Left ventricular volume estimation for real-time three-dimensional echocardiography

Cardiac Segmentation Method Based on Domain Knowledge

Echocardiography plays an important role in the clinical diagnosis of cardiovascular diseases. Cardiac function assessment by echocardiography is a crucial process in daily cardiology. However, cardiac segmentation in echocardiography is a challenging task due to shadows and speckle noise. The traditional manual segmentation method is a time-consuming process and limited by inter-observer variability. In this paper, we present a fast and accurate echocardiographic automatic segmentation framework based on Convolutional neural networks (CNN). We propose FAUet, a segmentation method serially integrated U-Net with coordinate attention mechanism and domain feature loss from VGG19 pre-trained on the ImageNet dataset. The coordinate attention mechanism can capture long-range dependencies along one spatial direction and meanwhile preserve precise positional information along the other spatial direction. And the domain feature loss is more concerned with the topology of cardiac structures by exploiting their higher-level features. In this research, we use a two-dimensional echocardiogram (2DE) of 88 patients from two devices, Philips Epiq 7C and Mindray Resona 7T, to segment the left ventricle (LV), interventricular septal (IVS), and posterior left ventricular wall (PLVW). We also draw the gradient weighted class activation mapping (Grad-CAM) to improve the interpretability of the segmentation results. Compared with the traditional U-Net, the proposed segmentation method shows better performance. The mean Dice Score Coefficient (Dice) of LV, IVS, and PLVW of FAUet can achieve 0.932, 0.848, and 0.868, and the average Dice of the three objects can achieve 0.883. Statistical analysis showed that there is no significant difference between the segmentation results of the two devices. The proposed method can realize fast and accurate segmentation of 2DE with a low time cost. Combining coordinate attention module and feature loss with the original U-Net framework can significantly increase the performance of the algorithm.

One Shot Segmentation: Unifying Rigid Detection and Non-Rigid Segmentation Using Elastic Regularization

This paper proposes a novel approach for the non-rigid segmentation of deformable objects in image sequences, which is based on one-shot segmentation that unifies rigid detection and non-rigid segmentation using elastic regularization. The domain of application is the segmentation of a visual object that temporally undergoes a rigid transformation (e.g., affine transformation) and a non-rigid transformation (i.e., contour deformation). The majority of segmentation approaches to solve this problem are generally based on two steps that run in sequence: a rigid detection, followed by a non-rigid segmentation. In this paper, we propose a new approach, where both the rigid and non-rigid segmentation are performed in a single shot using a sparse low-dimensional manifold that represents the visual object deformations. Given the multi-modality of these deformations, the manifold partitions the training data into several patches, where each patch provides a segmentation proposal during the inference process. These multiple segmentation proposals are merged using the classification results produced by deep belief networks (DBN) that compute the confidence on each segmentation proposal. Thus, an ensemble of DBN classifiers is used for estimating the final segmentation. Compared to current methods proposed in the field, our proposed approach is advantageous in four aspects: (i) it is a unified framework to produce rigid and non-rigid segmentations; (ii) it uses an ensemble classification process, which can help the segmentation robustness; (iii) it provides a significant reduction in terms of the number of dimensions of the rigid and non-rigid segmentations search spaces, compared to current approaches that divide these two problems; and (iv) this lower dimensionality of the search space can also reduce the need for large annotated training sets to be used for estimating the DBN models. Experiments on the problem of left ventricle endocardial segmentation from ultrasound images, and lip segmentation from frontal facial images using the extended Cohn-Kanade (CK+) database, demonstrate the potential of the methodology through qualitative and quantitative evaluations, and the ability to reduce the search and training complexities without a significant impact on the segmentation accuracy.

Anatomical structure segmentation from early fetal ultrasound sequences using global pollination CAT swarm optimizer-based Chan-Vese model

The structure of an early fetal heart provides essential information for the diagnosis of fetus defects. Accurate segmentation of anatomical structure is a major challenging task because of the small size, low signal-to-noise ratio, and rapid movement of the ultrasound images. In recent years, active contour methods have found applications to ultrasound image segmentation. The familiar region-based Chan-Vese (RCV) model is a strong and flexible technique that is able to segment many types of images compared to other active contours. However, the solution trapping in local minima is the main drawback determined on the RCV model with the exposure of improper initial contours. Also, the RCV model showed poor results with this situation. More probably, the images having large intensity differences between global and local structures usually suffered from this problem. To solve this issue, we develop an improved version of the RCV model which is expected to achieve satisfactory segmentation performance, irrespective of the initial selection of the contour. We have formulated a new and hybrid meta-heuristic optimization algorithm namely global pollination-based CAT swarm (GPCATS) optimizer to solve the fitting energy minimization problem. In the GPCATS method, the global pollination step of the flower pollination algorithm (FPA) is used for improving the distance averaging of the CATS algorithm. The performance of the proposed method was analyzed on different fetal heart ultrasound videos acquired from 12 subjects. Each frame of each video was manually annotated in order to provide labels for training and validating the model. Experimental results of the proposed model proved that the precision of locating boundaries is improved greatly and requires only a reduced number of iterations (75% less) for convergence compared to the traditional RCV model. This proposed method also proved that our model not only enhances the accuracy of locating boundaries but also works stronger robustness than some other active contour methods. Graphical Abstract Anatomical structure segmentation from early fetal ultrasound sequences using GPCATS based Chan-Vese Model.

Synergy between Diastolic Mitral Valve Function and Left Ventricular Flow Aids in Valve Closure and Blood Transport during Systole

Highly resolved three-dimensional (3D) fluid structure interaction (FSI) simulation using patient-specific echocardiographic data can be a powerful tool for accurately and thoroughly elucidating the biomechanics of mitral valve (MV) function and left ventricular (LV) fluid dynamics. We developed and validated a strongly coupled FSI algorithm to fully characterize the LV flow field during diastolic MV opening under physiologic conditions. Our model revealed that distinct MV deformation and LV flow patterns developed during different diastolic stages. A vortex ring that strongly depended on MV deformation formed during early diastole. At peak E wave, the MV fully opened, with a local Reynolds number of ~5500, indicating that the flow was in the laminar-turbulent transitional regime. Our results  showed that during diastasis, the vortex structures caused the MV leaflets to converge, thus increasing mitral jet’s velocity. The vortex ring became asymmetrical, with the vortex structures on the anterior side being larger than on the posterior side. During the late diastolic stages, the flow structures advected toward the LV outflow tract, enhancing fluid transport to the aorta. This 3D-FSI study demonstrated the importance of leaflet dynamics, their effect on the vortex ring, and their influence on MV function and fluid transport within the LV during diastole.

Theory of energy harvesting from heartbeat including the effects of pleural cavity and respiration

Self-powered implantable devices with flexible energy harvesters are of significant interest due to their potential to solve the problem of limited battery life and surgical replacement. The flexible electronic devices made of piezoelectric materials have been employed to harvest energy from the motion of biological organs. Experimental measurements show that the output voltage of the device mounted on porcine left ventricle in chest closed environment decreases significantly compared to the case of chest open. A restricted-space deformation model is proposed to predict the impeding effect of pleural cavity, surrounding tissues, as well as respiration on the efficiency of energy harvesting from heartbeat using flexible piezoelectric devices. The analytical solution is verified by comparing theoretical predictions to experimental measurements. A simple scaling law is established to analyse the intrinsic correlations between the normalized output power and the combined system parameters, i.e. the normalized permitted space and normalized electrical load. The results may provide guidelines for optimization of in vivo energy harvesting from heartbeat or the motions of other biological organs using flexible piezoelectric energy harvesters.

