Inge H. Gerrits

Performance evaluation of methods for two-dimensional displacement and strain estimation using ultrasound radio frequency data

In elastography, several methods for 2-D strain imaging have been introduced, based on both raw frequency (RF) data and speckle-tracking. Although the precision and lesion detectability of axial strain imaging in terms of elastographic signal-to-noise ratio (SNRe) and elastographic contrast-to-noise ratio (CNRe) have been reported extensively, analysis of lateral precision is still lacking. In this paper, the performance of different 2-D correlation RF- and envelope-based strain estimation methods was evaluated using simulation data and phantom experiments. Besides window size and interpolation methods for subsample displacement estimation, the influence of recorrelation techniques was examined. Precision and contrast of the measured displacements and strains were assessed using the difference between modeled and measured displacements, SNRe and CNRe. In general, a 2-D coarse-to-fine displacement estimation method is favored, using envelope data for window sizes exceeding the theoretical upper bound for strain estimation. Using 2-D windows of RF data resulted in better displacement estimates for both the axial and lateral direction than 1-D RF-based or envelope-based techniques. Obtaining subsample lateral displacement estimates by fitting a predefined shape through the cross-correlation function (CCF) yielded results similar to those obtained with up-sampling of RF data in the lateral direction. Using a CCF model was favored because of the decreased computation time. Local aligning and stretching of the windows (recorrelation) resulted in an increase of 2-17 and 6-7 dB in SNRe for axial and lateral strain estimates, respectively, over a range of strains (0.5 to 5.0%). For a simulated inhomogeneous phantom (2.0% applied strain), the measured axial and lateral SNRes were 29.2 and 20.2 dB, whereas the CNRes were 50.2 dB and 31.5 dB, respectively. For the experimental data, lower SNRe (axial: 28.5 dB; lateral: 17.5 dB) and CNRe (axial: 39.3 dB; lateral: 31 dB) were found. In conclusion, a coarse-to-fine approach is favored using RF data on a fine scale. The use of 2D parabolic interpolation is favored to obtain subsample displacement estimates. Recorrelation techniques, such as local aligning and stretching, increase SNRe and CNRe in both directions.

PERFORMANCE OF TWO DIMENSIONAL DISPLACEMENT AND STRAIN ESTIMATION TECHNIQUES USING A PHASED ARRAY TRANSDUCER

The goal of this study was to investigate the applicability of conventional 2-D displacement and strain imaging techniques to phased array radiofrequency (RF) data. Furthermore, the possible advantages of aligning and stretching techniques for the reduction of decorrelation artefacts was examined. Data from both realistic simulations and phantoms were used in this study. Recently, the used processing concepts were successfully applied to linear array data. However, their applicability to sector scan data is not trivial because of the polar grid. Homogeneous and inhomogeneous tissue phantoms were simulated at a range of strains (0 to 5%) using Field II((c)). The inhomogeneous phantom, a commonly used tumor/lesion model, was also constructed using gelatin/agar solutions. A coarse-to-fine displacement algorithm was applied, using aligning and stretching to enhance re-correlation. Vertical and horizontal strains were reconstructed from the axial and lateral displacements. Results revealed that the error on displacement estimates was lower when using 2-D data windows rather than 1-D windows. For regions at large depths and large insonification angles, the allowed lateral window size was limited. Still, 1-D windows resulted in larger errors. The re-correlation techniques resulted in a significant increase in the elastographic signal-to-noise ratio (SNRe) and elastographic contrast-to-noise ratio (CNRe) of the vertical and horizontal strain components. An increase of the SNRe of 5-20 dB was observed over a range of strains (0.5 to 5.0%). In the inhomogeneous phantom, a vertical SNRe of 27.7 dB and a horizontal SNRe of 16.7 dB were measured in the background. The vertical and horizontal CNRe were 35 dB and 23.1 dB, respectively. For the experimental data, lower SNRe (vertical: 19.1 dB; horizontal: 11.4 dB) and CNRe (vertical: 33.3 dB; horizontal: 12.5 dB) were found. In conclusion, 2-D window matching of sector scan data is feasible and outperforms 1-D window matching. Furthermore, the use of re-correlation techniques enhances both precision and contrast of strain images.

