Dataset Information

CINENet: deep learning-based 3D cardiac CINE MRI reconstruction with multi-coil complex-valued 4D spatio-temporal convolutions.

ABSTRACT: Cardiac CINE magnetic resonance imaging is the gold-standard for the assessment of cardiac function. Imaging accelerations have shown to enable 3D CINE with left ventricular (LV) coverage in a single breath-hold. However, 3D imaging remains limited to anisotropic resolution and long reconstruction times. Recently deep learning has shown promising results for computationally efficient reconstructions of highly accelerated 2D CINE imaging. In this work, we propose a novel 4D (3D?+?time) deep learning-based reconstruction network, termed 4D CINENet, for prospectively undersampled 3D Cartesian CINE imaging. CINENet is based on (3?+?1)D complex-valued spatio-temporal convolutions and multi-coil data processing. We trained and evaluated the proposed CINENet on in-house acquired 3D CINE data of 20 healthy subjects and 15 patients with suspected cardiovascular disease. The proposed CINENet network outperforms iterative reconstructions in visual image quality and contrast (+?67% improvement). We found good agreement in LV function (bias?±?95% confidence) in terms of end-systolic volume (0?±?3.3 ml), end-diastolic volume (-?0.4?±?2.0 ml) and ejection fraction (0.1?±?3.2%) compared to clinical gold-standard 2D CINE, enabling single breath-hold isotropic 3D CINE in less than 10 s scan and?~?5 s reconstruction time.

SUBMITTER: Kustner T

PROVIDER: S-EPMC7426830 | biostudies-literature | 2020 Aug

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

CINENet: deep learning-based 3D cardiac CINE MRI reconstruction with multi-coil complex-valued 4D spatio-temporal convolutions.

Küstner Thomas T Fuin Niccolo N Hammernik Kerstin K Bustin Aurelien A Qi Haikun H Hajhosseiny Reza R Masci Pier Giorgio PG Neji Radhouene R Rueckert Daniel D Botnar René M RM Prieto Claudia C

Scientific reports 20200813 1

Cardiac CINE magnetic resonance imaging is the gold-standard for the assessment of cardiac function. Imaging accelerations have shown to enable 3D CINE with left ventricular (LV) coverage in a single breath-hold. However, 3D imaging remains limited to anisotropic resolution and long reconstruction times. Recently deep learning has shown promising results for computationally efficient reconstructions of highly accelerated 2D CINE imaging. In this work, we propose a novel 4D (3D + time) deep learn ...[more]

PMID: 32792507

Similar Datasets

Project description:PurposeTo generate fully automated and fast 4D-flow MRI-based 3D segmentations of the aorta using deep learning for reproducible quantification of aortic flow, peak velocity, and dimensions.MethodsA total of 1018 subjects with aortic 4D-flow MRI (528 with bicuspid aortic valve, 376 with tricuspid aortic valve and aortic dilation, 114 healthy controls) comprised the data set. A convolutional neural network was trained to generate 3D aortic segmentations from 4D-flow data. Manual segmentations served as the ground truth (N = 499 training, N = 101 validation, N = 418 testing). Dice scores, Hausdorff distance, and average symmetrical surface distance were calculated to assess performance. Aortic flow, peak velocity, and lumen dimensions were quantified at the ascending, arch, and descending aorta and compared using Bland-Altman analysis. Interobserver variability of manual analysis was assessed on a subset of 40.ResultsConvolutional neural network segmentation required 0.438 ± 0.355 seconds versus 630 ± 254 seconds for manual analysis and demonstrated excellent performance with a median Dice score of 0.951 (0.930-0.966), Hausdorff distance of 2.80 (2.13-4.35), and average symmetrical surface distance of 0.176 (0.119-0.290). Excellent agreement was found for flow, peak velocity, and dimensions with low bias and limits of agreement less than 10% difference versus manual analysis. For aortic volume, limits of agreement were moderate within 16.3%. Interobserver variability (median Dice score: 0.950; Hausdorff distance: 2.45; and average symmetrical surface distance: 0.145) and convolutional neural network-based analysis (median Dice score: 0.953-0.959; Hausdorff distance: 2.24-2.91; and average symmetrical surface distance: 0.145-1.98 to observers) demonstrated similar reproducibility.ConclusionsDeep learning enabled fast and automated 3D aortic segmentation from 4D-flow MRI, demonstrating its potential for efficient clinical workflows. Future studies should investigate its utility for other vasculature and multivendor applications.

