Search

The aims of this study were to investigate the dynamic changes in the vena contracta (VC) and proximal isovelocity surface area (PISA) through systole in patients with hypoplastic left heart syndrome and tricuspid regurgitation and to identify the stage of systole (early, mid, or late) in which VC and PISA radius are optimal.

Ineffective right ventricular (RV) adaptation to increasing pulmonary arterial (PA) afterload in pulmonary vascular disease (PVD) significantly contributes to morbidity and mortality. PVD in systemic sclerosis (SSc) arises through various mechanisms, yet detecting abnormal contractile response remains challenging. Here, we examine whether echocardiographic RV-PA coupling metrics correlate with invasive pressure-volume (PV) loops, enhancing the prediction of adverse clinical outcomes in SSc-PVD patients.

Patients with tetralogy of Fallot (TOF) may undergo pulmonary valve replacement (PVR) after initial full repair. We investigated indices of biventricular function, work and efficiency of TOF patients' using noninvasive pressure-volume (PV) loop analysis on cardiovascular magnetic resonance (CMR) images and compared pre-and post PVR groups. Biventricular segmentations of steady state free precession CMR images were performed using custom validated software (Segment version 2.0 R7067). Brachial cuff pressure estimated left ventricular (LV) systolic pressure. Right ventricular (RV) inputs were obtained from pre-PVR cardiac catheterization data. Biventricular PV loops were then derived using a time-varying elastance model. Twenty seven patients were studied: (22 pre-PVR, 5 post-PVR), mean age of 20±10.5 years and 83% male. RV stroke volume significantly differed before and after PVR (73.2±25 ml vs. 41±10 mL, P=0.01). RV stroke work (SW) and mean external power (MEP) were significantly less post-PVR, but there were no significant differences in the LV hemodynamic indices. TOF patients have reduced RV SW and MEP post-PVR suggesting improved hemodynamics. Noninvasive biventricular PV loop analysis shows potential for integration into standard CMR imaging of TOF and provides hemodynamic data that could influence management decisions.

Intro:Right ventricular (RV) dysfunction worsens outcomes in pulmonary hypertension (PH) due to heart failure with reduced ejection fraction (HF), yet underlying mechanisms are unknown. We previously reported RV myofilament max calcium-activated tension (Tmax) is reduced in human PH-HF. Here, we reveal that this decline is due to more myosin heads in the super-relaxed (SRX) state which lack force-production due to low ATP turnover.Methods:Frozen RV myocardium from 25 PH-HF and 9 control patients were permeabilized and myocyte mechanics and small angle X-ray diffraction measured.Results:RV Tmax(16±4 vs. 23±3 mN/mm2, p<1x10-15) and peak power (0.06±0.01 vs. 0.11±0.03, p=7.9x10-5) were 50% lower in HF. Unsupervised machine learning (infinite mixture Dirichlet process) using mechanics identified two HF subgroups distinguished by RV Tmax: Group 1 with slightly reduced Tmax(19±1 mN/mm2, p=3.8x10-6) and Group 2 with 40% reduced Tmax(14±3 mN/mm2, p<1x10-15), with Group 2 exhibiting clinically overt RV failure vs. Group 1 (right atrial pressure, RAP 14±6 vs. 9±4, p=0.01; RAP/pulmonary capillary wedge ratio 0.7±0.3 vs. 0.4±0.2, p=0.047). To test if low RV Tmaxwas associated with more SRX myosin, X-ray diffraction patterns from RV myofibers were used to determine the equatorial intensity ratio (I1,1/I1,0), an indicator of myosin proximity to the thick filament backbone. The ratio was lower (more SRX), which was associated with reduced myosin ATP turnover in Group 2 vs. both 1 (p=9.0x10-7) and controls (p=1.8x10-9). We then tested the effect of SRX on Tmaxby perturbing the proportion of SRX myosin using mavacamten (M), a myosin inhibitor that stabilizes the SRX state, and deoxy-ATP (dATP), a nucleotide that destabilizes SRX. In group 2, dATP augmented Tmaxby 30% (p=0.0002) while M had no effect (p=0.1); meanwhile in group 1, M suppressed Tmaxby 30% (p=0.0002) while dATP had no effect (p=0.9). Finally, a myosin activator EMD-57033, which increases probability of actin-myosin binding, augmented Tmax(correlation between ΔSRX-ΔTmaxr=-0.55, p=0.01).Conclusion:We find increased SRX myosin contributes to RV myofilament and chamber-level failure in PH-HF. Small molecules that recruit SRX myosin may have therapeutic benefit in patients with RV dysfunction.