3D Cardiac Strain Imaging using a Novel Tracking Method

Current developments in 3D ultrasound imaging are boosting research in 3D strain imaging. 3D cardiac strain imaging seems promising considering the complex 3D structure and deformation of the heart. Segmentation and tracking of the tissue-of-interest is necessary for proper strain imaging. This study focuses on tracking of myocardial tissue using displacements and strain, derived from radio frequency ultrasound data. A novel tracking method was implemented to reconstruct strain from inter-frame strain measurements (i.e. strain rate). A region-of-interest (ROI) was deformed using displacement measurements and regularized using the behavior of neighbouring pixels. The forces were weighted using the difference between theoretical and measured cross-correlation values. The tracking method was tested using a tube phantom, which was heavily translated and dilated using a pressure column. Displacements and strain rate were estimated using the radio frequency data. Next, Biplane images of a canine heart were acquired. The strain algorithm was applied on the data. ROIs in the lateral wall in both the short and long axis view were manually segmented and subsequently tracked. In the phantom, each pixel within the ROI was tracked separately. The local strains were reconstructed correctly after tracking, despite large translations and deformations. Using the translational behavior of neighbouring pixels was favored over mesh regularization. In the animals, the strain curves were consistent with other animal and human studies on strain imaging. Furthermore, the strain curves showed less drift and higher peak strains after tracking.

3D strain estimation for isotropic and anisotropic materials

A study was undertaken to evaluate the feasibility of 3D strain imaging in muscle tissue. Simulations and phantom experiments were conducted for evaluation of iterative single-plane strain estimation algorithms. Rf-data is acquired with a 3D transducer (X4, Philips Sonos 7500) in bi-plane mode from the gelatin phantoms and a single volunteer’s calf muscle during passive compression. Iterative strain algorithms were applied using axial and non-axial correction and local stretching. The Field II simulation (a pre- and post-compression recording using artificially stretched scatterer positions (axial stretch: 1%; lateral stretch: 0.5%, cq. a Poisson-ratio of 0.5)) revealed excellent correlation between applied and measured axial strain. The phantom data showed similar results in both azimuth and lateral planes, although the Poisson’s ratio was slightly underestimated (0.46). Elastograms of the calf muscle showed different strain values in lateral and elevational direction due to the anisotropy of the calf muscle, despite the fact that the passive compression of the tissue was not optimal. 3D Elastography is feasible but further research is needed to improve lateral strain estimation and assess full-volume 3D strain.

In vivo validation of cardiac output assessment in non-standard 3D echocardiographic images.

Automatic segmentation of the endocardial surface in three-dimensional (3D) echocardiographic images is an important tool to assess left ventricular (LV) geometry and cardiac output (CO). The presence of speckle noise as well as the nonisotropic characteristics of the myocardium impose strong demands on the segmentation algorithm. In the analysis of normal heart geometries of standardized (apical) views, it is advantageous to incorporate a priori knowledge about the shape and appearance of the heart. In contrast, when analyzing abnormal heart geometries, for example in children with congenital malformations, this a priori knowledge about the shape and anatomy of the LV might induce erroneous segmentation results. This study describes a fully automated segmentation method for the analysis of non-standard echocardiographic images, without making strong assumptions on the shape and appearance of the heart. The method was validated in vivo in a piglet model. Real-time 3D echocardiographic image sequences of five piglets were acquired in radiofrequency (rf) format. These ECG-gated full volume images were acquired intra-operatively in a non-standard view. Cardiac blood flow was measured simultaneously by an ultrasound transit time flow probe positioned around the common pulmonary artery. Three-dimensional adaptive filtering using the characteristics of speckle was performed on the demodulated rf data to reduce the influence of speckle noise and to optimize the distinction between blood and myocardium. A gradient-based 3D deformable simplex mesh was then used to segment the endocardial surface. A gradient and a speed force were included as external forces of the model. To balance data fitting and mesh regularity, one fixed set of weighting parameters of internal, gradient and speed forces was used for all data sets. End-diastolic and end-systolic volumes were computed from the segmented endocardial surface. The cardiac output derived from this automatic segmentation was validated quantitatively by comparing it with the CO values measured from the volume flow in the pulmonary artery. Relative bias varied between 0 and -17%, where the nominal accuracy of the flow meter is in the order of 10%. Assuming the CO measurements from the flow probe as a gold standard, excellent correlation (r = 0.99) was observed with the CO estimates obtained from image segmentation.

