Association between echocardiography-derived haemodynamic force parameters and left ventricular reverse remodelling after cardiac resynchronization therapy.

AIMS: Cardiac resynchronization therapy (CRT) may induce left ventricular (LV) reverse remodelling (=LV response) in patients with heart failure. Intraventricular pressure gradients can be quantified using echocardiography-derived haemodynamic forces (HDF). The aim was to evaluate the association between baseline HDF and LV response and to compare the change of HDF after CRT between LV responders and LV non-responders. METHODS AND RESULTS: The following HDF parameters were assessed: 1)apical-basal (AB) strength, 2)lateral-septal strength, 3)force vector angle, 4)systolic AB impulse, 5)systolic force vector angle. LV response was defined as a reduction of LV end-systolic volume ≥15% at six months. One hundred ninety-six patients were included (64±11 years, 122(62%) men), 136(69%) showed LV response. On multivariable logistic regression analysis, the force vector angle in the complete heart cycle (OR 1.083 (95%CI 1.018, 1.153), p=0.012) and the systolic force vector angle (OR 1.089 (95%CI 1.021, 1.161), p=0.009), both included in separate models, were independently associated with LV response. Six months after CRT, LV responders had greater AB strength, AB impulse and higher force vector angles, while LV non-responders only showed improvement in the force vector angle in the complete heart cycle. CONCLUSION: The orientation of HDF at baseline is associated with LV response to CRT. Six months after CRT, the orientation of HDF improves in LV responders and LV non-responders, while the magnitude of AB HDF only improves in LV responders.

Evolution of Echocardiography-Derived Hemodynamic Force Parameters After Cardiac Resynchronization Therapy.

Echocardiography-derived hemodynamic forces (HDF) allow calculation of intraventricular pressure gradients from routine transthoracic echocardiographic images. The evolution of HDF after cardiac resynchronization therapy (CRT) has not been investigated in large cohorts. The aim was to assess HDF in patients with heart failure implanted with CRT versus healthy controls. HDF were assessed before and 6 months after CRT. The following HDF parameters were calculated: (1) apical-basal strength, (2) lateral-septal strength, (3) the ratio of lateral-septal to apical-basal strength ratio, and (4) the force vector angle (1 and 2 representing the magnitude of HDF, 3 and 4 representing the orientation of HDF). In the propulsive phase of systole, the apical-basal impulse and the systolic force vector angle were measured. A total of 197 patients were included (age 64 ± 11 years, 62% male), with left ventricular ejection fraction ≤35%, QRS duration ≥130 ms and left bundle branch block. The magnitude of HDF was significantly lower and the orientation was significantly worse in patients with heart failure versus healthy controls. Immediately after CRT implantation, the apical-basal impulse and systolic force vector angle were significantly increased. Six months after CRT, improvement of apical-basal strength, lateral-septal to apical-basal strength ratio and the force vector angle occurred. When CRT was deactivated at 6 months, the increase in the magnitude of apical-basal HDF remained unchanged while the systolic force vector angle worsened significantly. In conclusion, HDF in CRT recipients reflect the acute effect of CRT and the effect of left ventricular reverse remodeling on intraventricular pressure gradients. Whether HDF analysis provides incremental value over established echocardiographic parameters, remains to be determined.

Interpretation and further development of the hypnodensity representation of sleep structure.

Objective.The recently-introduced hypnodensity graph provides a probability distribution over sleep stages per data window (i.e. an epoch). This work explored whether this representation reveals continuities that can only be attributed to intra- and inter-rater disagreement of expert scorings, or also to co-occurrence of sleep stage-dependent features within one epoch.Approach.We proposed a simplified model for time series like the ones measured during sleep, and a second model to describe the annotation process by an expert. Generating data according to these models, enabled controlled experiments to investigate the interpretation of the hypnodensity graph. Moreover, the influence of both the supervised training strategy, and the used softmax non-linearity were investigated. Polysomnography recordings of 96 healthy sleepers (of which 11 were used as independent test set), were subsequently used to transfer conclusions to real data.Main results.A hypnodensity graph, predicted by a supervised neural classifier, represents the probability with which the sleep expert(s) assigned a label to an epoch. It thus reflects annotator behavior, and is thereby only indirectly linked to the ratio of sleep stage-dependent features in the epoch. Unsupervised training was shown to result in hypnodensity graph that were slightly less dependent on this annotation process, resulting in, on average, higher-entropy distributions over sleep stages (Hunsupervised= 0.41 versusHsupervised= 0.29). Moreover, pre-softmax predictions were, for both training strategies, found to better reflect the ratio of sleep stage-dependent characteristics in an epoch, as compared to the post-softmax counterparts (i.e. the hypnodensity graph). In real data, this was observed from the linear relation between pre-softmax N3 predictions and the amount of delta power.Significance.This study provides insights in, and proposes new, representations of sleep that may enhance our comprehension about sleep and sleep disorders.

