On punctate white matter lesions in preterm infants: Is ultrasound diagnosis feasible?

OBJECTIVES: To observe hyperechoic nodular or punctate white matter lesions (HNPL) in a population of preterm infants using routine cranial ultrasound (cUS), to describe the characteristics of HNPL, and to compare them with punctate white matter lesions (PWML) detected in magnetic resonance imaging (MRI). DESIGN: Retrospective observational single-center cohort study. SETTING: Level 2B neonatal unit in France. PATIENTS: 307 infants born <33 weeks gestation undergoing routine cUS with a total of 961 cUS performed. MAIN OUTCOME MEASURES: Description of lesions (HNPL/PWML): presence or absence, number, size, location, and structural distribution. RESULTS: Among the 307 included infants, 63 (20.5%) had at least one cerebral lesion, with 453 HNPL for 63 infants. HNPL were numerous (more than three in 66.6% of cases), primarily grouped in clusters (76.2%), located near the lateral ventricles (96.8%), and measuring more than 2 mm (79%). HNPL were diagnosed on day 29 on average and persisted until term. Overall, 43 MRI were performed in 307 infants, on average 18.9 days after last cUS, in 21 of those the indication was presence of HPNL on cUS. Of these 21 MRI, 14/21 presented 118 PWML compared to 173 HNPL on cUS. In the remaining MRI (7/21), no PWML were detected compared to 47 HNPL on cUS. CONCLUSIONS: In our population of 307 preterm infants, cUS allowed the diagnosis of HNPL, with a large similarity to PWML in MRI and a better sensitivity. But in the absence of data on inter-observer variability, we cannot exclude overdiagnosis of HNPL.

Semi-automatic data annotation based on feature-space projection and local quality metrics: An application to cerebral emboli characterization.

We propose a semi-supervised learning approach to annotate a dataset with reduced requirements for manual annotation and with controlled annotation error. The method is based on feature-space projection and label propagation using local quality metrics. First, an auto-encoder extracts the features of the samples in an unsupervised manner. Then, the extracted features are projected by a t-distributed stochastic neighbor embedding algorithm into a two-dimensional (2D) space. A selection of the best 2D projection is introduced based on the silhouette score. The expert annotator uses the obtained 2D representation to manually label samples. Finally, the labels of the labeled samples are propagated to the unlabeled samples using a K-nearest neighbor strategy and local quality metrics. We compare our method against semi-supervised optimum-path forest and K-nearest neighbor label propagation (without considering local quality metrics). Our method achieves state-of-the-art results on three different datasets by labeling more than 96% of the samples with an annotation error from 7% to 17%. Additionally, our method allows to control the trade-off between annotation error and number of labeled samples. Moreover, we combine our method with robust loss functions to compensate for the label noise introduced by automatic label propagation. Our method allows to achieve similar, and even better, classification performances compared to those obtained using a fully manually labeled dataset, with up to 6% in terms of classification accuracy.

Automatic segmentation and location learning of neonatal cerebral ventricles in 3D ultrasound data combining CNN and CPPN.

Preterm neonates are highly likely to suffer from ventriculomegaly, a dilation of the Cerebral Ventricular System (CVS). This condition can develop into life-threatening hydrocephalus and is correlated with future neuro-developmental impairments. Consequently, it must be detected and monitored by physicians. In clinical routing, manual 2D measurements are performed on 2D ultrasound (US) images to estimate the CVS volume but this practice is imprecise due to the unavailability of 3D information. A way to tackle this problem would be to develop automatic CVS segmentation algorithms for 3D US data. In this paper, we investigate the potential of 2D and 3D Convolutional Neural Networks (CNN) to solve this complex task and propose to use Compositional Pattern Producing Network (CPPN) to enable Fully Convolutional Networks (FCN) to learn CVS location. Our database was composed of 25 3D US volumes collected on 21 preterm nenonates at the age of 35.8±1.6 gestational weeks. We found that the CPPN enables to encode CVS location, which increases the accuracy of the CNNs when they have few layers. Accuracy of the 2D and 3D FCNs reached intraobserver variability (IOV) in the case of dilated ventricles with Dice of 0.893±0.008 and 0.886±0.004 respectively (IOV = 0.898±0.008) and with volume errors of 0.45±0.42 cm3 and 0.36±0.24 cm3 respectively (IOV = 0.41±0.05 cm3). 3D FCNs were more accurate than 2D FCNs in the case of normal ventricles with Dice of 0.797±0.041 against 0.776±0.038 (IOV = 0.816±0.009) and volume errors of 0.35±0.29 cm3 against 0.35±0.24 cm3 (IOV = 0.2±0.11 cm3). The best segmentation time of volumes of size 320×320×320 was obtained by a 2D FCN in 3.5±0.2 s.