Cardiac and respiratory-gated volumetric murine ultrasound

Current cardiovascular ultrasound mainly employs planar imaging techniques to assess function and physiology. These techniques rely on geometric assumptions, which are dependent on the imaging plane, susceptible to inter-observer variability, and may be inaccurate when studying complex diseases. Here, we developed a gated volumetric murine ultrasound technique to visualize cardiovascular motion with high spatiotemporal resolution and directly evaluate cardiovascular health. Cardiac and respiratory-gated cine loops, acquired at 1000 frames-per-second from sequential positions, were temporally registered to generate a four-dimensional (4D) dataset. We applied this technique to (1) evaluate left ventricular (LV) function from both healthy mice and mice with myocardial infarction and (2) characterize aortic wall strain of angiotensin II-induced dissecting abdominal aortic aneurysms in apolipoprotein E-deficient mice. Combined imaging and processing times for the 4D technique was approximately 2-4 times longer than conventional 2D approaches, but substantially more data is collected with 4D ultrasound and further optimization can be implemented to reduce imaging times. Direct volumetric measurements of 4D cardiac data aligned closely with those obtained from MRI, contrary to conventional methods, which were sensitive to transducer alignment, leading to overestimation or underestimation of estimated LV parameters in infarcted hearts. Green-Lagrange circumferential strain analysis revealed higher strain values proximal and distal to the aneurysm than within the aneurysmal region, consistent with published reports. By eliminating the need for geometrical assumptions, the presented 4D technique can be used to more accurately evaluate cardiac function and aortic pulsatility. Furthermore, this technique allows for the visualization of regional differences that may be overlooked with conventional 2D approaches.

Evaluation of a New Ensemble Learning Framework for Mass Classification in Mammograms

BACKGROUND

Mammography is the most common screening method for diagnosis of breast cancer.

MATERIALS AND METHODS

In this study, a computer-aided system for diagnosis of benignity and malignity of the masses was implemented in mammogram images. In the computer aided diagnosis system, we first reduce the noise in the mammograms using an effective noise removal technique. After the noise removal, the mass in the region of interest must be segmented and this segmentation is done using a deformable model. After the mass segmentation, a number of features are extracted from it. These features include: features of the mass shape and border, tissue properties, and the fractal dimension. After extracting a large number of features, a proper subset must be chosen from among them. In this study, we make use of a new method on the basis of a genetic algorithm for selection of a proper set of features. After determining the proper features, a classifier is trained.

RESULTS

To classify the samples, a new architecture for combination of the classifiers is proposed. In this architecture, easy and difficult samples are identified and trained using different classifiers. Finally, the proposed mass diagnosis system was also tested on mini-Mammographic Image Analysis Society and digital database for screening mammography databases.

CONCLUSION

The obtained results indicate that the proposed system can compete with the state-of-the-art methods in terms of accuracy.

Deep Learning on Sparse Manifolds for Faster Object Segmentation

We propose a new combination of deep belief networks and sparse manifold learning strategies for the 2D segmentation of non-rigid visual objects. With this novel combination, we aim to reduce the training and inference complexities while maintaining the accuracy of machine learning-based non-rigid segmentation methodologies. Typical non-rigid object segmentation methodologies divide the problem into a rigid detection followed by a non-rigid segmentation, where the low dimensionality of the rigid detection allows for a robust training (i.e., a training that does not require a vast amount of annotated images to estimate robust appearance and shape models) and a fast search process during inference. Therefore, it is desirable that the dimensionality of this rigid transformation space is as small as possible in order to enhance the advantages brought by the aforementioned division of the problem. In this paper, we propose the use of sparse manifolds to reduce the dimensionality of the rigid detection space. Furthermore, we propose the use of deep belief networks to allow for a training process that can produce robust appearance models without the need of large annotated training sets. We test our approach in the segmentation of the left ventricle of the heart from ultrasound images and lips from frontal face images. Our experiments show that the use of sparse manifolds and deep belief networks for the rigid detection stage leads to segmentation results that are as accurate as the current state of the art, but with lower search complexity and training processes that require a small amount of annotated training data.

Real-Time Ultrasound Segmentation, Analysis and Visualisation of Deep Cervical Muscle Structure

Despite widespread availability of ultrasound and a need for personalised muscle diagnosis (neck/back pain-injury, work related disorder, myopathies, neuropathies), robust, online segmentation of muscles within complex groups remains unsolved by existing methods. For example, Cervical Dystonia (CD) is a prevalent neurological condition causing painful spasticity in one or multiple muscles in the cervical muscle system. Clinicians currently have no method for targeting/monitoring treatment of deep muscles. Automated methods of muscle segmentation would enable clinicians to study, target, and monitor the deep cervical muscles via ultrasound. We have developed a method for segmenting five bilateral cervical muscles and the spine via ultrasound alone, in real-time. Magnetic Resonance Imaging (MRI) and ultrasound data were collected from 22 participants (age: 29.0±6.6, male: 12). To acquire ultrasound muscle segment labels, a novel multimodal registration method was developed, involving MRI image annotation, and shape registration to MRI-matched ultrasound images, via approximation of the tissue deformation. We then applied polynomial regression to transform our annotations and textures into a mean space, before using shape statistics to generate a texture-to-shape dictionary. For segmentation, test images were compared to dictionary textures giving an initial segmentation, and then we used a customized Active Shape Model to refine the fit. Using ultrasound alone, on unseen participants, our technique currently segments a single image in [Formula: see text] to over 86% accuracy (Jaccard index). We propose this approach is applicable generally to segment, extrapolate and visualise deep muscle structure, and analyse statistical features online.

Standardized Evaluation System for Left Ventricular Segmentation Algorithms in 3D Echocardiography

Real-time 3D Echocardiography (RT3DE) has been proven to be an accurate tool for left ventricular (LV) volume assessment. However, identification of the LV endocardium remains a challenging task, mainly because of the low tissue/blood contrast of the images combined with typical artifacts. Several semi and fully automatic algorithms have been proposed for segmenting the endocardium in RT3DE data in order to extract relevant clinical indices, but a systematic and fair comparison between such methods has so far been impossible due to the lack of a publicly available common database. Here, we introduce a standardized evaluation framework to reliably evaluate and compare the performance of the algorithms developed to segment the LV border in RT3DE. A database consisting of 45 multivendor cardiac ultrasound recordings acquired at different centers with corresponding reference measurements from three experts are made available. The algorithms from nine research groups were quantitatively evaluated and compared using the proposed online platform. The results showed that the best methods produce promising results with respect to the experts' measurements for the extraction of clinical indices, and that they offer good segmentation precision in terms of mean distance error in the context of the experts' variability range. The platform remains open for new submissions.