Introduction to hemodynamic forces by echocardiography.

Hemodynamic force (HDF) analysis represents a novel approach to quantify intraventricular pressure gradients, responsible for blood flow. A new mathematical model allows the derivation of HDF parameters from routine transthoracic echocardiography, making this tool more accessible for clinical use. HDF analysis is considered the fluid dynamics correlate of deformation imaging and may be even more sensitive to detect mechanical abnormalities. This has the potential to add incremental clinical value, allowing earlier detection of pathology or immediate evaluation of response to treatment. In this article, the theoretical background and physiological patterns of HDF in the left ventricle are provided. In pathological situations, the HDF pattern might alter, which is illustrated with a case of ST segment elevation myocardial infarction and non-ischemic cardiomyopathy with typical left bundle branch block.

Automatic sleep scoring with LSTM networks: impact of time granularity and input signals.

Supervised automatic sleep scoring algorithms are usually trained using sleep stage labels manually annotated on 30 s epochs of PSG data. In this study, we investigate the impact of using shorter epochs with various PSG input signals for training and testing a Long Short Term Memory (LSTM) neural network. An LSTM model is evaluated on the provided 30 s epoch sleep stage labels from a publicly available dataset, as well as on 10 s subdivisions. Additionally, three independent scorers re-labeled a subset of the dataset on shorter time windows. The automatic sleep scoring experiments were repeated on the re-annotated subset.The highest performance is achieved on features extracted from 30 s epochs of a single channel frontal EEG. The resulting accuracy, precision and recall were of 92.22%, 67.58% and 66.00% respectively. When using a shorter epoch as input, the performance decreased by approximately 20%. Re-annotating a subset of the dataset on shorter time epochs did not improve the results and further altered the sleep stage detection performance. Our results show that our feature-based LSTM classification algorithm performs better on 30 s PSG epochs when compared to 10 s epochs used as input. Future work could be oriented to determining whether varying the epoch size improves classification outcomes for different types of classification algorithms.

Dimensionality reduction for EEG-based sleep stage detection: comparison of autoencoders, principal component analysis and factor analysis.

Methods developed for automatic sleep stage detection make use of large amounts of data in the form of polysomnographic (PSG) recordings to build predictive models. In this study, we investigate the effect of several dimensionality reduction techniques, i.e., principal component analysis (PCA), factor analysis (FA), and autoencoders (AE) on common classifiers, e.g., random forests (RF), multilayer perceptron (MLP), long-short term memory (LSTM) networks, for automated sleep stage detection. Experimental testing is carried out on the MGH Dataset provided in the "You Snooze, You Win: The PhysioNet/Computing in Cardiology Challenge 2018". The signals used as input are the six available (EEG) electoencephalographic channels and combinations with the other PSG signals provided: ECG - electrocardiogram, EMG - electromyogram, respiration based signals - respiratory efforts and airflow. We observe that a similar or improved accuracy is obtained in most cases when using all dimensionality reduction techniques, which is a promising result as it allows to reduce the computational load while maintaining performance and in some cases also improves the accuracy of automated sleep stage detection. In our study, using autoencoders for dimensionality reduction maintains the performance of the model, while using PCA and FA the accuracy of the models is in most cases improved.

Automatic Sleep Stage Detection: A Study on the Influence of Various PSG Input Signals.

Automatic sleep stage detection can be performed using a variety of input signals from a polysomnographic (PSG) recording. In this study, we investigate the effect of different input signals on the performance of feature-based automatic sleep stage classification algorithms with both a Random Forest (RF) and Multilayer Perceptron (MLP) classifier. Combinations of the EEG (electroencephalographic) signal and ECG (electrocardiographic), EMG (electromyographic) and respiratory signals as input are investigated as input with respect to using single channel and multi-channel EEG as input. The Physionet "You Snooze, You Win" dataset is used for the study. The RF classifier consistently outperforms our MLP implementation in all cases and is positively affected by specific signal combinations. The overall classification performance using a single channel EEG is high (an accuracy, precision and recall of 86.91 %, 89.52%, 86.91% respectively) using RF. The results are comparable to the performance obtained using six EEG channels as input. Adding respiratory signals to the inputs processed by RF increases the N2 stage detection performance with 20%, while adding the EMG signal improves the accuracy of the REM stage detection with 5%. Our analysis shows that adding specific signals as input to RF improves the accuracy of specific sleep stages and increases the overall performance. Using a combination of EEG and respiratory signals we achieved an accuracy of 93% for the RF classifier.

Assessment and comparison of left ventricular shear in normal and situs inversus totalis hearts by means of magnetic resonance tagging.

