A machine learning-derived risk score to predict left ventricular diastolic dysfunction from clinical cardiovascular magnetic resonance imaging.

Introduction: The evaluation of left ventricular diastolic dysfunction (LVDD) by clinical cardiac magnetic resonance (CMR) remains a challenge. We aimed to train and evaluate a machine-learning (ML) algorithm for the assessment of LVDD by clinical CMR variables and to investigate its prognostic value for predicting hospitalized heart failure and all-cause mortality. Methods: LVDD was characterized by echocardiography following the ASE guidelines. Eight demographic and nineteen common clinical CMR variables including delayed enhancement were used to train Random Forest models with a Bayesian optimizer. The model was evaluated using bootstrap and five-fold cross-validation. Area under the ROC curve (AUC) was utilized to evaluate the model performance. An ML risk score was used to stratify the risk of heart failure hospitalization and all-cause mortality. Results: A total of 606 consecutive patients underwent CMR and echocardiography within 7 days for cardiovascular disease evaluation. LVDD was present in 303 subjects by echocardiography. The performance of the ML algorithm was good using the CMR variables alone with an AUC of 0.868 (95% CI: 0.811-0.917), which was improved by combining with demographic data yielding an AUC 0.895 (95% CI: 0.845-0.939). The algorithm performed well in an independent validation cohort with AUC 0.810 (0.731-0.874). Subjects with higher ML scores (>0.4121) were associated with increased adjusted hazard ratio for a composite outcome than subjects with lower ML scores (1.72, 95% confidence interval 1.09-2.71). Discussion: An ML algorithm using variables derived from clinical CMR is effective in identifying patients with LVDD and providing prognostication for adverse clinical outcomes.

Left Ventricular Mass, Myocardial Structure, and Function in Severe Aortic Stenosis: an Echocardiographic and Cardiac Magnetic Resonance Imaging Study.

In severe aortic stenosis (AS), there are conflicting data on the prognostic implications of left ventricular (LV) hypertrophy (LVH). We aimed to characterize the LV geometry, myocardial matrix structural changes, and prognostic stratification using cardiac magnetic resonance imaging (CMR) and echocardiography in subjects with severe AS with and without LVH. Consecutive patients who had severe isolated AS and sufficient quality echocardiography and CMR within 6 months of each other were evaluated for LVH, cardiac structure, morphology, and late gadolinium-enhancement imaging. Kaplan-Meier curves, linear models, and proportional hazards models were used for prognostic stratification. There were 93 patients enrolled (mean age 74 ± 11 years, 48% female), of whom 38 (41%) had a normal LV mass index (LVMI), 41 (44%) had LVH defined at CMR by LVMI >2 SD higher than normal, and 14 (15% of the total) with >4 SD higher than the reference LVMI (severely elevated). The Society of Thoracic Surgeons scores were similar among the LVMI groups. Compared with those with normal LVMI, patients with LVH had higher LV end-diastolic and end-systolic volumes, increased late gadolinium-enhancement burden, and lower LV ejection fraction. Most notably, CMR feature-tracking global radial strain, 2-dimensional speckle-tracking echocardiography global longitudinal strain, and left atrial reservoir function were significantly worse. On the survival analyses, LVMI was not associated with a composite of all-cause mortality and/or heart failure hospitalization. In conclusion, compared with normal LVMI, elevated LVMI was not associated with a higher risk of adverse outcomes.

Reproducibility of global and segmental myocardial strain using cine DENSE at 3 T: a multicenter cardiovascular magnetic resonance study in healthy subjects and patients with heart disease.