Investigation of PVA cryogel Young's modulus stability with time, controlled by a simple reliable technique.

We describe a quasistatic method for mechanical characterization of tissue-mimicking material used in elastography. We demonstrate that it is possible to assess the elasticity modulus with a reasonable reproducibility using simple and easy tools and methods. Possessing a simple relevant technique with evaluated relative error to assess Young's modulus of these phantoms could deeply improve the quality of the research in the field of elastography. The method was tested and validated with four samples of polyvinyl alcohol (PVA) cryogel with different elasticity values corresponding to those of stiffer soft biological tissues. Young's moduli, varying from 70 to 180 kPa depending on the number of freeze-thaw cycles (two to five), were measured within strict measurement conditions and found to have a reproducibility varying from 4% to 8%. Relative error, estimated as the ratio between observed and reference values, varied from 16% to 32%. Besides, measurement stability over 4 months was evaluated. The method demonstrated good feasibility and acceptable reproducibility for mechanically characterizing and controlled over time phantoms used for validating new potential ultrasound imaging techniques in the field of elastography. Nevertheless, in this study, investigation was performed on gel possessing young's modulus values ranging from 80 to 215 kPa. Some tissue values of Young'modulus were reported to be lower, ranging from 0.6 to 28 kPa as liver or glandular values. Consequently, further validation of this static method for mechanical characterization of phantom gels should be performed using softer PVA cryogel.

Analytic estimation of subsample spatial shift using the phases of multidimensional analytic signals.

In this correspondence, a method of analytic subsample spatial shift estimation based on an a priori n-D signal model is proposed. The estimation uses the linear phases of n analytic signals defined with the multidimensional Hilbert transform. This estimation proposes: i) an analytic solution to the n-D shift estimation and ii) an estimation without processing complex cross-correlation function or cross-spectra between signals contrary to most phase shift estimators. The method provides better performance in estimating subsample shifts than two classical estimators, one using the maximum of cross-correlation function and the other seeking the zero of the complex correlation function phase. Two delay estimators using the in-phase and quadrature-phase components of signals are also compared to our estimator. Like most estimators using the complex signal phases, the estimator proposed herein presents the advantage of unaltered accuracy when low sampled signals are used. Moreover, we show that this method can be applied to motion tracking with ultrasound images. Thus, included in a block-based motion estimation method and tested with ultrasound data, this estimator provides an analytical solution to the translation estimation problem.

Multi-Grid Phase Field Skin Tumor Segmentation in 3D Ultrasound Images.

The aim of this paper is to present a new method for skin tumor segmentation in the 3D ultrasound images. We consider a variational formulation, the energy of which combines a diffuse interface phase field model (regularization term) and a log-likelihood computed using nonparametric estimates (data attachment term). We propose a multi-grid implementation with the exact solutions which has the advantage to avoid space discretization and numerical instabilities. The resulting algorithm is simple and easy to implement in multi-dimensions. Concerning applications, we focus on skin tumor segmentation. The clinical dataset used for the experiments is composed of 12 images with the ground truth given by a dermatologist. Comparisons with the reference methods show that the proposed method is more robust to the choice of the volume initialization. Moreover, thanks to the flexibility introduced by the diffuse interface, the sensitivity increases by 12% if the initialization is inside the lesion, and the Dice index increases by 59%, if the initialization covers the entire lesion. These results show that this new method is well designed to tackle the problem of underestimation of tumor volumes.

A gradient-based optical-flow cardiac motion estimation method for cine and tagged MR images.

A new method is proposed to quantify the myocardial motion from both 2D C(ine)-MRI and T(agged)-MRI sequences. The tag pattern offers natural landmarks within the image that makes it possible to accurately quantify the motion within the myocardial wall. Therefore, several methods have been proposed for T-MRI. However, the lack of salient features within the cardiac wall in C-MRI hampers local motion estimation. Our method aims to ensure the local intensity and shape features invariance during motion through the iterative minimization of a cost function via a random walk scheme. The proposed approach is evaluated on realistic simulated C-MRI and T-MRI sequences. The results show more than 53% improvements on displacement estimation, and more than 24% on strain estimation for both C-MRI and T-MRI sequences, as compared to state-of-the-art cardiac motion estimators. Preliminary experiments on clinical data have shown a good ability of the proposed method to detect abnormal motion patterns related to pathology. If those results are confirmed on large databases, this would open up the possibility for more accurate diagnosis of cardiac function from standard C-MRI examinations and also the retrospective study of prior studies.