Methods for 2-D and 3-D Endobronchial Ultrasound Image Segmentation

Endobronchial ultrasound (EBUS) is now commonly used for cancer-staging bronchoscopy. Unfortunately, EBUS is challenging to use and interpreting EBUS video sequences is difficult. Other ultrasound imaging domains, hampered by related difficulties, have benefited from computer-based image-segmentation methods. Yet, so far, no such methods have been proposed for EBUS. We propose image-segmentation methods for 2-D EBUS frames and 3-D EBUS sequences. Our 2-D method adapts the fast-marching level-set process, anisotropic diffusion, and region growing to the problem of segmenting 2-D EBUS frames. Our 3-D method builds upon the 2-D method while also incorporating the geodesic level-set process for segmenting EBUS sequences. Tests with lung-cancer patient data showed that the methods ran fully automatically for nearly 80% of test cases. For the remaining cases, the only user-interaction required was the selection of a seed point. When compared to ground-truth segmentations, the 2-D method achieved an overall Dice index = 90.0% ±4.9%, while the 3-D method achieved an overall Dice index = 83.9 ± 6.0%. In addition, the computation time (2-D, 0.070 s/frame; 3-D, 0.088 s/frame) was two orders of magnitude faster than interactive contour definition. Finally, we demonstrate the potential of the methods for EBUS localization in a multimodal image-guided bronchoscopy system.

Semi-supervised segmentation of ultrasound images based on patch representation and continuous min cut

Ultrasound segmentation is a challenging problem due to the inherent speckle and some artifacts like shadows, attenuation and signal dropout. Existing methods need to include strong priors like shape priors or analytical intensity models to succeed in the segmentation. However, such priors tend to limit these methods to a specific target or imaging settings, and they are not always applicable to pathological cases. This work introduces a semi-supervised segmentation framework for ultrasound imaging that alleviates the limitation of fully automatic segmentation, that is, it is applicable to any kind of target and imaging settings. Our methodology uses a graph of image patches to represent the ultrasound image and user-assisted initialization with labels, which acts as soft priors. The segmentation problem is formulated as a continuous minimum cut problem and solved with an efficient optimization algorithm. We validate our segmentation framework on clinical ultrasound imaging (prostate, fetus, and tumors of the liver and eye). We obtain high similarity agreement with the ground truth provided by medical expert delineations in all applications (94% DICE values in average) and the proposed algorithm performs favorably with the literature.

New developments in paediatric cardiac functional ultrasound imaging

Ultrasound imaging can be used to estimate the morphology as well as the motion and deformation of tissues. If the interrogated tissue is actively deforming, this deformation is directly related to its function and quantification of this deformation is normally referred as 'strain imaging'. Tissue can also be deformed by applying an internal or external force and the resulting, induced deformation is a function of the mechanical tissue characteristics. In combination with the load applied, these strain maps can be used to estimate or reconstruct the mechanical properties of tissue. This technique was named 'elastography' by Ophir et al. in 1991. Elastography can be used for atherosclerotic plaque characterisation, while the contractility of the heart or skeletal muscles can be assessed with strain imaging. Rather than using the conventional video format (DICOM) image information, radio frequency (RF)-based ultrasound methods enable estimation of the deformation at higher resolution and with higher precision than commercial methods using Doppler (tissue Doppler imaging) or video image data (2D speckle tracking methods). However, the improvement in accuracy is mainly achieved when measuring strain along the ultrasound beam direction, so it has to be considered a 1D technique. Recently, this method has been extended to multiple directions and precision further improved by using spatial compounding of data acquired at multiple beam steered angles. Using similar techniques, the blood velocity and flow can be determined. RF-based techniques are also beneficial for automated segmentation of the ventricular cavities. In this paper, new developments in different techniques of quantifying cardiac function by strain imaging, automated segmentation, and methods of performing blood flow imaging are reviewed and their application in paediatric cardiology is discussed.

Automatic 4D reconstruction of patient-specific cardiac mesh with 1-to-1 vertex correspondence from segmented contours lines

We propose an automatic algorithm for the reconstruction of patient-specific cardiac mesh models with 1-to-1 vertex correspondence. In this framework, a series of 3D meshes depicting the endocardial surface of the heart at each time step is constructed, based on a set of border delineated magnetic resonance imaging (MRI) data of the whole cardiac cycle. The key contribution in this work involves a novel reconstruction technique to generate a 4D (i.e., spatial-temporal) model of the heart with 1-to-1 vertex mapping throughout the time frames. The reconstructed 3D model from the first time step is used as a base template model and then deformed to fit the segmented contours from the subsequent time steps. A method to determine a tree-based connectivity relationship is proposed to ensure robust mapping during mesh deformation. The novel feature is the ability to handle intra- and inter-frame 2D topology changes of the contours, which manifests as a series of merging and splitting of contours when the images are viewed either in a spatial or temporal sequence. Our algorithm has been tested on five acquisitions of cardiac MRI and can successfully reconstruct the full 4D heart model in around 30 minutes per subject. The generated 4D heart model conforms very well with the input segmented contours and the mesh element shape is of reasonably good quality. The work is important in the support of downstream computational simulation activities.

Combining multiple dynamic models and deep learning architectures for tracking the left ventricle endocardium in ultrasound data

We present a new statistical pattern recognition approach for the problem of left ventricle endocardium tracking in ultrasound data. The problem is formulated as a sequential importance resampling algorithm such that the expected segmentation of the current time step is estimated based on the appearance, shape, and motion models that take into account all previous and current images and previous segmentation contours produced by the method. The new appearance and shape models decouple the affine and nonrigid segmentations of the left ventricle to reduce the running time complexity. The proposed motion model combines the systole and diastole motion patterns and an observation distribution built by a deep neural network. The functionality of our approach is evaluated using a dataset of diseased cases containing 16 sequences and another dataset of normal cases comprised of four sequences, where both sets present long axis views of the left ventricle. Using a training set comprised of diseased and healthy cases, we show that our approach produces more accurate results than current state-of-the-art endocardium tracking methods in two test sequences from healthy subjects. Using three test sequences containing different types of cardiopathies, we show that our method correlates well with interuser statistics produced by four cardiologists.

Novel indices for left-ventricular dyssynchrony characterization based on highly automated segmentation from real-time 3-d echocardiography

Cardiac resynchronization therapy (CRT) using a biventricular pacemaker is an invasive and expensive treatment option for left ventricular mechanical dyssynchrony (LVMD). The CRT candidate selection is a crucial issue due to the unreliability of the current standard CRT indicators. Real-time three-dimensional (3-D) echocardiography (RT3DE) provides four-dimensional (4-D) (3-D+time) information about the LV and is suitable for LVMD assessment. In this article, the complex left ventricle (LV) shape and motion of 50 RT3DE datasets are represented by novel 4-D descriptors - 4-D sphericity, volume and shape, from which novel indices were derived by principal component analysis (PCA) and subsequently analyzed by a support vector machine (SVM) classifier to assess their capability of LVMD characterization and CRT outcome prediction. These novel indices outperformed clinical indices and have promising capabilities in disease characterization and great potential in CRT outcome prediction. To enable efficient quantitative RT3DE analysis, a segmentation method was developed to combine the powers of active shape models and optimal graph search. Various aspects of the method were designed to handle varying RT3DE image quality among datasets and LV segments. An application with graphical user interface was developed to provide the user with simple and intuitive control. The developed method was robust to inter-observer variability and produced very good accuracy - 3.2±1.1 mm absolute surface positioning error, <1 mm mean signed error and <5% mean volume difference. The computer method's classification performance was compared with the independent standard, showing that the 4-D shape modal indices were not only the most capable of all tested options when employed for disease characterization but also the least sensitive to segmentation imperfections.