BACKGROUND: While multiple cardiovascular magnetic resonance (CMR) methods provide excellent reproducibility of global circumferential and global longitudinal strain, achieving highly reproducible segmental strain is more challenging. Previous single-center studies have demonstrated excellent reproducibility of displacement encoding with stimulated echoes (DENSE) segmental circumferential strain. The present study evaluated the reproducibility of DENSE for measurement of whole-slice or global circumferential (Ecc), longitudinal (Ell) and radial (Err) strain, torsion, and segmental Ecc at multiple centers. METHODS: Six centers participated and a total of 81 subjects were studied, including 60 healthy subjects and 21 patients with various types of heart disease. CMR utilized 3 T scanners, and cine DENSE images were acquired in three short-axis planes and in the four-chamber long-axis view. During one imaging session, each subject underwent two separate DENSE scans to assess inter-scan reproducibility. Each subject was taken out of the scanner and repositioned between the scans. Intra-user, inter-user-same-site, inter-user-different-site, and inter-user-Human-Deep-Learning (DL) comparisons assessed the reproducibility of different users analyzing the same data. Inter-scan comparisons assessed the reproducibility of DENSE from scan to scan. The reproducibility of whole-slice or global Ecc, Ell and Err, torsion, and segmental Ecc were quantified using Bland-Altman analysis, the coefficient of variation (CV), and the intraclass correlation coefficient (ICC). CV was considered excellent for CV ≤ 10%, good for 10% < CV ≤ 20%, fair for 20% < CV ≤ 40%, and poor for CV > 40. ICC values were considered excellent for ICC > 0.74, good for ICC 0.6 < ICC ≤ 0.74, fair for ICC 0.4 < ICC ≤ 0.59, poor for ICC < 0.4. RESULTS: Based on CV and ICC, segmental Ecc provided excellent intra-user, inter-user-same-site, inter-user-different-site, inter-user-Human-DL reproducibility and good-excellent inter-scan reproducibility. Whole-slice Ecc and global Ell provided excellent intra-user, inter-user-same-site, inter-user-different-site, inter-user-Human-DL and inter-scan reproducibility. The reproducibility of torsion was good-excellent for all comparisons. For whole-slice Err, CV was in the fair-good range, and ICC was in the good-excellent range. CONCLUSIONS: Multicenter data show that 3 T CMR DENSE provides highly reproducible whole-slice and segmental Ecc, global Ell, and torsion measurements in healthy subjects and heart disease patients.

A comparison of both DENSE and feature tracking techniques with tagging for the cardiovascular magnetic resonance assessment of myocardial strain.

BACKGROUND: Myocardial strain is increasingly recognized as an important assessment for myocardial function. In addition, it also improves outcome prediction. However, there is lack of standardization in strain evaluation by cardiovascular magnetic resonance (CMR). In this study we compared strain values using multiple techniques and multiple vendor products. METHODS: Prospectively recruited patients with cardiomyopathy of diverse etiology (N = 77) and healthy controls (N = 10) underwent CMR on a 1.5 T scanner. Tagging, displacement encoding with stimulated echoes (DENSE) and balanced stead state free precession cine imaging were acquired on all subjects. A single matched mid left ventricular (LV) short axis plane was used for the comparisons of peak circumferential (Ecc) and radial strain (Err) and a 4-chamber view for longitudinal strain (Ell). Tagging images were analyzed using harmonic phase (HARP) and displacement encoding with stimulated echoes (DENSE) images using a proprietary program. Feature tracking (FT) was evaluated using 3 commercially available software from Tomtec Imaging Systems, Cardiac Image Modeller (CIM), and Circle Cardiovascular Imaging. Tagging data were used as reference. Statistic analyses were performed using paired t-test, intraclass correlation coefficient (ICC), Bland Altman limits of agreement and coefficient of variations. RESULTS: Average LV ejection fraction was 50% (range 32 to 62%). Regional LV wall motion abnormalities were present in 48% of the analyzed planes. The average Ecc was - 13 ± 4%, - 13 ± 4%, - 16 ± 6%, - 10 ± 3% and - 14 ± 4% for tagging, DENSE, Tomtec, CIM and Circle, respectively, with the best agreement seen in DENSE and Circle with tagging. The Err was highly varied with poor agreement across the techniques, 32 ± 24%, 40 ± 28%, 47 ± 26%, 64 ± 33% and 23 ± 9% for tagging, DENSE, Tomtec, CIM and Circle, respectively. The average Ell was - 14 ± 4%, - 8 ± 3%, - 13 ± 5%, - 11 ± 3% and - 12 ± 4% for tagging, DENSE, Tomtec, CIM and Circle, respectively with the best agreement seen in Tomtec and Circle with tagging. In the intra- and inter-observer agreement analysis the reproducibility of each technique was good except for Err by HARP. CONCLUSIONS: Small but important differences are evident in Ecc and Ell comparisons among vendors while large differences are seen in Err assessment. Our findings suggest that CMR strain values are technique and vendor dependent. Hence, it is essential to develop reference standard from each technique and analytical product for clinical use, and to sequentially compare patient data using the same software.