Sequential Emboli Detection From Ultrasound Outpatient Data.

This paper addresses the detection of emboli from signals acquired with a new miniaturized and portable transcranial Doppler ultrasound device. The use of this device enables outpatient monitoring but increases the number of artifacts. These artifacts usually come from the patient voice and motion and can be superimposed to emboli. For this reason and because of the scarcity of emboli compared to artifacts, reliably detect emboli is a challenging task. As an example, the 11809 s of signal used in this study contained 0.06 % of embolic events and 10.14 % of artifacts. Herein, we propose an automatic and sequential approach. The method is based on sequential determination of high intensity transient signals. We also define efficient features to describe emboli in the time frequency representation. On our database, the number of artifacts detected as emboli is divided by more than 10 compared to the other algorithms reported in the literature.

Joint segmentation and characterization of the dermis in 50 MHz ultrasound 2D and 3D images of the skin.

We propose a novel joint segmentation and characterization algorithm for the assessment of skin aging using 50 MHz high-frequency ultrasound images. The proposed segmentation method allows a fine determination of the envelope signal's statistics in the dermis as a function of depth. The sequence of statistical estimates obtained is then combined into a single aging score. The segmentation is based on tailored recursive non-linear filters. The epidermis and the dermis are jointly segmented with a non-parametric active contour combining a texture criterion, an epidermis indicator map and the geometric constraint of horizontal continuity. The algorithm is designed to apply to 2D and 3D images as well. We evaluated skin photo-aging on ultrasound images with an experimental study on a cohort of 76 women separated into 2 groups of different ages. Two aging scores are computed from the images: local dermal contrast and skin roughness. We show that these scores are much better at identifying the two groups (p-value ≈10-6) than the previously used MGVR indicator (p-value 0.046). Moreover, we find that a combined score more reliably evaluates skin photo-aging, with 84% success, than a scoring of the ultrasound images by 4 experts.

PSF dedicated to estimation of displacement vectors for tissue elasticity imaging with ultrasound.

This paper investigates a new approach devoted to displacement vector estimation in ultrasound imaging. The main idea is to adapt the image formation to a given displacement estimation method to increase the precision of the estimation. The displacement is identified as the zero crossing of the phase of the complex cross-correlation between signals extracted from the lateral direction of the ultrasound RF image. For precise displacement estimation, a linearity of the phase slope is needed as well as a high phase slope. Consequently, a particular point spread function (PSF) dedicated to this estimator is designed. This PSF, showing oscillations in the lateral direction, leads to synthesis of lateral RF signals. The estimation is included in a 2-D displacement vector estimation method. The improvement of this approach is evaluated quantitatively by simulation studies. A comparison with a speckle tracking technique is also presented. The lateral oscillations improve both the speckle tracking estimation and our 2-D estimation method. Using our dedicated images, the precision of the estimation is improved by reducing the standard deviation of the lateral displacement error by a factor of 2 for speckle tracking and more than 3 with our method compared to using conventional images. Our method performs 7 times better than speckle tracking. Experimentally, the improvement in the case of a pure lateral translation reaches a factor of 7. Finally, the experimental feasibility of the 2-D displacement vector estimation is demonstrated on data acquired from a Cryogel phantom.

Discrimination between emboli and artifacts for outpatient transcranial Doppler ultrasound data.

This paper addresses the detection of emboli in transcranial Doppler ultrasound data acquired from an original portable device. The challenge is the removal of several artifacts (motion and voice) intrinsically related to long-duration (up to 1 h 40 mn per patient) outpatient signals monitoring from this device, as well as high intensities due to the stochastic nature of blood flow. This paper proposes an adapted removal procedure. This firstly consists of reducing the background noise and detecting the blood flow in the time-frequency domain using a likelihood method for contour detection. Then, a hierarchical extraction of features from magnitude and bounding boxes is achieved for the discrimination of emboli and artifacts. After processing of the long-duration outpatient signals, the number of artifacts predicted as emboli is considerably reduced (by 92% for some parameter values) between the first and the last step of our algorithm.