The segmentation of the left ventricle of the heart from ultrasound data using deep learning architectures and derivative-based search methods

We present a new supervised learning model designed for the automatic segmentation of the left ventricle (LV) of the heart in ultrasound images. We address the following problems inherent to supervised learning models: 1) the need of a large set of training images; 2) robustness to imaging conditions not present in the training data; and 3) complex search process. The innovations of our approach reside in a formulation that decouples the rigid and nonrigid detections, deep learning methods that model the appearance of the LV, and efficient derivative-based search algorithms. The functionality of our approach is evaluated using a data set of diseased cases containing 400 annotated images (from 12 sequences) and another data set of normal cases comprising 80 annotated images (from two sequences), where both sets present long axis views of the LV. Using several error measures to compute the degree of similarity between the manual and automatic segmentations, we show that our method not only has high sensitivity and specificity but also presents variations with respect to a gold standard (computed from the manual annotations of two experts) within interuser variability on a subset of the diseased cases. We also compare the segmentations produced by our approach and by two state-of-the-art LV segmentation models on the data set of normal cases, and the results show that our approach produces segmentations that are comparable to these two approaches using only 20 training images and increasing the training set to 400 images causes our approach to be generally more accurate. Finally, we show that efficient search methods reduce up to tenfold the complexity of the method while still producing competitive segmentations. In the future, we plan to include a dynamical model to improve the performance of the algorithm, to use semisupervised learning methods to reduce even more the dependence on rich and large training sets, and to design a shape model less dependent on the training set.

Correlation based 3-D segmentation of the left ventricle in pediatric echocardiographic images using radio-frequency data

Clinical diagnosis of heart disease might be substantially supported by automated segmentation of the endocardial surface in three-dimensional (3-D) echographic images. Because of the poor echogenicity contrast between blood and myocardial tissue in some regions and the inherent speckle noise, automated analysis of these images is challenging. A priori knowledge on the shape of the heart cannot always be relied on, e.g., in children with congenital heart disease, segmentation should be based on the echo features solely. The objective of this study was to investigate the merit of using temporal cross-correlation of radio-frequency (RF) data for automated segmentation of 3-D echocardiographic images. Maximum temporal cross-correlation (MCC) values were determined locally from the RF-data using an iterative 3-D technique. MCC values as well as a combination of MCC values and adaptive filtered, demodulated RF-data were used as an additional, external force in a deformable model approach to segment the endocardial surface and were tested against manually segmented surfaces. Results on 3-D full volume images (Philips, iE33) of 10 healthy children demonstrate that MCC values derived from the RF signal yield a useful parameter to distinguish between blood and myocardium in regions with low echogenicity contrast and incorporation of MCC improves the segmentation results significantly. Further investigation of the MCC over the whole cardiac cycle is required to exploit the full benefit of it for automated segmentation.

A regularization technique for closed contour segmentation in ultrasound images

Segmentation of a target object in the form of a closed curve has many potential applications in medical imaging because it provides quantitative information related to the target objext's size and shape. However, ultrasound image segmentation for boundary delineation of the target object is a very difficult task because of its inherent drawbacks, including uncertainty of the segmentation boundary caused by speckle noise, relatively low SNR, and low contrast. Indeed, in automatic ultrasound image segmentation, conventional techniques with standard regularization often fail to reach the desired segmentation in the form of a simple closed curve because of the weakness of edge detector functions in finding the likely target boundary. In this paper, we propose a new regularization model which has the property of encouraging a closed curve by deliberately controlling the curve smoothness. The new model may be combined with various fitting terms to enhance segmentation results. The key features of the proposed model are demonstrated in detail. Numerical simulations and experiments show that the proposed model enhances the segmentation ability for extracting the target boundary as a closed contour.

Multiview fusion 3-D echocardiography: improving the information and quality of real-time 3-D echocardiography

The advent of real-time 3-D echocardiography (RT3DE) promised dynamic 3-D image acquisition with the potential of a more objective and complete functional analysis. In spite of that, 2-D echocardiography remains the backbone of echocardiography imaging in current clinical practice, with RT3DE mainly used for clinical research. The uptake of RT3DE has been slow because of missing anatomic information, limited field-of-view (FOV) and tedious analysis procedures. This paper presents multiview fusion 3D echocardiography, where multiple images with complementary information are acquired from different probe positions. These multiple images are subsequently aligned and fused together for preserving salient structures in a single, multiview fused image. A novel and simple wavelet-based fusion algorithm is proposed that exploits the low- and high-frequency separation capability of the wavelet analysis. The results obtained show that the proposed multiview fusion considerably improves the contrast (31.1%), contrast-to-noise ratio (46.7%), signal-to-noise ratio (44.7%) and anatomic features (12%) in 3-D echocardiography, and enlarges the FOV (28.2%). This indicates that multiview fusion substantially enhances the image quality and information.

Left ventricular border tracking using cardiac motion models and optical flow

The use of automated methods is becoming increasingly important for assessing cardiac function quantitatively and objectively. In this study, we propose a method for tracking three-dimensional (3-D) left ventricular contours. The method consists of a local optical flow tracker and a global tracker, which uses a statistical model of cardiac motion in an optical-flow formulation. We propose a combination of local and global trackers using gradient-based weights. The algorithm was tested on 35 echocardiographic sequences, with good results (surface error: 1.35 ± 0.46 mm, absolute volume error: 5.4 ± 4.8 mL). This demonstrates the method's potential in automated tracking in clinical quality echocardiograms, facilitating the quantitative and objective assessment of cardiac function.

The evaluation of single-view and multi-view fusion 3D echocardiography using image-driven segmentation and tracking

Real-time 3D echocardiography (RT3DE) promises a more objective and complete cardiac functional analysis by dynamic 3D image acquisition. Despite several efforts towards automation of left ventricle (LV) segmentation and tracking, these remain challenging research problems due to the poor-quality nature of acquired images usually containing missing anatomical information, speckle noise, and limited field-of-view (FOV). Recently, multi-view fusion 3D echocardiography has been introduced as acquiring multiple conventional single-view RT3DE images with small probe movements and fusing them together after alignment. This concept of multi-view fusion helps to improve image quality and anatomical information and extends the FOV. We now take this work further by comparing single-view and multi-view fused images in a systematic study. In order to better illustrate the differences, this work evaluates image quality and information content of single-view and multi-view fused images using image-driven LV endocardial segmentation and tracking. The image-driven methods were utilized to fully exploit image quality and anatomical information present in the image, thus purposely not including any high-level constraints like prior shape or motion knowledge in the analysis approaches. Experiments show that multi-view fused images are better suited for LV segmentation and tracking, while relatively more failures and errors were observed on single-view images.

A 3D freehand ultrasound system for multi-view reconstructions from sparse 2D scanning planes

Background

A significant limitation of existing 3D ultrasound systems comes from the fact that the majority of them work with fixed acquisition geometries. As a result, the users have very limited control over the geometry of the 2D scanning planes.

Methods

We present a low-cost and flexible ultrasound imaging system that integrates several image processing components to allow for 3D reconstructions from limited numbers of 2D image planes and multiple acoustic views. Our approach is based on a 3D freehand ultrasound system that allows users to control the 2D acquisition imaging using conventional 2D probes.

