Impact of late gadolinium enhancement image acquisition resolution on neural network based automatic scar segmentation.

BACKGROUND: Automatic myocardial scar segmentation from late gadolinium enhancement (LGE) images using neural networks promises an alternative to time-consuming and observer-dependent semi-automatic approaches. However, alterations in data acquisition, reconstruction as well as post-processing may compromise network performance. The objective of the present work was to systematically assess network performance degradation due to a mismatch of point-spread function between training and testing data. METHODS: Thirty-six high-resolution (0.7×0.7×2.0 mm3) LGE k-space datasets were acquired post-mortem in porcine models of myocardial infarction. The in-plane point-spread function and hence in-plane resolution Δx was retrospectively degraded using k-space lowpass filtering, while field-of-view and matrix size were kept constant. Manual segmentation of the left ventricle (LV) and healthy remote myocardium was performed to quantify location and area (% of myocardium) of scar by thresholding (≥ SD5 above remote). Three standard U-Nets were trained on training resolutions Δxtrain = 0.7, 1.2 and 1.7 mm to predict endo- and epicardial borders of LV myocardium and scar. The scar prediction of the three networks for varying test resolutions (Δxtest = 0.7 to 1.7 mm) was compared against the reference SD5 thresholding at 0.7 mm. Finally, a fourth network trained on a combination of resolutions (Δxtrain = 0.7 to 1.7 mm) was tested. RESULTS: The prediction of relative scar areas showed the highest precision when the resolution of the test data was identical to or close to the resolution used during training. The median fractional scar errors and precisions (IQR) from networks trained and tested on the same resolution were 0.0 percentage points (p.p.) (1.24 - 1.45), and - 0.5 - 0.0 p.p. (2.00 - 3.25) for networks trained and tested on the most differing resolutions, respectively. Deploying the network trained on multiple resolutions resulted in reduced resolution dependency with median scar errors and IQRs of 0.0 p.p. (1.24 - 1.69) for all investigated test resolutions. CONCLUSION: A mismatch of the imaging point-spread function between training and test data can lead to degradation of scar segmentation when using current U-Net architectures as demonstrated on LGE porcine myocardial infarction data. Training networks on multi-resolution data can alleviate the resolution dependency.

CMRsim-A python package for cardiovascular MR simulations incorporating complex motion and flow.

PURPOSE: To present an open-source MR simulation framework that facilitates the incorporation of complex motion and flow for studying cardiovascular MR (CMR) acquisition and reconstruction. METHODS: CMRsim is a Python package that allows simulation of CMR images using dynamic digital phantoms with complex motion as input. Two simulation paradigms are available, namely, numerical and analytical solutions to the Bloch equations, using a common motion representation. Competitive simulation speeds are achieved using TensorFlow for GPU acceleration. To demonstrate the capability of the package, one introductory and two advanced CMR simulation experiments are presented. The latter showcase phase-contrast imaging of turbulent flow downstream of a stenotic section and cardiac diffusion tensor imaging on a contracting left ventricle. Additionally, extensive documentation and example resources are provided. RESULTS: The Bloch simulation with turbulent flow using approximately 1.5 million particles and a sequence duration of 710 ms for each of the seven different velocity encodings took a total of 29 min on a NVIDIA Titan RTX GPU. The results show characteristic phase contrast and magnitude modulation present in real data. The analytical simulation of cardiac diffusion tensor imaging with bulk-motion phase sensitivity took approximately 10 s per diffusion-weighted image, including preparation and loading steps. The results exhibit the expected alteration of diffusion metrics due to strain. CONCLUSION: CMRsim is the first simulation framework that allows one to feasibly incorporate complex motion, including turbulent flow, to systematically study advanced CMR acquisition and reconstruction approaches. The open-source package features modularity and transparency, facilitating maintainability and extensibility in support of reproducible research.

Synthetically trained convolutional neural networks for improved tensor estimation from free-breathing cardiac DTI.

Cardiac diffusion tensor imaging (cDTI) provides invaluable information about the state of myocardial microstructure. For further clinical dissemination, free-breathing acquisitions are desired, which however require image registration prior to tensor estimation. Due to the varying contrast and the intrinsically low signal-to-noise ratio (SNR), registration is very challenging and thus can introduce additional errors in the tensor estimation. In the work at hand it is hypothesized, that by incorporating spatial information and physiologically plausible priors into the fitting algorithm, the robustness of diffusion tensor estimation can be improved. To this end, we present a parameterized pipeline to generate synthetic data, that captures the statistics including spatial correlations of diffusion tensors and motion of the heart. The synthetic data is used to train a residual convolutional neural network (CNN) to estimate diffusion tensors from unregistered in-vivo cDTI data. Using in-silico data, the synthetically trained CNN is demonstrated to yield increased tensor estimation accuracy and precision when compared to conventional registration followed by least squares fitting. The network outputs fewer outliers especially at the myocardial borders. In-vivo feasibility using data from five healthy subjects demonstrates the utility of the synthetically trained network. The in-vivo results predicted by the synthetically trained CNN are found to be consistent with the registered least-squares estimates while showing fewer outliers and reduced noise. Even in low SNR regimes, the network results in robust tensor estimation, enabling scan time reduction by reduced-average acquisition in-vivo. Finally, to investigate the network's capability of discriminating between healthy and lesioned tissue, the in-vivo data was artificially augmented showing preserved classification of tissue states based on diffusion metrics.

Water-Mediated Protein-Protein Interactions at High Pressures are Controlled by a Deep-Sea Osmolyte.

The influence of natural cosolvent mixtures on the pressure-dependent structure and protein-protein interaction potential of dense protein solutions is studied and analyzed using small-angle X-ray scattering in combination with a liquid-state theoretical approach. The deep-sea osmolyte trimethylamine-N-oxide is shown to play a crucial and singular role in its ability to not only guarantee sustainability of the native protein's folded state under harsh environmental conditions, but it also controls water-mediated intermolecular interactions at high pressure, thereby preventing contact formation and hence aggregation of proteins.

Phase behavior of lysozyme solutions in the liquid-liquid phase coexistence region at high hydrostatic pressures.

We present results from small-angle X-ray scattering and turbidity measurements on the effect of high hydrostatic pressure on the phase behavior of dense lysozyme solutions in the liquid-liquid phase separation region, and characterize the underlying intermolecular protein-protein interactions as a function of temperature and pressure under charge-screening conditions (0.5 M NaCl). A reentrant liquid-liquid phase separation region is observed at elevated pressures, which may originate in the pressure dependence of the solvent-mediated protein-protein interaction. A temperature-pressure-concentration phase diagram was constructed for highly concentrated lysozyme solutions over a wide range of temperatures, pressures and protein concentrations including the critical region of the liquid-liquid miscibility gap.