Segmentation of Skin Tumors in High-Frequency 3-D Ultrasound Images.

High-frequency 3-D ultrasound imaging is an informative tool for diagnosis, surgery planning and skin lesion examination. The purpose of this article was to describe a semi-automated segmentation tool providing easy access to the extent, shape and volume of a lesion. We propose an adaptive log-likelihood level-set segmentation procedure using non-parametric estimates of the intensity distribution. The algorithm has a single parameter to control the smoothness of the contour, and we describe how a fixed value yields satisfactory segmentation results with an average Dice coefficient of D = 0.76. The algorithm is implemented on a grid, which increases the speed by a factor of 100 compared with a standard pixelwise segmentation. We compare the method with parametric methods making the hypothesis of Rayleigh or Nakagami distributed signals, and illustrate that our method has greater robustness with similar computational speed. Benchmarks are made on realistic synthetic ultrasound images and a data set of nine clinical 3-D images acquired with a 50-MHz imaging system. The proposed algorithm is suitable for use in a clinical context as a post-processing tool.

Estimation of cardiac motion in cine-MRI sequences by correlation transform optical flow of monogenic features distance.

Cine-MRI is widely used for the analysis of cardiac function in clinical routine, because of its high soft tissue contrast and relatively short acquisition time in comparison with other cardiac MRI techniques. The gray level distribution in cardiac cine-MRI is relatively homogenous within the myocardium, and can therefore make motion quantification difficult. To ensure that the motion estimation problem is well posed, more image features have to be considered. This work is inspired by a method previously developed for color image processing. The monogenic signal provides a framework to estimate the local phase, orientation, and amplitude, of an image, three features which locally characterize the 2D intensity profile. The independent monogenic features are combined into a 3D matrix for motion estimation. To improve motion estimation accuracy, we chose the zero-mean normalized cross-correlation as a matching measure, and implemented a bilateral filter for denoising and edge-preservation. The monogenic features distance is used in lieu of the color space distance in the bilateral filter. Results obtained from four realistic simulated sequences outperformed two other state of the art methods even in the presence of noise. The motion estimation errors (end point error) using our proposed method were reduced by about 20% in comparison with those obtained by the other tested methods. The new methodology was evaluated on four clinical sequences from patients presenting with cardiac motion dysfunctions and one healthy volunteer. The derived strain fields were analyzed favorably in their ability to identify myocardial regions with impaired motion.

Analytic signal phase-based myocardial motion estimation in tagged MRI sequences by a bilinear model and motion compensation.

Different mathematical tools, such as multidimensional analytic signals, allow for the calculation of 2D spatial phases of real-value images. The motion estimation method proposed in this paper is based on two spatial phases of the 2D analytic signal applied to cardiac sequences. By combining the information of these phases issued from analytic signals of two successive frames, we propose an analytical estimator for 2D local displacements. To improve the accuracy of the motion estimation, a local bilinear deformation model is used within an iterative estimation scheme. The main advantages of our method are: (1) The phase-based method allows the displacement to be estimated with subpixel accuracy and is robust to image intensity variation in time; (2) Preliminary filtering is not required due to the bilinear model. The proposed algorithm, integrating phase-based optical flow motion estimation and the combination of global motion compensation with local bilinear transform, allows spatio-temporal cardiac motion analysis, e.g. strain and dense trajectory estimation over the cardiac cycle. Results from 7 realistic simulated tagged magnetic resonance imaging (MRI) sequences show that our method is more accurate compared with state-of-the-art method for cardiac motion analysis and with another differential approach from the literature. The motion estimation errors (end point error) of the proposed method are reduced by about 33% compared with that of the two methods. In our work, the frame-to-frame displacements are further accumulated in time, to allow for the calculation of myocardial Lagrangian cardiac strains and point trajectories. Indeed, from the estimated trajectories in time on 11 in vivo data sets (9 patients and 2 healthy volunteers), the shape of myocardial point trajectories belonging to pathological regions are clearly reduced in magnitude compared with the ones from normal regions. Myocardial point trajectories, estimated from our phase-based analytic signal approach, seem therefore a good indicator of the local cardiac dynamics. Moreover, they are shown to be coherent with the estimated deformation of the myocardium.

Characterization of PVA cryogel for intravascular ultrasound elasticity imaging.

