Artifact Suppression in Cine DENSE MRI Using Deep Learning 304 views

Myocardial strain imaging adds diagnostic and prognostic value in the assessment of many types of heart disease. Cine displacement encoding with stimulated echoes (DENSE) is among the most accurate and reproducible myocardial strain imaging methods with growing clinical applications. This dissertation research seeks to develop methods to shorten the scan time for cine DENSE and to develop free-breathing cine DENSE methods, both of which would facilitate greater clinical usage of the method. During DENSE data acquisition, a signal due to T1-relaxation (T1-echo) is simultaneously acquired along with the displacement-encoded stimulated echo which generates stripe artifacts and leads to inaccurate strain measurement. The T1-echo is typically suppressed by acquiring additional phase-cycled data, which despite its effectiveness, leads to increased scan time. In addition to the T1-echo, respiratory motion also leads to undesired artifacts. The standard DENSE image acquisition protocol requires breath-holding to avoid respiratory motion artifacts, and this can be challenging for heart failure and pediatric patients and for those under sedation. This creates a broad need for free-breathing methods. In free-breathing cine DENSE acquisitions, three types of artifacts arise: (a) those due to incomplete suppression of the T1-echo, (b) those due to (approximately) rigid translation of the tissue, and (c) encoding of breathing-induced tissue motion into the phase of the stimulated-echo. Previously, methods were developed to compensate for the first and the second types of artifacts in DENSE. The third type of artifact causes unique respiratory-induced k-space phase errors which correspond to phase shifts in the image domain and lead to signal loss and phase corruption artifacts. A deep learning model was developed for suppression of the artifact-generating T1-echo in cine DENSE for the purpose of eliminating the phase-cycling acquisitions and reducing the scan time limitation. A U-Net (DAS-Net) was trained to suppress the artifact-generating T1-echo using phase-cycled data as the ground truth. A data augmentation method was developed that generates synthetic DENSE images with arbitrary displacement encoding frequencies to suppress the T1-echo modulated for a range of frequencies. DAS-Net was evaluated on non-phase-cycling cine DENSE images from healthy subjects. Comparisons between DAS-Net processed images and the corresponding phase-cycling reference data using signal-to-noise ratio and strain measurements demonstrated that DAS-Net provides an effective alternative approach for suppression of the artifact-generating T1-echo in DENSE MRI, enabling a 42% reduction in scan time compared to DENSE with phase cycling. A new model was introduced that describes artifacts due to encoding of respiratory motion into the phase of the stimulated echo. Phantom experiment and Bloch-equation simulations were performed to validate the model. The model was used along with the simulation of respiratory motion to generate synthetic images with phase shift artifacts to train a U-Net, DENSE-RESP-NET for compensation of signal loss and phase corruption artifacts. Evaluations of the DENSE-RESP-NET on self-navigated free-breathing cine DENSE from healthy volunteers showed that the DENSE-RESP-NET is an effective method to compensate for breathing-associated signal loss and phase corruption artifacts. The developed motion compensation method, DENSE-RESP-NET, was used and evaluated in concert with adaptive free-breathing acquisitions and self-navigation applied on healthy volunteers and heart disease patients. Assessment of motion compensated images for strain and signal-to-noise ratio demonstrated that the proposed motion compensation method outperforms the conventional diaphragm navigator-gated method and provides reliable free-breathing cine DENSE acquisitions for measurement of systolic and diastolic parameters.

PHD (Doctor of Philosophy)

Artifact Suppression; Deep Learning; Magnetic Resonance Imaging; Mayocardial Strain Imaging; Motion Compensation; Free-breathing

English

All rights reserved (no additional license for public reuse)

2022-07-21