Myocyte disarray is a hallmark of many cardiac disorders. However, the relationship between alterations in the orientation of individual myofibrils and myofilaments to disease progression has been largely underexplored. This oversight has predominantly been due to a paucity of methods for objective and quantitative analysis. Here we introduce a novel, less-biased approach to quantify myofibrillar and myofilament orientation in cardiac muscle under near physiological conditions and demonstrate its superiority as compared to conventional histological assessments. Using small-angle X-ray diffraction, we first investigated changes in myofibrillar orientation at increasing sarcomere lengths in permeabilized, relaxed, wildtype mouse myocardium from the left ventricle by assessing the angular spread of the 1,0 equatorial reflection (angle σ). At a sarcomere length (SL) of 1.9 μm, the angle σ was 0.23±0.01 rad, decreased to 0.19±0.01 rad at a SL of 2.1 μm, and further decreased to 0.15±0.01 rad at a SL of 2.3 μm (p<0.0001). Angle σ was significantly larger in R403Q, a MYH7 hypertrophic cardiomyopathy (HCM) model, porcine myocardium (0.24±0.01 rad) compared to WT myocardium (0.14±0.005 rad, p<0.0001) as well as in human heart failure tissue (0.19±0.006 rad) when compared to non-failing samples (0.17±0.007 rad, p=0.01). These data indicate that diseased myocardium suffers from greater myofibrillar disorientation compared to healthy controls. Finally, we showed that conventional, histology-based analysis of disarray can be subject to user bias and/or sampling error and lead to false positives. Our method for directly assessing myofibrillar orientation avoids the artifacts introduced by conventional histological approaches that assess myocyte orientation and only indirectly evaluate myofibrillar orientation, and provides a precise and objective metric for phenotypically characterizing myocardium. The ability to obtain excellent X-ray diffraction patterns from frozen human myocardium provides a new tool for investigating structural anomalies associated with cardiac diseases.

We used two-dimensional echocardiographic speckle-tracking to investigate whether left and right atrial (LA and RA) phasic function in pediatric heart transplantation (HT) patients is altered and explored the relationship to HT-related clinical variables.

Introduction:Myocardial fibrosis is associated with worsening left ventricular (LV) systolic function and has been found to increase with age. Clinical assessment of myocardial replacement fibrosis (scar) is routinely performed by means of late gadolinium enhanced MRI (LGE).Hypothesis:We hypothesize that a novel convolution neural network (CNN) framework with loss function initialization can be used for accurate and reproducible automatic quantification of LV myocardium and scar.Methods:This study included 164 3D LGE short axis studies (10-16 images/study) across 40 sites. LV myocardium and ischemic scar were segmented by one expert reader on the MEDIS QMass software. 3D U-Nets were trained for segmentation of the LV myocardium and scar tissue and binary segmentation of the LV myocardium on 2910 studies, augmented from 97 unique studies. Weights for U-Net were initialized with weighted categorical cross entropy (WCCE), which penalizes pixels that were misclassified as background, and then trained with categorical cross entropy (CCE). The network was evaluated on a validation and test set of 34 and 33 studies. Accuracy was assessed with the dice coefficient, intraclass correlation coefficients (ICC), and linear regression for LV myocardial mass and percent LV scar.Results:The network was found to converge after 35 epochs, with 25 epochs on WCCE and 10 epochs on CCE. Mean dice coefficient was found to be 0.580 ?0.110 for multi-class segmentation and0.671 ?0.091 for binary segmentation. ICC for LV mass was found to be 0.8299 (?= 0.9583) and 0.8504 (?= 1.064) for multi-class and binary segmentation methods (Figure 1). ICC for scar percentage in multi-class segmentation was found to be 0.5364 (?= 0.9815).Conclusions:In conclusion, 3D U-Net segmentation with WCCE initialization accurately segments LV myocardium and also demonstrates moderate success for LV ischemic scar quantification.