For reliable performance, we develop new methods for image segmentation and robust multi-view registration. We first present a new hybrid geometric level-set approach that provides reliable segmentation performance with relatively simple initializations and minimum edge leakage. Optimization of the segmentation model parameters and its effect on performance is carefully discussed. Second, using the segmented images, a new coarse to fine automatic multi-view registration method is introduced. The approach uses a 3D Hotelling transform to initialize an optimization search. Then, the fine scale feature-based registration is performed using a robust, non-linear least squares algorithm. The robustness of the multi-view registration system allows for accurate 3D reconstructions from sparse 2D image planes.

Results

Volume measurements from multi-view 3D reconstructions are found to be consistently and significantly more accurate than measurements from single view reconstructions. The volume error of multi-view reconstruction is measured to be less than 5% of the true volume. We show that volume reconstruction accuracy is a function of the total number of 2D image planes and the number of views for calibrated phantom. In clinical in-vivo cardiac experiments, we show that volume estimates of the left ventricle from multi-view reconstructions are found to be in better agreement with clinical measures than measures from single view reconstructions.

Conclusions

Multi-view 3D reconstruction from sparse 2D freehand B-mode images leads to more accurate volume quantification compared to single view systems. The flexibility and low-cost of the proposed system allow for fine control of the image acquisition planes for optimal 3D reconstructions from multiple views.

Patient-specific modeling and quantification of the aortic and mitral valves from 4-D cardiac CT and TEE

As decisions in cardiology increasingly rely on noninvasive methods, fast and precise image processing tools have become a crucial component of the analysis workflow. To the best of our knowledge, we propose the first automatic system for patient-specific modeling and quantification of the left heart valves, which operates on cardiac computed tomography (CT) and transesophageal echocardiogram (TEE) data. Robust algorithms, based on recent advances in discriminative learning, are used to estimate patient-specific parameters from sequences of volumes covering an entire cardiac cycle. A novel physiological model of the aortic and mitral valves is introduced, which captures complex morphologic, dynamic, and pathologic variations. This holistic representation is hierarchically defined on three abstraction levels: global location and rigid motion model, nonrigid landmark motion model, and comprehensive aortic-mitral model. First we compute the rough location and cardiac motion applying marginal space learning. The rapid and complex motion of the valves, represented by anatomical landmarks, is estimated using a novel trajectory spectrum learning algorithm. The obtained landmark model guides the fitting of the full physiological valve model, which is locally refined through learned boundary detectors. Measurements efficiently computed from the aortic-mitral representation support an effective morphological and functional clinical evaluation. Extensive experiments on a heterogeneous data set, cumulated to 1516 TEE volumes from 65 4-D TEE sequences and 690 cardiac CT volumes from 69 4-D CT sequences, demonstrated a speed of 4.8 seconds per volume and average accuracy of 1.45 mm with respect to expert defined ground-truth. Additional clinical validations prove the quantification precision to be in the range of inter-user variability. To the best of our knowledge this is the first time a patient-specific model of the aortic and mitral valves is automatically estimated from volumetric sequences.

Ultrasound image segmentation and tissue characterization

Ultrasound image segmentation deals with delineating the boundaries of structures, as a step towards semi-automated or fully automated measurement of dimensions or for characterizing tissue regions. Ultrasound tissue characterization (UTC) is driven by knowledge of the physics of ultrasound and its interactions with biological tissue, and has traditionally used signal modelling and analysis to characterize and differentiate between healthy and diseased tissue. Thus, both aim to enhance the capabilities of ultrasound as a quantitative tool in clinical medicine, and the two end goals can be the same, namely to characterize the health of tissue. This article reviews both research topics, and finds that the two fields are becoming more tightly coupled, even though there are key challenges to overcome in each area, influenced by factors such as more open software-based ultrasound system architectures, increased computational power, and advances in imaging transducer design.

Automated border detection in three-dimensional echocardiography: principles and promises

Several automated border detection approaches for three-dimensional echocardiography have been developed in recent years, allowing quantification of a range of clinically important parameters. In this review, the background and principles of these approaches and the different classes of methods are described from a practical perspective, as well as the research trends to achieve a robust method.

Computational hemodynamic modeling based on transesophageal echocardiographic imaging

We address the problem of hemodynamic computational modeling in the left heart complex. The novelty of our approach lies in the exploitation of prior patient specific data resulting from image analysis of Transesophageal Echocardiographic Imagery (TEE). Kinematic and anatomical information in the form of left heart chambers and valve boundaries is recovered through a level-set-based user-in-the-loop segmentation on 2D TEE. The resulting boundaries in the TEE sequence are then interpolated to prescribe the motion displacements in a computational fluid dynamics (CFD) model implemented using Finite Element Modeling (FEM) applied on Arbitrary Lagrangian-Eulerian (ALE) meshes. Experimental results are presented.

Feasibility of left ventricular shape analysis from transthoracic real-time 3-D echocardiographic images

Despite the potential ability of left ventricular (LV) shape analysis to provide independent information complementary to ventricular size and function, in clinical practice only ejection fraction (EF) is currently assessed while LV shape is not routinely quantified. Moreover, geometric assumptions in the computation of EF from multiple two-dimensional (2-D) cut-planes by disc summation or area-length methods, introduce inaccuracies in the estimates. Also, previous approaches for the quantification of LV shape were based on geometric modeling and, as a result, proved inaccurate. Our aims were (1) to develop and test a three-dimensional (3-D) technique for direct quantification of LV shape from real-time 3-D echocardiographic (RT3DE) images without the need for geometric modeling using a new class of LV shape indices; and (2) to study the relationship between these indices and ventricular size and function in normal and abnormal ventricles. Spherical (S), ellipsoidal (E) and conical (C) shape indices were calculated using custom software for analysis of transthoracic RT3DE images on both global and regional levels and initially tested on computer simulated objects of different shapes. The feasibility of using these indices to differentiate between normal and abnormal ventricles was tested in three groups of patients: normal volunteers (NL, n=9), dilated cardiomyopathy (DCM, n=9) and coronary artery disease with apical regional wall motion abnormalities (RWMA, n=9). Computer simulation demonstrated that these shape indices are size-independent and can correctly classify the simulated objects. In human ventricles, both S and C but not E correlated well with LV volumes and EF. Also, S and C changed throughout the cardiac cycle while E remained almost constant. In addition, both regional and global S and C were able to identify differences between NL and abnormal ventricles: normal ventricles were less spherical and more conical than those of patients with DCM at both end-systole and end-diastole (p<0.05) both globally and regionally. In contrast, in patients with RWMA, similar differences were noted only at end-systole, both on a global level and in the apical region. In this study, we demonstrated the feasibility of quantifying LV shape from transthoracic RT3DE images at both global and regional levels. Potentially, such 3-D shape analysis could be combined with conventional evaluation of LV volume and function to provide a more comprehensive assessment of left ventricular performance.

Left Ventricle Segmentation Using Model Fitting and Active Surfaces

A method to perform 4D (3D over time) segmentation of the left ventricle of a mouse heart using a set of B mode cine slices acquired in vivo from a series of short axis scans is described. We incorporate previously suggested methods such as temporal propagation, the gradient vector flow active surface, superquadric models, etc. into our proposed 4D segmentation of the left ventricle. The contributions of this paper are incorporation of a novel despeckling method and the use of locally fitted superellipsoid models to provide a better initialization for the active surface segmentation algorithm. Average distances of the improved surface segmentation to a manually segmented surface throughout the entire cardiac cycle and cross-sectional contours are provided to demonstrate the improvements produced by the proposed 4D segmentation.

