Deep Learning for Automatic Volumetric Segmentation of Left Ventricular Myocardium and Ischemic Scar from Multi-Slice LGE-CMR.

PURPOSE: This study details application of deep learning for automatic volumetric segmentation of left ventricular myocardium and scar and automated quantification of myocardial ischemic scar burden from late-gadolinium enhancement cardiovascular magnetic resonance (LGE-CMR). MATERIALS AND METHODS: We included 501 images and manual segmentations of short-axis LGE-CMR from over 20 multinational sites, from which 377 studies were used for training and 124 studies from unique participants for internal validation. A third test set of 52 images was used for external evaluation. Three models, U-Net, Cascaded U-Net, and U-Net++, were trained with a novel adaptive weighted categorical cross entropy loss function. Model performance was evaluated using concordance correlation coefficients (CCC) for left ventricular (LV) mass and percent myocardial scar burden. RESULTS: Cascaded U-Net was found to be the best model for quantification of LV mass and scar percentage. The model exhibited a mean difference of -5 ± 23 g for LV mass, -0.4 ± 11.2 g for scar mass, and -0.8 ± 7% for percent scar. CCC were 0.87, 0.77, and 0.78 for LV mass, scar mass, and percent scar burden, respectively, in the internal validation set and 0.75, 0.71, and 0.69, respectively, in the external test set. For segmental scar mass, CCC was 0.74 for apical scar, 0.91 for midventricular scar, and 0.73 for basal scar, demonstrating moderate to strong agreement. CONCLUSION: We successfully trained a convolutional neural network for volumetric segmentation and analysis of left ventricular scar burden from LGE-CMR images in a large, multinational cohort of participants with ischemic scar.

Association of Cardiovascular Risk Factors and Myocardial Fibrosis With Early Cardiac Dysfunction in Type 1 Diabetes: The Diabetes Control and Complications Trial/Epidemiology of Diabetes Interventions and Complications Study.

OBJECTIVE: We investigated the association of cardiovascular risk factors and myocardial fibrosis with early cardiac dysfunction in type 1 diabetes. RESEARCH DESIGN AND METHODS: Participants with type 1 diabetes aged 13-39 years without a known history of cardiovascular disease (CVD) (n = 1,441) were recruited into the Diabetes Control and Complications Trial (1983-1993) and subsequently followed in the Epidemiology of Diabetes Interventions and Complications study (1994 to present). Seven hundred fourteen participants underwent cardiac magnetic resonance (CMR) imaging (2007-2009) with late gadolinium enhancement sequences to assess ischemic and nonischemic scars and tagging sequences to evaluate circumferential strain. CMR-derived T1 mapping also was used to assess interstitial fibrosis. The influence of cardiovascular risk factors and myocardial scar on circumferential strain was assessed using linear regression. RESULTS: Circumferential dysfunction was consistently associated with older age, male sex, smoking history, obesity, higher blood pressure, lower HDL cholesterol, and higher mean HbA1c. Participants with nonischemic scars (n = 16) had the worst circumferential function compared with those without scars (ß ± SE 1.32 ± 0.60; P = 0.03). In sex-adjusted models, the correlation between T1 times and circumferential strain was not significant. In the fully adjusted models, a trend toward circumferential dysfunction in participants with nonischemic scars was found. Left ventricular ejection fraction was not associated with risk factors but was significantly lower if a myocardial scar was present. CONCLUSIONS: Traditional CVD risk factors and elevated HbA1c levels are major factors related to early cardiac dysfunction in type 1 diabetes. Nonischemic myocardial scar, possibly as a marker of chronic exposure to known risk factors, may predict early cardiac dysfunction mediated by diffuse myocardial fibrosis as seen in diabetic cardiomyopathy.

Reproducibility of functional aortic analysis using magnetic resonance imaging: the MESA.