Project description:PurposeCurrent cardiovascular magnetic resonance (CMR) examinations require expert planning, multiple breath holds, and 2D imaging. To address this, we sought to develop and validate a comprehensive free-breathing 3D cine function and flow CMR examination using a steady-state free precession (SSFP) sequence to depict anatomy fused with a spatially registered phase contrast (PC) sequence for blood flow analysis.MethodsIn a prospective study, 25 patients underwent a CMR examination which included a 3D cine SSFP sequence and a 3D cine PC (also known as 4D flow) sequence acquired during free-breathing and after the administration of a gadolinium-based contrast agent. Both 3D sequences covered the heart and mediastinum, and used retrospective vectorcardiogram gating (20 phases/beat interpolated to 30 phases/beat) and prospective respiratory motion compensation confining data acquisition to end-expiration. Cardiovascular measurements derived from the 3D cine SSFP and PC images were then compared with those from standard 2D imaging.ResultsAll 3D cine SSFP and PC acquisitions were completed successfully. The mean time for the 3D cine sequences including prescription was shorter than that for the corresponding 2D sequences (21 min vs. 36 min, P-value <0.001). Left and right ventricular end-diastolic volumes and stroke volumes by 3D cine SSFP were slightly smaller than those from 2D cine SSFP (all biases ≤5%). The blood flow measurements from the 3D and 2D sequences had close agreement in the ascending aorta (bias -2.6%) but main pulmonary artery flow was lower with the 3D cine sequence (bias -11.2%).ConclusionCompared to the conventional 2D cine approach, a comprehensive 3D cine function and flow examination was faster and yielded slightly lower left and right end-diastolic volumes, stroke volumes, and main pulmonary artery blood flow. This free-breathing 3D cine approach allows flexible post-examination data analysis and has the potential to make examinations more comfortable for patients and easier to perform for the operator.

Project description:Two-dimensional (2D) strain analysis is constrained by geometry-dependent reference directions of deformation (i.e. radial, circumferential, and longitudinal) following the assumption of cylindrical chamber architecture. Three-dimensional (3D) principal strain analysis may overcome such limitations by referencing intrinsic (i.e. principal) directions of deformation. This study aimed to demonstrate clinical feasibility of 3D principal strain analysis from routine 2D cine MRI with validation to strain from 2D tagged cine analysis and 3D speckle tracking echocardiography. Thirty-one patients undergoing cardiac MRI were studied. 3D strain was measured from routine, multi-planar 2D cine SSFP images using custom software designed to apply 4D deformation fields to 3D cardiac models to derive principal strain. Comparisons of strain estimates versus those by 2D tagged cine, 2D non-tagged cine (feature tracking), and 3D speckle tracking echocardiography (STE) were performed. Mean age was 51?±?14 (36% female). Mean LV ejection fraction was 66?±?10% (range 37-80%). 3D principal strain analysis was feasible in all subjects and showed high inter- and intra-observer reproducibility (ICC range 0.83-0.97 and 0.83-0.98, respectively-p?<?0.001 for all directions). Strong correlations of minimum and maximum principal strain were respectively observed versus the following: 3D STE estimates of longitudinal (r?=?0.81 and r?=?-0.64), circumferential (r?=?0.76 and r?=?-0.58) and radial (r?=?-0.80 and r?=?0.63) strain (p?<?0.001 for all); 2D tagged cine estimates of longitudinal (r?=?0.81 and r?=?-0.81), circumferential (r?=?0.87 and r?=?-0.85), and radial (r?=?-0.76 and r?=?0.81) strain (p?<?0.0001 for all); and 2D cine (feature tracking) estimates of longitudinal (r?=?0.85 and -0.83), circumferential (r?=?0.88 and r?=?-0.87), and radial strain (r?=?-0.79 and r?=?0.84, p?<?0.0001 for all). 3D principal strain analysis is feasible using routine, multi-planar 2D cine MRI and shows high reproducibility with strong correlations to 2D conventional strain analysis and 3D STE-based analysis. Given its independence from geometry-related directions of deformation this technique may offer unique benefit for the detection and prognostication of myocardial disease, and warrants expanded investigation.

Dataset Information

CINENet: deep learning-based 3D cardiac CINE MRI reconstruction with multi-coil complex-valued 4D spatio-temporal convolutions.

Publications

CINENet: deep learning-based 3D cardiac CINE MRI reconstruction with multi-coil complex-valued 4D spatio-temporal convolutions.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure