Cardiac Valve Event Timing in Echocardiography Using Deep Learning and Triplane Recordings.

Cardiac valve event timing plays a crucial role when conducting clinical measurements using echocardiography. However, established automated approaches are limited by the need of external electrocardiogram sensors, and manual measurements often rely on timing from different cardiac cycles. Recent methods have applied deep learning to cardiac timing, but they have mainly been restricted to only detecting two key time points, namely end-diastole (ED) and end-systole (ES). In this work, we propose a deep learning approach that leverages triplane recordings to enhance detection of valve events in echocardiography. Our method demonstrates improved performance detecting six different events, including valve events conventionally associated with ED and ES. Of all events, we achieve an average absolute frame difference (aFD) of maximum 1.4 frames (29 ms) for start of diastasis, down to 0.6 frames (12 ms) for mitral valve opening when performing a ten-fold cross-validation with test splits on triplane data from 240 patients. On an external independent test consisting of apical long-axis data from 180 other patients, the worst performing event detection had an aFD of 1.8 (30 ms). The proposed approach has the potential to significantly impact clinical practice by enabling more accurate, rapid and comprehensive event detection, leading to improved clinical measurements.

Automated Segmentation and Quantification of the Right Ventricle in 2-D Echocardiography.

OBJECTIVE: The right ventricle receives less attention than its left counterpart in echocardiography research, practice and development of automated solutions. In the work described here, we sought to determine that the deep learning methods for automated segmentation of the left ventricle in 2-D echocardiograms are also valid for the right ventricle. Additionally, here we describe and explore a keypoint detection approach to segmentation that guards against erratic behavior often displayed by segmentation models. METHODS: We used a data set of echo images focused on the right ventricle from 250 participants to train and evaluate several deep learning models for segmentation and keypoint detection. We propose a compact architecture (U-Net KP) employing the latter approach. The architecture is designed to balance high speed with accuracy and robustness. RESULTS: All featured models achieved segmentation accuracy close to the inter-observer variability. When computing the metrics of right ventricular systolic function from contour predictions of U-Net KP, we obtained the bias and 95% limits of agreement of 0.8 ± 10.8% for the right ventricular fractional area change measurements, -0.04 ± 0.54 cm for the tricuspid annular plane systolic excursion measurements and 0.2 ± 6.6% for the right ventricular free wall strain measurements. These results were also comparable to the semi-automatically derived inter-observer discrepancies of 0.4 ± 11.8%, -0.37 ± 0.58 cm and -1.0 ± 7.7% for the aforementioned metrics, respectively. CONCLUSION: Given the appropriate data, automated segmentation and quantification of the right ventricle in 2-D echocardiography are feasible with existing methods. However, keypoint detection architectures may offer higher robustness and information density for the same computational cost.

Real-Time Echocardiography Guidance for Optimized Apical Standard Views.

Measurements of cardiac function such as left ventricular ejection fraction and myocardial strain are typically based on 2-D ultrasound imaging. The reliability of these measurements depends on the correct pose of the transducer such that the 2-D imaging plane properly aligns with the heart for standard measurement views and is thus dependent on the operator's skills. We propose a deep learning tool that suggests transducer movements to help users navigate toward the required standard views while scanning. The tool can simplify echocardiography for less experienced users and improve image standardization for more experienced users. Training data were generated by slicing 3-D ultrasound volumes, which permits simulation of the movements of a 2-D transducer. Neural networks were further trained to calculate the transducer position in a regression fashion. The method was validated and tested on 2-D images from several data sets representative of a prospective clinical setting. The method proposed the adequate transducer movement 75% of the time when averaging over all degrees of freedom and 95% of the time when considering transducer rotation solely. Real-time application examples illustrate the direct relation between the transducer movements, the ultrasound image and the provided feedback.

3D Myocardial Mechanical Wave Measurements: Toward In Vivo 3D Myocardial Elasticity Mapping.

OBJECTIVES: This study aimed to investigate the potential of a novel 3-dimensional (3D) mechanical wave velocity mapping technique, based on the natural mechanical waves produced by the heart itself, to approach a noninvasive 3D stiffness mapping of the left ventricle. BACKGROUND: Myocardial fibrosis is recognized as a pathophysiological substrate of major cardiovascular disorders such as cardiomyopathies and valvular heart disease. As fibrosis leads to increased myocardial stiffness, ultrasound elastography measurements could provide important clinical information. METHODS: A 3D high frame rate imaging sequence was implemented on a high-end clinical ultrasound scanner to achieve 820 volumes/s when gating over 4 consecutive cardiac cycles. Five healthy volunteers and 10 patients with various degrees of aortic stenosis were included to evaluate feasibility and reproducibility. Mechanical waves were detected using the novel Clutter Filter Wave Imaging approach, shown to be highly sensitive to the weak tissue displacements caused by natural mechanical waves. RESULTS: 3D spatiotemporal maps of mechanical wave velocities were produced for all subjects. Only the specific mechanical wave at atrial contraction provided a full 3D coverage of the left ventricle (LV). The average atrial kick propagation velocity was 1.6 ± 0.2 m/s in healthy volunteers and 2.8 ± 0.8 m/s in patients (p = 0.0016). A high correlation was found between mechanical wave velocity and age (R2 = 0.88, healthy group), septal wall thickness (R2 = 0.73, entire group), and peak jet velocity across the aortic valve (R2 = 0.70). For 3 of the patients, the higher mechanical wave velocity coexisted with the presence of late gadolinium enhancement on cardiac magnetic resonance. CONCLUSIONS: In this study, 3D LV mechanical wave velocities were visualized and measured in healthy volunteers and patients with aortic stenosis. The proposed imaging sequence and measurement technique allowed, for the first time, the measurement of full spatiotemporal 3D elasticity maps of the LV using ultrasound. (Ultrasonic markers for myocardial fibrosis and prognosis in aortic stenosis; NCT03422770).

User-Intended Doppler Measurement Type Prediction Combining CNNs With Smart Post-Processing.

Spectral Doppler measurements are an important part of the standard echocardiographic examination. These measurements give insight into myocardial motion and blood flow, providing clinicians with parameters for diagnostic decision making. Many of these measurements are performed automatically with high accuracy, increasing the efficiency of the diagnostic pipeline. However, full automation is not yet available because the user must manually select which measurement should be performed on each image. In this work, we develop a pipeline based on convolutional neural networks (CNNs) to automatically classify the measurement type from cardiac Doppler scans. We show how the multi-modal information in each spectral Doppler recording can be combined using a meta parameter post-processing mapping scheme and heatmaps to encode coordinate locations. Additionally, we experiment with several architectures to examine the tradeoff between accuracy, speed, and memory usage for resource-constrained environments. Finally, we propose a confidence metric using the values in the last fully connected layer of the network and show that our confidence metric can prevent many misclassifications. Our algorithm enables a fully automatic pipeline from acquisition to Doppler spectrum measurements. We achieve 96% accuracy on a test set drawn from separate clinical sites, indicating that the proposed method is suitable for clinical adoption.

Estimation of End-Diastole in Cardiac Spectral Doppler Using Deep Learning.

Electrocardiogram (ECG) is often used together with a spectral Doppler ultrasound to separate heart cycles by determining the end-diastole locations. However, the ECG signal is not always recorded. In such cases, the cardiac cycles can be estimated manually from the ultrasound data retrospectively. We present a deep learning-based method for automatic detection of the end-diastoles in spectral Doppler spectrograms. The method uses a combination of a convolutional neural network (CNN) for extracting features and a recurrent neural network (RNN) for modeling temporal relations. In echocardiography, there are three Doppler spectrogram modalities, continuous wave, pulsed wave, and tissue velocity Doppler. Both the training and test data sets include all three modalities. The model was tested on 643 spectrograms coming from different hospitals than in the training data set. For the purposes described in this work, a valid end-diastole detection is defined as a prediction being closer than 60 ms to the reference value. We will refer to these as true detections. Similarly, a prediction farther away is defined as nonvalid or false detections. The method automatically rejects spectrograms where the detection of an end-diastole has low confidence. When setting the algorithm to reject 1.9%, the method achieved 97.7% true detections with a mean error of 14 ms and had 2.5% false detections on the remaining spectrograms.

Clutter Filter Wave Imaging.

The elastic properties of human tissue can be evaluated through the study of mechanical wave propagation captured using high frame rate ultrasound imaging. Methods such as block-matching or phase-based motion estimation have been used to estimate the displacement induced by the mechanical waves. In this paper, a new method for detecting mechanical wave propagation without motion estimation is presented, where the motion of interest is accentuated by an appropriate clutter filter. Thus, the mechanical wave propagation will directly appear as bands of the attenuated signal moving in the B-mode sequence and corresponding anatomical M-mode images. While only the locality of tissue velocity induced by the mechanical wave is detected, it is shown that the method is more sensitive to subtle tissue displacements when compared to motion estimation techniques. The technique was evaluated for the propagation of the pulse wave in a carotid artery, mechanical waves on the left ventricle, and shear waves induced by radiation force on a tissue-mimicking phantom. The results were compared to tissue Doppler imaging (TDI) and demonstrated that clutter filter wave imaging (CFWI) was able to detect the mechanical wave propagating in tissue with a relative temporal and spatial resolution 30% higher and a relative consistency 40% higher than TDI. The results showed that CFWI was able to detect mechanical waves with a relative frequency content 40% higher than TDI in a shear wave imaging experiment.

Real-Time Standard View Classification in Transthoracic Echocardiography Using Convolutional Neural Networks.

Transthoracic echocardiography examinations are usually performed according to a protocol comprising different probe postures providing standard views of the heart. These are used as a basis when assessing cardiac function, and it is essential that the morphophysiological representations are correct. Clinical analysis is often initialized with the current view, and automatic classification can thus be useful in improving today's workflow. In this article, convolutional neural networks (CNNs) are used to create classification models predicting up to seven different cardiac views. Data sets of 2-D ultrasound acquired from studies totaling more than 500 patients and 7000 videos were included. State-of-the-art accuracies of 98.3% ± 0.6% and 98.9% ± 0.6% on single frames and sequences, respectively, and real-time performance with 4.4 ± 0.3 ms per frame were achieved. Further, it was found that CNNs have the potential for use in automatic multiplanar reformatting and orientation guidance. Using 3-D data to train models applicable for 2-D classification, we achieved a median deviation of 4° ± 3° from the optimal orientations.

Myocardial Strain Rate by Anatomic Doppler Spectrum: First Clinical Experience Using Retrospective Spectral Tissue Doppler from Ultra-High Frame Rate Imaging.

Strain rate imaging by tissue Doppler (TDI) is vulnerable to stationary reverberations and noise (clutter). Anatomic Doppler spectrum (ADS) presents retrospective spectral Doppler from ultra-high frame rate imaging (UFR-TDI) data for a region of interest, that is, ventricular wall or segment, at one time instance. This enables spectral assessment of strain rate (SR) without the influence of clutter. In this study, we assessed SR with ADS and conventional TDI in 20 patients with a recent myocardial infarction and 10 healthy volunteers. ADS-based SR correlated with fraction of scarred myocardium of the left ventricle (r = 0.68, p < 0.001), whereas SR by conventional TDI did not (r = 0.23, p = 0.30). ADS identified scarred myocardium and ADS Visual was the only method that differentiated transmural from non-transmural distribution of myocardial scar on a segmental level (p = 0.002). Finally, analysis of SR by ADS was feasible in a larger number of segments compared with SR by conventional TDI (p < 0.001).

Strain rate imaging combined with wall motion analysis gives incremental value in direct quantification of myocardial infarct size.

BACKGROUND: The study aimed to evaluate the diagnostic accuracy of a new method for direct echocardiographic quantification of the myocardial infarct size, using late enhancement magnetic resonance imaging (LE-MRI) as a reference method. METHODS AND RESULTS: Echocardiography and LE-MRI were performed on average 31 days after first-time myocardial infarction in 58 patients. Echocardiography was also performed on 35 healthy controls. Direct echocardiographic quantification of the infarct size was based on automated selection and quantification of areas with hypokinesia and akinesia from colour-coded strain rate data, with manual correction based on visual wall motion analysis. The left ventricular (LV) ejection fraction, speckle-tracking-based longitudinal global strain, wall motion score index (WMSI), longitudinal systolic motion and velocity, and the ratio of early mitral inflow velocity to mitral annular early diastolic velocity were also measured by echocardiography. The area under the receiver-operating characteristic curves for the identification of the infarct size >12% by LE-MRI was 0.84, using the new method for direct echocardiographic quantification of the infarct size. The new method showed significantly a higher correlation with the infarct size by LE-MRI both at the global (r = 0.81) and segmental (r = 0.59) level compared with other indices of LV function. CONCLUSION: Direct quantification of the percentage infarct size by strain rate imaging combined with wall motion analysis yields high diagnostic accuracy and better correlation to LE-MRI compared with other echocardiographic indices of global LV function. Echocardiography performed ~1 month after myocardial infarction showed ability to identify the patients with the infarct size >12%.

Echocardiography without electrocardiogram.

AIMS: automatic detection of the QRS complex on electrocardiogram (ECG) is used on cardiac ultrasound scanners to separate ultrasound image series into cardiac cycles for playback and storage. On small hand-held scanners it is unpractical to connect ECG cables. We therefore aim to do automatic cardiac cycle separation using apical B-mode ultrasound images. METHODS AND RESULTS: cardiac cycle length is estimated by cyclicity analysis of B-mode intensities. To determine a cycle start estimate near QRS, a deformable model is fitted to the left ventricle in real-time. The model is used to initialize and constrain a speckle tracker positioned near the mitral annulus. In the displacement curve generated by the speckle tracker, a time point near maximum distance from the probe is detected as a cardiac cycle start estimate. Validation against ECG was done on 233 recordings from normal subjects and 46 recordings from subjects with coronary pathology. Several test cases were run for each recording to emulate B-mode series starting at all time points in the cardiac cycle. Totally, 11 886 test cases were run. Cycle length estimation was feasible in 98% of normal subject cases and 91% of pathology cases. Median difference in cycle length by ECG was 0 and -3 ms, respectively. Cycle start estimation was feasible in 90% of normal subject cases and 77% of pathology cases. Median difference to cycle start by ECG was 62 and 76 ms, respectively. CONCLUSION: apical B-mode series can automatically be separated into cardiac cycles without using ECG.

Reproducibility in echocardiographic assessment of the left ventricular global and regional function, the HUNT study.

AIMS: The study aimed to compare the inter-observer reproducibility of new and traditional measurements of the left ventricular (LV) global and regional function. METHODS AND RESULTS: Two experienced echocardiographers performed 20 complete echo/Doppler examinations and 50 analyses on ten healthy subjects. All recordings were analysed for systolic and diastolic conventional and deformation measurements by both echocardiographers. Inter-observer mean error (absolute difference divided by the mean) was 4% and lowest (P = 0.001) for systolic M-mode annulus excursion. Mean error for the regional deformation indices was significantly higher than for all the global measurements (all P < 0.001). Mean error for analyses of the same recording was 34% (P = 0.002) lower for global systolic indices and 44% (P < 0.001) lower for global diastolic indices than inter-observer mean error for analyses made in separate recordings. CONCLUSION: Systolic M-mode annulus excursion showed better inter-observer reproducibility than other traditional and newer measurements of LV systolic and diastolic function. Repeated analyses of the same recordings underestimate the more clinically relevant inter-observer reproducibility by approximately 40% for most measurements of LV function.

Automatic timing of aortic valve closure in apical tissue Doppler images.

Ultrasound color tissue Doppler imaging (TDI) can be used to estimate velocities of moving left ventricular cardiac tissue. Aortic valve closure (AVC) can be observed as a notch in apical TDI velocity/time curves occurring after ejection, but before early relaxation. This work sought to evaluate automatic and automated algorithms using TDI for timing AVC. Mitral valve position and the time point of early relaxation were extracted and used to accomplish the task. To test the algorithms, phonocardiogram of the second heart sound was recorded simultaneously with TDI and used as a reference method. The algorithms were tested on apical views of 16 healthy subjects. In 98% of the cardiac cycles, the automatic algorithm estimated the time point of AVC within 25 ms of the reference. Automatic detection of AVC might save manual effort and provide a marker separating ejection and diastole for further automated analysis.

Automated analysis of strain rate and strain: feasibility and clinical implications.

BACKGROUND: This study evaluated 3 new automated methods, based on a combination of speckle tracking and tissue Doppler, for the analysis of strain rate (SR) and strain. Feasibility and values for peak systolic strain rate (SR s ) and end-systolic strain (S es ) were assessed. METHODS: Thirty patients with myocardial infarction and 30 normal subjects were examined. Customized software with automatic definition of segments was used for automated measurements. SR s and SR es were measured over each segment simultaneously and identified automatically. The study compared tissue Doppler-based SR and strain measurements without (method 1) and with segment tracking (method 2) to speckle tracking-based measurements (method 3). For tracking, speckle tracking and tissue Doppler were used in combination. Standard manual analysis was used as a reference. RESULTS: The automated analysis (16 segments, 3 apical views) required 2 minutes; manual analysis took 11 minutes. Accuracy was compared in 56 segments (28 mid-infarcted and 28 normal) from 28 patients and was 93.9% for method 1, 93.8% for method 2, 95.8% for method 3, and 96.2% for the manual method. In the normal group, mean SR s (0.27 s -1 ) was less with method 3 than with the other methods ( P < .001). CONCLUSIONS: Our findings indicate that automated analysis of SR and strain, with some manual adjustment, is feasible and quicker than manual analysis. Diagnostic accuracy was similar with all methods. SR s was lower in the speckle tracking-based method than in the Doppler-based methods.