employing CNNs and four cGAN approaches was developed.
- The framework maps two-dimensional Eulerian principal strains directly from 1-1 spatial modulation of magnetization (SPAMM) tMRI.
- Validation involved Monte Carlo simulations and in-vivo datasets, with comparisons to the HARP method across various filter settings.
Main Results:
- The proposed cGAN approach achieved high correlation with ground-truth strain maps (R=0.90 and 0.92) in simulations, significantly outperforming HARP (R=0.12 and 0.73).
- The cGAN framework demonstrated substantially lower error compared to the best HARP method across all strain ranges.
- In-vivo results from healthy subjects and patients with pulmonary hypertension showed unprecedented clarity in anatomical, functional, and temporal details of strain maps.
Conclusions:
- The developed deep learning cGAN framework enables accurate, artifact-free Eulerian strain mapping directly from tMRI at native resolution.
- This approach overcomes limitations of conventional methods like HARP, offering superior accuracy and reduced error.
- The study demonstrates the feasibility and effectiveness of cGANs for myocardial and liver strain analysis, paving the way for improved diagnostic capabilities.