Image-driven cardiac left ventricle segmentation for the evaluation of multiview fused real-time 3-dimensional echocardiography images

Real-time 3-dimensional echocardiography (RT3DE) permits the acquisition and visualization of the beating heart in 3D. Despite a number of efforts to automate the left ventricle (LV) delineation from RT3DE images, this remains a challenging problem due to the poor nature of the acquired images usually containing missing anatomical information and high speckle noise. Recently, there have been efforts to improve image quality and anatomical definition by acquiring multiple single-view RT3DE images with small probe movements and fusing them together after alignment. In this work, we evaluate the quality of the multiview fused images using an image-driven semiautomatic LV segmentation method. The segmentation method is based on an edge-driven level set framework, where the edges are extracted using a local-phase inspired feature detector for low-contrast echocardiography boundaries. This totally image-driven segmentation method is applied for the evaluation of end-diastolic (ED) and end-systolic (ES) single-view and multiview fused images. Experiments were conducted on 17 cases and the results show that multiview fused images have better image segmentation quality, but large failures were observed on ED (88.2%) and ES (58.8%) single-view images.

Incorporating temporal information into level set functional for robust ventricular boundary detection from echocardiographic image sequence

Echocardiographic images often suffer from dropouts that lead to loss of signals on the ventricular boundary and cause the level set curve used to detect the boundary leaking out from the gaps on the boundary. In this paper, a novel method that incorporates temporal information into the level set functional is proposed to solve the leakage problem encountered when detecting the heart wall boundary from the echocardiographic image sequence. The ventricular boundary is quantitatively partitioned and classified into strong and weak segments. The weak segments are considered to be weakened by dropouts and there is low confidence on the presence of boundary. Temporal information from neighboring frames is exploited as a regularizer into the level set equation. Hence, the original boundary information in the weak segments can be reconstructed and the curve leakage problem can be remedied. Experimental results demonstrate the advantages of the proposed method for the intended task.

Four-chamber heart modeling and automatic segmentation for 3-D cardiac CT volumes using marginal space learning and steerable features

We propose an automatic four-chamber heart segmentation system for the quantitative functional analysis of the heart from cardiac computed tomography (CT) volumes. Two topics are discussed: heart modeling and automatic model fitting to an unseen volume. Heart modeling is a nontrivial task since the heart is a complex nonrigid organ. The model must be anatomically accurate, allow manual editing, and provide sufficient information to guide automatic detection and segmentation. Unlike previous work, we explicitly represent important landmarks (such as the valves and the ventricular septum cusps) among the control points of the model. The control points can be detected reliably to guide the automatic model fitting process. Using this model, we develop an efficient and robust approach for automatic heart chamber segmentation in 3-D CT volumes. We formulate the segmentation as a two-step learning problem: anatomical structure localization and boundary delineation. In both steps, we exploit the recent advances in learning discriminative models. A novel algorithm, marginal space learning (MSL), is introduced to solve the 9-D similarity transformation search problem for localizing the heart chambers. After determining the pose of the heart chambers, we estimate the 3-D shape through learning-based boundary delineation. The proposed method has been extensively tested on the largest dataset (with 323 volumes from 137 patients) ever reported in the literature. To the best of our knowledge, our system is the fastest with a speed of 4.0 s per volume (on a dual-core 3.2-GHz processor) for the automatic segmentation of all four chambers.

Semi-automated analysis of dynamic changes in myocardial contrast from real-time three-dimensional echocardiographic images as a basis for volumetric quantification of myocardial perfusion

AIMS

Despite the potential of real-time three-dimensional (3D) echocardiography (RT3DE) to assess myocardial perfusion, there is no quantification method available for perfusion analysis from RT3DE images. Such method would require 3D regions of interest (ROI) to be defined and adjusted frame-by-frame to compensate for cardiac translation and deformation. Our aims were to develop and test a technique for automated identification of 3D myocardial ROI suitable for translation-free quantification of myocardial videointensity over time, MVI(t), from contrast-enhanced RT3DE images.

METHODS AND RESULTS

Twelve transthoracic RT3DE (Philips) data sets obtained in pigs during transition from no contrast to steady-state enhancement (Definity) were analysed using custom software. Analysis included: (i) semi-automated detection of left ventricular endo- and epicardial surfaces using level-set techniques in one frame to define a 3D myocardial ROI, (ii) rigid 3D registration to reduce translation and rotation, (iii) elastic 3D registration to compensate for deformation, and (iv) quantification of MVI(t) in the 3D ROI from the registered and non-registered data sets to assess the effectiveness of registration. For each MVI(t) curve we computed % variability during steady-state enhancement (100 x SD/mean) and goodness of fit (r2) to the indicator dilution equation MVI(t) = A[1-exp(-betat)]. Analysis of myocardial contrast throughout contrast inflow was feasible in all data sets. Three-dimensional registration improved MVI(t) curves in terms of both % variability (2.8 +/- 1.8 to 1.5 +/- 0.9%; P < 0.05) and goodness of fit (r2 from 0.79 +/- 0.2 to 0.90 +/- 0.1; P < 0.05).

CONCLUSION

This is the first study to describe a new technique for semi-automated volumetric quantification of myocardial contrast from RT3DE images that includes registration and thus provides the basis for 3D measurement of myocardial perfusion.

Time continuous detection of the left ventricular long axis and the mitral valve plane in 3-D echocardiography

Automated segmentation approaches for the left ventricle (LV) in 3-D echocardiography (3DE) often rely on manual initialization. So far, little effort has been put into automating the initialization procedure to get to a fully automatic segmentation approach. We propose a fully automatic method for the detection of the LV long axis (LAX) and the mitral valve plane (MVP) over the full cardiac cycle, for the initialization of segmentation algorithms in 3DE. Our method exploits the cyclic motion of the LV and therefore detects salient structures in a time-continuous way. Probabilities to candidate LV center points are assigned through a Hough transform for circles. The LV LAX is detected by combining dynamic programming detections on these probabilities in 3-D and 2D + time to obtain a time continuous solution. Subsequently, the mitral valve plane is detected in a projection of the data on a plane through the previously detected LAX. The method easily adjusts to different acquisition routines and combines robustness with good accuracy and low computational costs. Automatic detection was evaluated using patient data acquired with the fast rotating ultrasound (FRU) transducer (n=11 patients) and with the Philips Sonos 7500 ultrasound system (Philips Medical Systems, Andover, MA, USA), with the X4 matrix transducer (n=14 patients). For the FRU-transducer data, the LAX was estimated with a distance error of 2.85+/-1.70 mm (mean+/-SD) and an angle of 5.25+/-3.17 degrees; the mitral valve plane was estimated with a distance of -1.54+/-4.31 mm. For the matrix data, these distances were 1.96+/-1.30 mm with an angle error of 5.95+/-2.11 and -1.66+/-5.27 mm for the mitral valve plane. These results confirm that the method is very suitable for automatic detection of the LV LAX and MVP. It provides a basis for further automatic exploration of the LV and could therefore serve as a replacement of manual initialization of 3-D segmentation approaches.