The present study characterizes the mechanical properties of polyvinyl alcohol (PVA) cryogel in order to show its utility for intravascular elastography. PVA cryogel becomes harder with an increasing number of freeze-thaw cycles, and Young's modulus and Poisson's ratio are measured for seven samples. Mechanical tests were performed on cylindrical samples with a pressure column and on a hollow cylinder with the calculation of an intravascular elastogram. An image of the Young's modulus was obtained from the elastogram using cylinder geometry properties. Results show the mechanical similitude of PVA cryogel with the biological tissues present in arteries. A good agreement between Young's modulus obtained from pressure column and from elastogram was also observed.

Internal strain estimation for quantification of human heel pad elastic modulus: A phantom study.

Shock absorption is the most important function of the human heel pad. However, changes in heel pad elasticity, as seen in e.g. long-distance runners, diabetes patients, and victims of Falanga torture are affecting this function, often in a painful manner. Assessment of heel pad elasticity is usually based on one or a few strain measurements obtained by an external load-deformation system. The aim of this study was to develop a technique for quantitative measurements of heel pad elastic modulus based on several internal strain measures from within the heel pad by use of ultrasound images. Nine heel phantoms were manufactured featuring a combination of three heel pad stiffnesses and three heel pad thicknesses to model the normal human variation. Each phantom was tested in an indentation system comprising a 7MHz linear array ultrasound transducer, working as the indentor, and a connected load cell. Load-compression data and ultrasound B-mode images were simultaneously acquired in 19 compression steps of 0.1mm each. The internal tissue displacement was for each step calculated by a phase-based cross-correlation technique and internal strain maps were derived from these displacement maps. Elastic moduli were found from the resulting stress-strain curves. The elastic moduli made it possible to distinguish eight of nine phantoms from each other according to the manufactured stiffness and showed very little dependence of the thickness. Mean elastic moduli for the three soft, the three medium, and the three hard phantoms were 89kPa, 153kPa, and 168kPa, respectively. The combination of ultrasound images and force measurements provided an effective way of assessing the elastic properties of the heel pad due to the internal strain estimation.

Motion estimation in ultrasound imaging applied to the diagnostic of pelvic floor disorders.

The main purpose of this paper is to show the potential of tissue motion estimation in ultrasound imaging for the diagnostic of pelvic floor disorders. We propose to evaluate the tissue motion using a method based on a local deformable model and on image features (local phase and orientation) extracted from the monogenic signal. The proposed method is well adapted to the pelvic organ deformations and estimates motion with subpixel precision without the need for interpolation. The estimated motion is used to visualize the bladder local deformation and to extract quantitative figures such as the deformation parameters and the bladder angle variation. These results could potentially be interesting to characterize the degree of the pelvic organ prolapse.

Motion estimation using the monogenic signal applied to ultrasound elastography.

This paper presents a phase-based method for estimating subpixel motion in medical ultrasound imaging. The main novelty is to propose a closed form expression for local motion estimation in space domain based on the model of the monogenic signal. First, the proposed method uses the properties of the monogenic signal in order to extract local orientation and phase data of ultrasound images. Second, using these image features, we show how the proposed method can analytically estimate the dense motion field between a pair of images. This way of estimating motion provides subpixel accuracy without interpolating the images. For this reason, we show that the proposed method has high computational efficiency. Results on simulated and experimental data show that this method provides a better Contrast to Noise Ratio than a classical block-matching method applied to images that are 25 times larger (in number of pixels).

Analysis of motion tracking in echocardiographic image sequences: influence of system geometry and point-spread function.