Introduction:Cardiovascular magnetic resonance (CMR) is used for clinical surveillance in tetralogy of Fallot (TOF) to assess right heart size, pulmonary regurgitation, and aortic size and function. However, application of deep learning for CMR analysis has been limited in this setting.Hypothesis:We hypothesized that application of transfer learning, a method of deep learning in which the network is previously trained on similar images, will allow for automated segmentation of TOF CMR studies despite variant anatomy.Methods:In total,3017 TOF images from 100 CMR studies of the prospective German Competence Network were included. Phase contrast ascending (AA) and descending (DA) aorta studies at the pulmonary artery bifurcation were segmented utilizing the ArtFUN software. Train, validation, and test datasets were 18400 images augmented from 1840 (62 studies), 590 images (20 studies), and 587 images (20 studies), respectively. Two 2D U-Nets previously trained on 24280 images (1214 participants) on the Multi-Ethnic Study for Atherosclerosis (MESA) dataset were trained on the TOF dataset using transfer learning. Segmentation accuracy was assessed using a dice coefficient, intraclass correlation coefficient (ICC) for AA and DA area and strain.Results:Transfer learning resulted in a dice coefficient of 0.86 ?0.16. ICC for AA and DA area were 0.7984 and 0.8115, respectively. ICC for AA and DA systolic area between DL and manual segmentations in the test set were found to be 0.8529 and 0.8508 respectively. ICC for AA and DA strain between DL and manual segmentations in the test set were found to be 0.8848 and 0.8482.Conclusions:In conclusion, transfer learning was found to be successful for automated segmentation and quantification of aortic mechanics in TOF patients. Transfer learning shows promise as a novel application of deep learning in congenital heart disease where larger datasets are unavailable for training.

Introduction:Direct measurement of central aortic stiffness by means of aortic distensibility has been shown to be an early marker of subclinical vascular alteration. However, current methods to measure aortic area from cardiovascular magnetic resonance (CMR) require manual intervention.Hypothesis:We hypothesized that application of 2D U-Net for segmentation of the ascending (AA) and descending (DA) aorta will have comparable performance with user-dependent semi-automated analysis.Methods:This study included 2D PC-cine CMR of the AA and DA at the pulmonary artery bifurcation from the MESA study across 6 sites. Four CNNs were trained with the U-Net architecture on 24282 unique images from 1123 MESA participants, namely (1) binary segmentation, (2) binary segmentation on augmented training data, (3) multi-class segmentation, (4) multi-class segmentation on augmented training data. CNN performance was evaluated utilizing a dice coefficient. Aortic flow parameters were determined utilizing the ArtFUN software and were compared utilizing the concordance-correlation coefficient (CCC), linear regression, and Bland-Altman analysis.Results:Dice coefficients were found to 0.9732 (CI: 0.9726-0.9739), 0.9760 (CI: 0.9752-0.9767), 0.9709 (CI: 0.9703-0.9716), and 0.9755 (CI: 0.9747-0.9762) for the binary segmentation model, the binary segmentation model with augmented training data set, the multi-class segmentation model, and the multi-class segmentation with augmented training data set, respectively. Aortic flow parameters from deep learning segmentation maps were well correlated with those from manual segmentation maps as assessed by linear regression and Bland-Altman analysis.Conclusions:In conclusion, the present study demonstrates uniformly successful ascending and descending aortic segmentation and aortic flow parameter quantification utilizing U-Net.