AIMS: To assess the test-retest, intra- and inter-reader reliability of thoracic aorta measurements by magnetic resonance imaging (MRI). METHODS AND RESULTS: Twenty-five participants underwent aortic MRI twice over 13 ± 7 days. All aortic variables from baseline and repeat MR were analysed using a semi-automated method by the ARTFUN software. To assess the inter-study reproducibility of aortic variables, we calculated intraclass correlation coefficient (ICC) for individual aortic measurements. Intra- and inter-observer variability was also assessed using the baseline MR data. Mean ascending aortic strain had moderate inter-study reproducibility (11.53 ± 6.44 vs. 10.55 ± 6.64, P = 0.443, ICC = 0.53, P < 0.01). Mean descending aortic strain and arch pulse wave velocity (PWV) had good inter-study reproducibility (descending aortic strain: 8.65 ± 5.30 vs. 8.35 ± 5.26, P = 0.706, ICC = 0.74, P < 0.001; PWV: 9.92 ± 4.18 vs. 9.94 ± 4.55, P = 0.968, ICC = 0.77, P < 0.001, respectively). All aortic variables had excellent intra- and inter-observer reproducibility (intra-: ICC range, 0.87-0.99, inter-: ICC range, 0.56-0.99, respectively). CONCLUSION: Inter-study reproducibility of all aortic variables was acceptable. Intra- and inter-observer reproducibility of all aortic variables was excellent. MRI can provide a repeatable method of measuring aortic structural and functional parameters.

Left atrial structure and functional quantitation using cardiovascular magnetic resonance and multimodality tissue tracking: validation and reproducibility assessment.

BACKGROUND: Left atrium (LA) strain, volume and function are important markers of cardiovascular disease and myocardial impairment. We aimed to assess the accuracy of LA biplane volume and function measured by Multimodality Tissue Tracking (MTT). Also we assessed the inter-study reproducibility for cardiovascular magnetic resonance (CMR) derived LA volume and function parameters. METHODS: Thirty subjects (mean age: 71.3 ± 8.7, 87% male) including twenty subjects with cardiovascular events and ten healthy subjects, with CMR were evaluated in the Multi-Ethnic Study of Atherosclerosis (MESA). LA volumes were computed by the modified biplane method from 2- and 4-chamber projections and the Simpson's method from short-axis slices using both methods - manual and semi-automated delineation using MTT. LA total, active and passive ejection fractions were calculated. Pearson's correlation and Bland-Altman analysis were used to compare the measurements. In a second sample of 25 subjects (age: 65.7 ± 7.1, 72% males) inter study, intra and inter reader reliability analysis was performed. The intra-class correlation coefficient (ICC) was evaluated. RESULTS: Left atrial MTT structural and functional parameters were not different from manual delineation, yet image analysis was only half as time consuming on average with MTT. Maximal volume MTT was not different between the Simpson's and Biplane methods, functional parameters, however were different. MTT allowed us to measure multiple LA parameters with good-excellent (ICC; 0.88- 0.98, p < 0.001) intra-and inter reader reproducibility and fair-good (ICC; 0.44-0.82, p < 0.05-0.001) inter study reproducibility. CONCLUSIONS: MTT derived LA biplane volume and function is accurate and reproducible and is suited for use in longitudinal studies.

Comparison of strain measurement from multimodality tissue tracking with strain-encoding MRI and harmonic phase MRI in pulmonary hypertension.

BACKGROUND: Pixel-based multimodality tissue tracking (MTT) is a new noninvasive method for the quantification of cardiac deformation from cine image of MRI. The aim of this study is to validate bi-ventricular strain measurement by MTT compared to strain-encoding (SENC) MRI and harmonic phase (HARP) MRI in pulmonary hypertension (PH) patients. METHODS: In 45 subjects (30 PH patients and 15 normal subjects), RV and LV peak global longitudinal strains (Ell) were measured from long axis 4 chamber view using MTT. LV peak global circumferential strains (Ecc) by MTT were measured from short axis. For validation, RV and LV Ell by MTT were compared to measures by SENC-MRI from short axis, and LV Ecc by MTT was compared to measures by short axis tagged MRI analysis (HARP). Reproducibility of MTT was also determined. RESULTS: MTT quantified RV Ell correlated closely to those of SENC (r=0.72, p<0.001), with good limits of agreement. LV Ell quantified by MTT showed moderate correlation with SENC (r=0.57, p=0.001), and LV Ecc by MTT also showed moderate correlation with HARP (-16.9±4.1 vs -14.3±3.5, p<0.001 for all, r=0.60, p<0.001). RV Ell negatively correlated with RVEF (r=-0.53, p=0.001) and also positively correlated with mean PAP in PH patients (r=0.60, p=0.001). Strain measurement by MTT showed high reproducibility. CONCLUSIONS: We demonstrate that MTT is a reproducible tool for quantification of cardiac deformation using cine images in PH patients. Hence, it could serve as a new rapid and comprehensive technique for clinical assessment of regional cardiac function.