Effects of hemodynamics on global and regional lung perfusion: a quantitative lung perfusion study by magnetic resonance imaging.

BACKGROUND: Cardiac hemodynamics affect pulmonary vascular pressure and flow, but little is known of the effects of hemodynamics on lung perfusion at the tissue level. We sought to investigate the relationship between hemodynamic abnormalities in patients with left heart failure and global and regional lung perfusion using lung perfusion quantification by magnetic resonance imaging. METHODS AND RESULTS: Lung perfusion was quantified in 10 normal subjects and 28 patients undergoing clinically indicated left and right heart catheterization and same day research cardiac magnetic resonance imaging. A total of 228 lung slices were evaluated. Global lung perfusion, determined as the average of 6 coronal lung slices through the anterior, mid, and posterior left and right lungs, was significantly lower in patients with reduced cardiac index (<2.5 L/min per m(2)): 94±30 mL/100 mL per minute versus 132±40 mL/100 mL per minute in those with preserved cardiac index (≥2.5 L/min per m(2); P=0.003). The gravitational anterior to posterior perfusion gradient was inversely associated with left ventricular end-diastolic pressure (r=-0.728; P<0.001), resulting in a blunted perfusion gradient in patients with elevated left ventricular end-diastolic pressure, a finding largely attributed to the perfusion reduction in posterior lung regions. In a multivariate regression analysis adjusting for all hemodynamic variables, altered lung perfusion gradient was most closely associated with increased mean pulmonary arterial pressure (P=0.016). CONCLUSIONS: Increased left ventricular filling pressure and the resultant increase in pulmonary arterial pressure are associated with disruption of the normal gravitational lung perfusion gradient. Our findings underscore the complexity of heart-lung interaction in determining pulmonary hemodynamics in left heart failure.

Effects of respiratory cycle and body position on quantitative pulmonary perfusion by MRI.

PURPOSE: To evaluate the performance of lung perfusion imaging using two-dimensional (2D) first pass perfusion MRI and a quantitation program based on model-independent deconvolution algorithm. MATERIALS AND METHODS: In eight healthy volunteers 2D first pass lung perfusion was imaged in coronal planes using a partial Fourier saturation recovery stead state free precession (SSFP) technique with a temporal resolution of 160 ms per slice acquisition. The dynamic signal in the lung was measured over time and absolute perfusion calculated based on a model-independent deconvolution program. RESULTS: In the supine position mean pulmonary perfusion was 287 ± 106 mL/min/100 mL during held expiration. It was significantly reduced to 129 ± 68 mL/min/100 mL during held inspiration. Similar differences due to respiration were observed in prone position with lung perfusion much greater during expiration than during inspiration (271 ± 101 versus 99 ± 38 mL/min/100 mL (P < 0.01)). There was a linear increase in pulmonary perfusion from anterior to posterior lung fields in supine position. The perfusion gradient reversed in the prone position with the highest perfusion in anterior lung and the lowest in posterior lung fields. CONCLUSION: Lung perfusion imaging using a 2D saturation recovery SSFP perfusion MRI coupled with a model-independent deconvolution algorithm demonstrated physiologically consistent dynamic heterogeneity of lung perfusion distribution.