Situs inversus totalis (SIT) is characterized by complete mirroring of gross cardiac anatomy and position combined with an incompletely mirrored myofiber arrangement, being normal at the apex but inverted at the base of the left ventricle (LV). This study relates myocardial structure to mechanical function by analyzing and comparing myocardial deformation patterns of normal and SIT subjects, focusing especially on circumferential-radial shear. In nine control and nine SIT normotensive human subjects, myocardial deformation was assessed from magnetic resonance tagging (MRT) image sequences of five LV short-axis slices. During ejection, no significant difference in either circumferential shortening (εcc) or its axial gradient (Δεcc) is found between corresponding LV levels in control and SIT hearts. Circumferential-radial shear (εcr) has a clear linear trend from apex-to-base in controls, while in SIT it hovers close to zero at all levels. Torsion as well as axial change in εcr (Δεcr) is as in controls in apical sections of SIT hearts but deviates significantly towards the base, changing sign close to the LV equator. Interindividual variability in torsion and Δεcr values is higher in SIT than in controls. Apex-to-base trends of torsion and Δεcr in SIT, changing sign near the LV equator, further substantiate a structural transition in myofiber arrangement close to the LV equator itself. Invariance of εcc and Δεcc patterns between controls and SIT subjects shows that normal LV pump function is achieved in SIT despite partial mirroring of myocardial structure leading to torsional and shear patterns that are far from normality.

Why SIT works: normal function despite typical myofiber pattern in Situs Inversus Totalis (SIT) hearts derived by shear-induced myofiber reorientation.

The left ventricle (LV) of mammals with Situs Solitus (SS, normal organ arrangement) displays hardly any interindividual variation in myofiber pattern and experimentally determined torsion. SS LV myofiber pattern has been suggested to result from adaptive myofiber reorientation, in turn leading to efficient pump and myofiber function. Limited data from the Situs Inversus Totalis (SIT, a complete mirror image of organ anatomy and position) LV demonstrated an essential different myofiber pattern, being normal at the apex but mirrored at the base. Considerable differences in torsion patterns in between human SIT LVs even suggest variation in myofiber pattern among SIT LVs themselves. We addressed whether different myofiber patterns in the SIT LV can be predicted by adaptive myofiber reorientation and whether they yield similar pump and myofiber function as in the SS LV. With a mathematical model of LV mechanics including shear induced myofiber reorientation, we predicted myofiber patterns of one SS and three different SIT LVs. Initial conditions for SIT were based on scarce information on the helix angle. The transverse angle was set to zero. During reorientation, a non-zero transverse angle developed, pump function increased, and myofiber function increased and became more homogeneous. Three continuous SIT structures emerged with a different location of transition between normal and mirrored myofiber orientation pattern. Predicted SIT torsion patterns matched experimentally determined ones. Pump and myofiber function in SIT and SS LVs are similar, despite essential differences in myocardial structure. SS and SIT LV structure and function may originate from same processes of adaptive myofiber reorientation.

Mapping displacement and deformation of the heart with local sine-wave modeling.

The new SinMod method extracts motion from magnetic resonance imaging (MRI)-tagged (MRIT) image sequences. Image intensity in the environment of each pixel is modeled as a moving sine wavefront. Displacement is estimated at subpixel accuracy. Performance is compared with the harmonic-phase analysis (HARP) method, which is currently the most common method used to detect motion in MRIT images. SinMod can handle line tags, as well as speckle patterns. In artificial images (tag distance six pixels), SinMod detects displacements accurately (error < 0.02 pixels). Effects of noise are suppressed effectively. Sharp transitions in motion at the boundary of an object are smeared out over a width of 0.6 tag distance. For MRIT images of the heart, SinMod appears less sensitive to artifacts, especially later in the cardiac cycle when image quality deteriorates. For each pixel, the quality of the sine-wave model in describing local image intensity is quantified objectively. If local quality is low, artifacts are avoided by averaging motion over a larger environment. Summarizing, SinMod is just as fast as HARP, but it performs better with respect to accuracy of displacement detection, noise reduction, and avoidance of artifacts.

Automatic localization of intimal and adventitial carotid artery layers with noninvasive ultrasound: a novel algorithm providing scan quality control.