This paper focuses on motion tracking in echocardiographic ultrasound images. The difficulty of this task is related to the fact that echographic image formation induces decorrelation between the underlying motion of tissue and the observed speckle motion. Since Meunier's seminal work, this phenomenon has been investigated in many simulation studies as part of speckle tracking or optical flow-based motion estimation techniques. Most of these studies modeled image formation using a linear convolution approach, where the system point-spread function (PSF) was spatially invariant and the probe geometry was linear. While these assumptions are valid over a small spatial area, they constitute an oversimplification when a complete image is considered. Indeed, echocardiographic acquisition geometry relies on sectorial probes and the system PSF is not perfectly invariant, even if dynamic focusing is performed. This study investigated the influence of sectorial geometry and spatially varying PSF on speckle tracking. This was done by simulating a typical 64 elements, cardiac probe operating at 3.5 MHz frequency, using the simulation software Field II. This simulation first allowed quantification of the decorrelation induced by the system between two images when simple motion such as translation or incompressible deformation was applied. We then quantified the influence of decorrelation on speckle tracking accuracy using a conventional block matching (BM) algorithm and a bilinear deformable block matching (BDBM) algorithm. In echocardiography, motion estimation is usually performed on reconstructed images where the initial sectorial (i.e., polar) data are interpolated on a cartesian grid. We therefore studied the influence of sectorial acquisition geometry, by performing block matching on cartesian and polar data. Simulation results show that decorrelation is spatially variant and depends on the position of the region where motion takes place relative to the probe. Previous studies did not consider translation in their experiments, since their simulation model (spatially invariant PSF and linear probe) yields by definition no decorrelation. On the opposite, our realistic simulation settings (i.e., sectorial probe and realistic beamforming) show that translation yields decorrelation, particularly when translation is large (above 6 mm) and when the moving regions is located close to the probe (distance to probe less than 50 mm). The tracking accuracy study shows that tracking errors are larger for the usual cartesian data, whatever the estimation algorithm, indicating that speckle tracking is more reliable when based on the unconverted polar data: for axial translations in the range 0-10 mm, the maximum error associated to conventional block matching (BM) is 4.2 mm when using cartesian data and 1.8 mm for polar data. The corresponding errors are 1.8 mm (cartesian data) and 0.4 mm (polar data) for an applied deformation in the range 0-10%. We also show that accuracy is improved by using the bilinear deformable block matching (BDBM) algorithm. For translation, the maximum error associated to the bilinear deformable block matching is indeed 3.6mm (cartesian data) and 1.2 mm (polar data). Regarding deformation, the error is 0.7 mm (cartesian data) and 0.3 mm (polar data). These figures also indicates that the larger improvement brought by the bilinear deformable block matching over standard block matching logically takes place when deformation on cartesian data is considered (the error drops from 1.8 to 0.7 mm is this case). We give a preliminary evaluation of this framework on a cardiac sequence acquired with a Toshiba Powervision 6000 imaging system using a probe operating at 3.25 MHz. As ground truth reference motion is not available in this case, motion estimation performance was evaluated by comparing a reference image (i.e., the first image of the sequence) and the subsequent images after motion compensation has been applied. The comparison was quantified by computing the normalized correlation between the reference and the motion-compensated images. The obtained results are consistent with the simulation data: correlation is smaller for cartesian data, whatever the estimation algorithm. The correlation associated to the conventional block matching (BM) is in the range 0.45-0.02 when using cartesian data and in the range 0.65-0.2 for polar data. The corresponding correlation ranges for the bilinear deformable block matching are 0.98-0.2 and 0.98-0.55. In the same way these figures indicate that the bilinear deformable block matching yield a larger improvement when cartesian data are considered (correlation range increases from 0.45-0.02 to 0.98-0.2 in this case).

Blood flow evaluation in high-frequency, 40 MHz imaging: a comparative study of four vector velocity estimation methods.

Ultrasonic imaging is often used to estimate blood flow velocity. Currently, estimates are carried out using Doppler-based techniques. However, there are a number of shortcomings such as the limited spatial resolution and the inability to estimate longitudinal flows. Thus, alternative methods have been proposed to overcome them. Difficulties are notably encountered with high-frequency imaging systems that use swept-scan techniques. In this article, we propose to compare four vector velocity estimation methods that are complementary to Doppler, focusing on 40 MHz, high-frequency imaging. The goal of this study is to evaluate which method could circumvent the limitations of Doppler methods for evaluation of microcirculation, in the vessels having diameter on the order of 1 mm. We used two region-based approaches, one decorrelation-based approach and one spatiotemporal approach. Each method has been applied to seven flow sequences with various orientations and mean velocities. Four sequences were simulated with a system approach based on a 3D set of moving scatterers. Three experimental sequences were carried out by injecting blood-mimicking fluid within a gelatin phantom and then acquiring images with Visualsonics, Vevo 660 system. From velocity estimates, several performance criteria such as the normalized mean error or the normalized mean standard deviation were defined to compare the performance of the four estimators. The results show that region-based methods are the most accurate exhibiting mean errors less than 10% and mean standard deviation less than 13%. However, region-based approaches are those that require the highest calculative cost compared to the decorrelation-based method, which is the fastest. Finally, the spatiotemporal approach appeared to be a trade-off in terms of computational complexity and accuracy of estimates. It provides estimates with errors less than 10% for mean velocity and the CPU time is approximately 17s for a ROI of size 40*80 pixels.