Real-Time Three-Dimensional Echocardiographic Acquisition and Quantification of Left Ventricular Indices in Children and Young Adults with Congenital Heart Disease: Comparison with Magnetic Resonance Imaging

BACKGROUND

Echocardiographic assessment of left ventricular (LV) contractility and dimensions is important in the management of patients with congenital heart disease. Conventional two-dimensional measures are limited because of volume or pressure-overloaded right ventricles that may distort the septal planes. Real-time three-dimensional echocardiography (RT3DE) has overcome these limitations; however, postprocessing image reconstruction and analysis are required. We compared LV indices calculated by new online RT3DE software with those obtained by magnetic resonance imaging (MRI) in patients with congenital heart disease.

METHODS

Twelve patients (ages 1-33 years, median age = 15.9 years) with congenital heart disease underwent RT3DE and cardiac MRI. End-diastolic and end-systolic LV volumes, stroke volume, ejection fraction, and mass were calculated online using biplane method-of-discs and semiautomated border detection echocardiographic techniques.

RESULTS

All RT3DE volumes correlated strongly with MRI (r = 0.93-0.99, P < .001). Ejection fraction had a lower correlation (r = 0.69, P = .013). There was no significant underestimation or overestimation of MRI values by RT3DE. Both biplane method-of-discs and semiautomated border detection echocardiographic techniques had excellent volume correlation (r = 0.94-0.99, P < .001). Interobserver variability was 7%.

CONCLUSIONS

Combined RT3DE acquisition and analysis machines can accurately assess the LV in patients with congenital heart disease, thus impacting clinical management and perhaps obviating the need for MRI in some cases.

Volumetric analysis of regional left ventricular function with real-time three-dimensional echocardiography: validation by magnetic resonance and clinical utility testing

BACKGROUND

Quantitative information on regional left ventricular volumes from real-time three-dimensional echocardiographic (RT3DE) images has significant clinical potential but needs validation.

AIM

To validate these measurements against cardiac magnetic resonance (CMR) and test the feasibility of automated detection of regional wall motion (RWM) abnormalities from RT3DE data.

METHODS

RT3DE (Philips) and CMR (Siemens) images were obtained from 31 patients and analysed by using prototype software to semiautomatically calculate indices of regional left ventricular function, which were compared between RT3DE and CMR (linear regression, Bland-Altman). Additionally, CMR images were reviewed by an expert, whose RWM grades were used as a reference for automated classification of segments as normal or abnormal from RT3DE and from CMR images. For each modality, normal regional ejection fraction (REF) values were obtained from 15 patients with normal wall motion. In the remaining 16 patients, REFs were compared with thresholds that were derived from patients with normal wall motion and optimised using receiver operating characteristic analysis.

RESULTS

RT3DE measurements resulted in good agreement with CMR. Regional indices calculated in patients with normal wall motion varied between segments, but overall were similar between modalities. In patients with abnormal wall motion, RWM was graded as abnormal in 74% segments. CMR and RT3DE thresholds were similar (16-segment average 55 (10)% and 56 (7)%, respectively). Automated interpretation resulted in good agreement with expert interpretation, similar for CMR and RT3DE (sensitivity 0.85, 0.84; specificity 0.81, 0.78; accuracy 0.84, 0.84, respectively).

CONCLUSION

Analysis of RT3DE data provides accurate quantification of regional left ventricular function and allows semiautomated detection of RWM abnormalities, which is as accurate as the same algorithm applied to CMR images.

Quantitative assessment of left ventricular size and function: side-by-side comparison of real-time three-dimensional echocardiography and computed tomography with magnetic resonance reference

BACKGROUND

Cardiac CT (CCT) and real-time 3D echocardiography (RT3DE) are being used increasingly in clinical cardiology. CCT offers superb spatial and contrast resolution, resulting in excellent endocardial definition. RT3DE has the advantages of low cost, portability, and live 3D imaging without offline reconstruction. We sought to compare both CCT and RT3DE measurements of left ventricular size and function with the standard reference technique, cardiac MR (CMR).

METHODS AND RESULTS

In 31 patients, RT3DE data sets (Philips 7500) and long-axis CMR (Siemens, 1.5 T) and CCT (Toshiba, 16-slice MDCT) images were obtained on the same day without beta-blockers. All images were analyzed to obtain end-systolic and end-diastolic volumes and ejection fractions using the same rotational analysis to eliminate possible analysis-related differences. Intertechnique agreement was tested through linear regression and Bland-Altman analyses. Repeated measurements were performed to determine intraobserver and interobserver variability. Both CCT and RT3DE measurements resulted in high correlation (r2 > 0.85) compared with CMR. However, CCT significantly overestimated end-diastolic and end-systolic volumes (26 and 19 mL; P < 0.05), resulting in a small but significant bias in ejection fraction (-2.8%). RT3DE underestimated end-diastolic and end-systolic volumes only slightly (5 and 6 mL), with no significant bias in EF (0.3%; P = 0.68). The limits of agreement with CMR were comparable for the 2 techniques. The variability in the CCT measurements was roughly half of that in either RT3DE or CMR values.

CONCLUSIONS

CCT provides highly reproducible measurements of left ventricular volumes, which are significantly larger than CMR values. RT3DE measurements compared more favorably with the CMR reference, albeit with higher variability.

Ultrasound image segmentation: a survey

This paper reviews ultrasound segmentation paper methods, in a broad sense, focusing on techniques developed for medical B-mode ultrasound images. First, we present a review of articles by clinical application to highlight the approaches that have been investigated and degree of validation that has been done in different clinical domains. Then, we present a classification of methodology in terms of use of prior information. We conclude by selecting ten papers which have presented original ideas that have demonstrated particular clinical usefulness or potential specific to the ultrasound segmentation problem.

Dual triggering improves the accuracy of left ventricular volume measurements by contrast-enhanced real-time 3-dimensional echocardiography

Real-time 3-dimensional echocardiographic continuous imaging (CIM) with contrast underestimates left ventricular (LV) volumes. We studied the effects of dual-triggered (DT) acquisition on the accuracy of LV volume measurements for patients with poor acoustic windows. Real-time 3-dimensional echocardiographic imaging was performed in 20 patients during LV opacification (Definity) on the same day as cardiac magnetic resonance imaging. Both CIM and DT data were analyzed using custom software to calculate end-systolic volume (ESV) and end-diastolic volume (EDV), which were compared with the cardiac magnetic resonance reference. CIM correlated well with the cardiac magnetic resonance reference (EDV: r = 0.89; ESV: r = 0.93), but underestimated EDV and ESV by 17% and 19%, respectively. In contrast, DT resulted in higher correlation (EDV: r = 0.95; ESV: r = 0.96) and smaller biases (9% and 6%, respectively). In conclusion, because the accuracy of LV volume measurements depends on the acquisition strategy of contrast-enhanced real-time 3-dimensional echocardiographic images, the use of DT instead of the conventional CIM acquisition is recommended.

Segmentation of elastographic images using a coarse-to-fine active contour model

Delineation of radiofrequency-ablation-induced coagulation (thermal lesion) boundaries is an important clinical problem that is not well addressed by conventional imaging modalities. Elastography, which produces images of the local strain after small, externally applied compressions, can be used for visualization of thermal coagulations. This paper presents an automated segmentation approach for thermal coagulations on 3-D elastographic data to obtain both area and volume information rapidly. The approach consists of a coarse-to-fine method for active contour initialization and a gradient vector flow, active contour model for deformable contour optimization with the help of prior knowledge of the geometry of general thermal coagulations. The performance of the algorithm has been shown to be comparable to manual delineation of coagulations on elastograms by medical physicists (r = 0.99 for volumes of 36 radiofrequency-induced coagulations). Furthermore, the automatic algorithm applied to elastograms yielded results that agreed with manual delineation of coagulations on pathology images (r = 0.96 for the same 36 lesions). This algorithm has also been successfully applied on in vivo elastograms.