Transcutaneous ultrasound measurements of common carotid artery (CCA) diameter and intima-media thickness (IMT) give insight on arterial dynamics and anatomy, both correlating well with atherosclerosis and risk of cardiovascular disease. We propose a novel automatic algorithm to estimate CCA diameter and IMT in ultrasound (US) images, based on separate analysis of anterior and posterior CCA walls and able to distinguish internal (intima-intima) and external (adventitia-adventitia) diameter. The method combines off-line signal- and image-processing techniques to accommodate echo images acquired at a frame rate of 30 Hz and composed directly from RF data, circumventing digital video-grabbing. Segmentation consists of automatic CCA recognition, followed by adventitial delineation performed with a sustain-attack filter with exponentially decaying reference functions. Intimal delineation is then based on the multiscale anisotropic barycenter (MAB), which is an extension of a known delineation method involving the "first order absolute central moment" of the echo amplitude. An automatic measure of the quality of the US beam incidence for each wall is superimosed on the CCA contour overlays for visual feedback. Validation is carried out on 36 US CCA acquisitions from 12 healthy volunteers, as well as on synthetic US images. Results indicate good accuracy on synthetic US images (within 1.3% for diameter and 3% for IMT). The in vivo intra-recording beat-to-beat variations are on average lower than 50 microm for external diameter and IMT, and lower than 100 microm for internal diameter. A comparison with a commercial device (ART.LAB system) shows that the proposed algorithm performs better in terms of inter-recording precision. The beam incidence control significantly improves the repeatability of IMT estimates, and motivates sonographers actively to maintain a proper scan plane throughout the acquisition to minimize the incidence of confounding factors. The method is clinically viable, providing robust estimates of CCA internal and external diameter and IMT waveforms for both CCA walls, even at a low B-mode update rate of 30 Hz.

Nonlinear processing in B-mode ultrasound affects carotid diameter assessment.

Noninvasive diameter assessment in the common carotid artery (CCA) by means of ultrasound is a useful technique for estimation of arterial mechanical and dynamic properties, clinical screening and treatment monitoring. Before presentation on screen, ultrasound images are subjected to nonlinear processing, e.g., logarithmic compression and noise-level thresholding, to improve visualization. In addition, signal saturation may occur, either in the received radiofrequency (RF) signals or in their envelopes. The objective of this study is to evaluate the effect of signal nonlinearities on CCA diameter measurements by means of noninvasive B-mode ultrasound, comparing the performance of two different edge detectors. In 14 healthy subjects, three repeated ultrasonic acquisitions (6 s) without saturation were performed. The acquired RF signals were subjected off-line to envelope detection, logarithmic compression and various degrees of saturation applied to the signals before or after envelope detection. For the purpose of CCA diameter estimation, artery walls were automatically outlined frame by frame. As automatic edge detectors, we considered the sustain attack filter (SAF), based on exponentially decaying reference functions, and a derivative approach (DER), relying on the positions of first derivative maxima. Both methods are applied within a region-of-interest located on the CCA. No regularization of the detected wall positions by means of pre- or postprocessing is presently applied to directly relate the outcome of the edge detectors to the applied nonlinear processing. Diameter values assessed with SAF are unaffected by logarithmic compression because of the possibility to integrate the compression characteristic of the ultrasound system into the method. The estimated diameters values obtained with DER instead show differences in the order of 10% because of compression. Saturation affects DER more than SAF; DER exhibits larger intrarecording and intrasubject variations in the estimated diameter values. Therefore, SAF gives more precise and robust CCA diameter estimates than DER, and is more suited for integration in algorithms meant for vascular ultrasound image segmentation. This study demonstrates the relevant effects of nonlinearities such as saturation and logarithmic compression on the quality of noninvasive US CCA diameter measurements.

Automatic recognition of the common carotid artery in longitudinal ultrasound B-mode scans.

Many morphological and dynamic properties of the common carotid artery (CCA), e.g. lumen diameter, distension and wall thickness, can be measured non-invasively with ultrasound (US) techniques. As common to other medical image segmentation processes, this requires as a preliminary step the manual recognition of the artery of interest within the ultrasound image. In real-time US imaging, such manual initialization procedure interferes with the difficult task of the sonographer to select and maintain a proper image scan plane. Even for off-line US segmentation the requirement for human supervision and interaction precludes full automation. To eliminate user interference and to speed up processing for both real-time and off-line applications, we developed an algorithm for the automatic artery recognition in longitudinal US scans of the CCA. It acts directly on the envelopes of received radio frequency echo signals, eventually composing the ultrasound image. In order to properly exploit the information content of the arterial structure the envelopes are decimated, according to the two-dimensional resolution characteristics of the echo system, thereby substantially decreasing computational load. Subsequently, based upon the expected diameter range and a priori knowledge of the typical pattern in the echo envelope of the arterial wall-lumen complex, parametrical template matching is performed, resulting in the location of the lumen position along each echo line considered. Finally, in order to reject incorrect estimates, a spatial and temporal clustering method is applied. Adequate values for the parameters involved in the processing are obtained via off-line testing of the proposed algorithm on 128 echo data recordings from 45 subjects. Using those robust parameter values, correct and fast recognition of the artery is achieved in more than 98% of the 6185 processed frames. Since these results are obtained via rigorous data decimation and using a cascade of rather simple steps, the proposed automatic algorithm is suitable for real-time recognition of the CCA.