Accuracy of Doppler echocardiography in the hemodynamic assessment of pulmonary hypertension.

RATIONALE: Transthoracic Doppler echocardiography is recommended for screening for the presence of pulmonary hypertension (PH). However, some recent studies have suggested that Doppler echocardiographic pulmonary artery pressure estimates may frequently be inaccurate. OBJECTIVES: Evaluate the accuracy of Doppler echocardiography for estimating pulmonary artery pressure and cardiac output. METHODS: We conducted a prospective study on patients with various forms of PH who underwent comprehensive Doppler echocardiography within 1 hour of a clinically indicated right-heart catheterization to compare noninvasive hemodynamic estimates with invasively measured values. MEASUREMENTS AND MAIN RESULTS: A total of 65 patients completed the study protocol. Using Bland-Altman analytic methods, the bias for the echocardiographic estimates of the pulmonary artery systolic pressure was -0.6 mm Hg with 95% limits of agreement ranging from +38.8 to -40.0 mm Hg. Doppler echocardiography was inaccurate (defined as being greater than +/-10 mm Hg of the invasive measurement) in 48% of cases. Overestimation and underestimation of pulmonary artery systolic pressure by Doppler echocardiography occurred with a similar frequency (16 vs. 15 instances, respectively). The magnitude of pressure underestimation was greater than overestimation (-30 +/- 16 vs. +19 +/- 11 mm Hg; P = 0.03); underestimates by Doppler also led more often to misclassification of the severity of the PH. For cardiac output measurement, the bias was -0.1 L/min with 95% limits of agreement ranging from +2.2 to -2.4 L/min. CONCLUSIONS: Doppler echocardiography may frequently be inaccurate in estimating pulmonary artery pressure and cardiac output in patients being evaluated for PH.

Tricuspid annular displacement predicts survival in pulmonary hypertension.

RATIONALE: Right ventricular (RV) function is an important determinant of prognosis in pulmonary hypertension. However, noninvasive assessment of the RV function is often limited by complex geometry and poor endocardial definition. OBJECTIVES: To test whether the degree of tricuspid annular displacement (tricuspid annular plane systolic excursion [TAPSE]) is a useful echo-derived measure of RV function with prognostic significance in pulmonary hypertension. METHODS: We prospectively studied 63 consecutive patients with pulmonary hypertension who were referred for a clinically indicated right heart catheterization. Patients underwent right heart catheterization immediately followed by transthoracic echocardiogram and TAPSE measurement. RESULTS: In the overall cohort, a TAPSE of less than 1.8 cm was associated with greater RV systolic dysfunction (cardiac index, 1.9 vs. 2.7 L/min/m2; RV % area change, 24 vs. 33%), right heart remodeling (right atrial area index, 17.0 vs. 12.1 cm(2)/m), and RV-left ventricular (LV) disproportion (RV/LV diastolic area, 1.7 vs. 1.2; all p < 0.001), versus a TAPSE of 1.8 cm or greater. In patients with pulmonary arterial hypertension (PAH; n = 47), survival estimates at 1 and 2 yr were 94 and 88%, respectively, in those with a TAPSE of 1.8 cm or greater versus 60 and 50%, respectively, in subjects with a TAPSE less than 1.8 cm. The unadjusted risk of death (hazard ratio) in patients with a TAPSE less than 1.8 versus 1.8 cm or greater was 5.7 (95% confidence interval, 1.3-24.9; p = 0.02) for the PAH cohort. For every 1-mm decrease in TAPSE, the unadjusted risk of death increased by 17% (hazard ratio, 1.17; 95% confidence interval, 1.05-1.30; p = 0.006), which persisted after adjusting for other echocardiographic and hemodynamic variables and baseline treatment status. CONCLUSIONS: TAPSE powerfully reflects RV function and prognosis in PAH.