Improved quantification of left ventricular mass based on endocardial and epicardial surface detection with real time three dimensional echocardiography

OBJECTIVE

To develop a technique for volumetric analysis of real time three dimensional echocardiography (RT3DE) data aimed at quantifying left ventricular (LV) mass and to validate the technique against magnetic resonance (MR) assumed as the reference standard.

DESIGN

RT3DE, which has recently become widely available, provides dynamic pyramidal data structures that encompass the entire heart and allows four dimensional assessment of cardiac anatomy and function. However, analysis techniques for the quantification of LV mass from RT3DE data are fundamentally two dimensional, rely on geometric modelling, and do not fully exploit the volumetric information contained in RT3DE datasets. Twenty one patients underwent two dimensional echocardiography (2DE), RT3DE, and cardiac MR. LV mass was measured from 2DE and MR images by conventional techniques. RT3DE data were analysed to semiautomatically detect endocardial and epicardial LV surfaces by the level set approach. From the detected surfaces, LV mass was computed directly in the three dimensional space as voxel counts.

RESULTS

RT3DE measurement was feasible in 19 of 21 patients and resulted in higher correlation with MR (r = 0.96) than did 2DE (r = 0.79). RT3DE measurements also had a significantly smaller bias (-2.1 g) and tighter limits of agreement (2SD = +/-23 g) with MR than did the 2DE values (bias (2SD) -34.9 (50) g). Additionally, interobserver variability of RT3DE (12.5%) was significantly lower than that of 2DE (24.1%).

CONCLUSIONS

Direct three dimensional model independent LV mass measurement from RT3DE images is feasible in the clinical setting and provides fast and accurate assessment of LV mass, superior to the two dimensional analysis techniques.

Cell-competition algorithm: a new segmentation algorithm for multiple objects with irregular boundaries in ultrasound images

Segmentation of multiple objects with irregular contours and surrounding sporadic spots is a common practice in ultrasound image analysis. A new region-based approach, called cell-competition algorithm, is proposed for simultaneous segmentation of multiple objects in a sonogram. The algorithm is composed of two essential ideas. One is simultaneous cell-based deformation of regions and the other is cell competition. The cells are generated by two-pass watershed transformations. The cell-competition algorithm has been validated with 13 synthetic images of different contrast-to-noise ratios and 71 breast sonograms. Three assessments have been carried out and the results show that the boundaries derived by the cell-competition algorithm are reasonably comparable to those delineated manually. Moreover, the cell-competition algorithm is robust to the variation of regions-of-interest and a range of thresholds required for the second-pass watershed transformation. The proposed algorithm is also shown to be superior to the region-competition algorithm for both types of images.

Left ventricular volume estimation in cardiac three-dimensional ultrasound: a semiautomatic border detection approach

RATIONALE AND OBJECTIVES

We propose a semiautomatic endocardial border detection method for three-dimensional (3D) time series of cardiac ultrasound (US) data based on pattern matching and dynamic programming, operating on two-dimensional (2D) slices of the 3D plus time data, for the estimation of full cycle left ventricular volume, with minimal user interaction.

MATERIALS AND METHODS

The presented method is generally applicable to 3D US data and evaluated on data acquired with the Fast Rotating Ultrasound (FRU-) Transducer, developed by Erasmus Medical Center (Rotterdam, the Netherlands), a conventional phased-array transducer, rotating at very high speed around its image axis. The detection is based on endocardial edge pattern matching using dynamic programming, which is constrained by a 3D plus time shape model. It is applied to an automatically selected subset of 2D images of the original data set, for typically 10 equidistant rotation angles and 16 cardiac phases (160 images). Initialization requires the drawing of four contours per patient manually. We evaluated this method on 14 patients against MRI end-diastole and end-systole volumes. Initialization requires the drawing of four contours per patient manually. We evaluated this method on 14 patients against MRI end-diastolic (ED) and end-systolic (ES) volumes.

RESULTS

The semiautomatic border detection approach shows good correlations with MRI ED/ES volumes (r = 0.938) and low interobserver variability (y = 1.005x - 16.7, r = 0.943) over full-cycle volume estimations. It shows a high consistency in tracking the user-defined initial borders over space and time.

CONCLUSIONS

We show that the ease of the acquisition using the FRU-transducer and the semiautomatic endocardial border detection method together can provide a way to quickly estimate the left ventricular volume over the full cardiac cycle using little user interaction.

Registration-assisted segmentation of real-time 3-D echocardiographic data using deformable models

Real-time three-dimensional (3-D) echocardiography is a new imaging modality that presents the unique opportunity to visualize the complex 3-D shape and motion of the left ventricle (LV) in vivo and to measure the associated global and local function parameters. To take advantage of this opportunity in routine clinical practice, automatic segmentation of the LV in the 3-D echocardiographic data, usually hundreds of megabytes large, is essential. We report a new segmentation algorithm for this task. Our algorithm has two distinct stages, initialization of a deformable model and its refinement, which are connected by a dual "voxel + wiremesh" template. In the first stage, mutual-information-based registration of the voxel template with the image to be segmented helps initialize the wiremesh template. In the second stage, the wiremesh is refined iteratively under the influence of external and internal forces. The internal forces have been customized to preserve the nonsymmetric shape of the wiremesh template in the absence of external forces, defined using the gradient vector flow approach. The algorithm was validated against expert-defined segmentation and demonstrated acceptable accuracy. Our segmentation algorithm is fully automatic and has the potential to be used clinically together with real-time 3-D echocardiography for improved cardiovascular disease diagnosis.

Segmentation of real-time three-dimensional ultrasound for quantification of ventricular function: a clinical study on right and left ventricles

Among screening modalities, echocardiography is the fastest, least expensive and least invasive method for imaging the heart. A new generation of three-dimensional (3-D) ultrasound (US) technology has been developed with real-time 3-D (RT3-D) matrix phased-array transducers. These transducers allow interactive 3-D visualization of cardiac anatomy and fast ventricular volume estimation without tomographic interpolation as required with earlier 3-D US acquisition systems. However, real-time acquisition speed is performed at the cost of decreasing spatial resolution, leading to echocardiographic data with poor definition of anatomical structures and high levels of speckle noise. The poor quality of the US signal has limited the acceptance of RT3-D US technology in clinical practice, despite the wealth of information acquired by this system, far greater than with any other existing echocardiography screening modality. We present, in this work, a clinical study for segmentation of right and left ventricular volumes using RT3-D US. A preprocessing of the volumetric data sets was performed using spatiotemporal brushlet denoising, as presented in previous articles Two deformable-model segmentation methods were implemented in 2-D using a parametric formulation and in 3-D using an implicit formulation with a level set implementation for extraction of endocardial surfaces on denoised RT3-D US data. A complete and rigorous validation of the segmentation methods was carried out for quantification of left and right ventricular volumes and ejection fraction, including comparison of measurements with cardiac magnetic resonance imaging as the reference. Results for volume and ejection fraction measurements report good performance of quantification of cardiac function on RT3-D data compared with magnetic resonance imaging with better performance of semiautomatic segmentation methods than with manual tracing on the US data.