Segmentation of 3D Echocardiographic Images using Deformable Simplex Meshes and Adaptive Filtering

Semi-automatic segmentation of the heart muscle in three-dimensional (3D) echographic images may substantially support clinical diagnosis of (congenital) heart disease. It may serve as an important preprocessing step for automated cardiac strain imaging and could be used for cardiac output measurement. Echocardiographic image sequences of the left ventricle of one healthy subject and one piglet were obtained in radiofrequency (rf) format, directly after beamforming, in Full Volume mode. 3D Adaptive Mean Squares (AMS) filtering was performed on the demodulated rf-data to optimize the distinction between blood and myocardium. A 3D deformable simplex mesh was then used to segment the endocardial surface. A gradient and a speed force were included as external forces of the model. Initial results show that segmentation of the endocardial surface using 3D deformable simplex meshes in combination with adaptive filtering is feasible. Cardiac output (CO) measurements in the piglet model, based on this segmentation technique were promising. The method can be applied to non-standard heart geometries without having to impose strong shape constraints.

Ultrasound strain imaging: From nano-scale motion detection to macro-scale functional imaging

With ultrasound strain imaging, the function of tissue and organs can be identified. The technique uses multiple images, acquired from tissue under different degrees of deformation. We recently applied this technique on hearts and skeletal muscles. Cardiac data was acquired in dogs with a valvar aorta stenosis. Muscle data was acquired from the orbicular oral muscle in the upper lip. For accurate assessment of deformation, the displacement of tissue can be determined at nanometer scale. Raw ultrasound data, containing the amplitude as well as the phase information is required for this analysis. A 2D coarse-to-fine strain estimation strategy is proposed to calculate the minute differential displacements in tissue, while the tissue itself is moving on a macro scale. The technique was validated using phantom experiments. These experiments demonstrated that accurate strain images can be determined using the proposed technique. Cardiac evaluation in dogs showed that the strain can be determined in three dimensions. The strain curves over the cardiac cycle are in correspondence with the severity of the stenosis of the aortic valve. In patients with a reconstructed cleft lip, the orbicular oral muscle in the reconstructed region showed decreased strain values. In normal individuals, similar strain values were found for all regions of the muscle. Ultrasound strain imaging is a promising technique enabling the addition of functional information to the geometrical information that is already provided by the conventional ultrasound imaging technique.

Segmentation of 3D cardiac ultrasound images using correlation of radio frequency data

Semi-automatic segmentation of the heart muscle in 3D echographic images may substantially support clinical diagnosis of heart disease. Especially in children with congenital heart disease, segmentation should be based on the echo features solely since a priori knowledge on the shape of the heart cannot be used. Segmentation of echocardiographic images is challenging because of the low echogenicity of the myocardium in some regions. High resolution information derived from radio frequency (rf) ultrasound data might be a useful additional feature in these regions. A semi-3D technique was used to determine maximum temporal cross-correlation values from the rf-data. To segment the endocardial surface, maximum cross-correlation values were used as additional external force in a deformable model approach and were tested against and combined with adaptive filtered, demodulated rf-data. The method was tested on pediatric full volume images (Philips, iE33) and evaluated by comparison with contours obtained from manual segmentation.