Marijn van Stralen

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 methods potential in automated tracking in clinical quality echocardiograms, facilitating the quantitative and objective assessment of cardiac function.

Automatic active appearance model segmentation of 3D echocardiograms

A fully automated segmentation for 3D echocardiography (3DE) using 3D Active Appearance Models (AAM) was developed and evaluated on 99 patients. The method used ultrasound specific grey value normalization and two matching algorithms were tested. To our knowledge this is the first report on a fully operational 3D AAM employed in 3DE on a large scale. The 3D AAM detected the endocardial contours accurately, even in the presence of large variations in left ventricular appearance and shape. Matching was successful in 91% of patients and resulted in a median point-tosurface error of 2.69 mm (av±sd: 2.91±1.03mm). Results indicate that fully automated AAM analysis of 3DE is practically feasible in datasets of mixed origin and quality.

Side-by-side viewing of anatomically aligned left ventricular segments in three-dimensional stress echocardiography.

Background: Dobutamine stress echocardiography (DSE) suffers from high interobserver and inter‐institution variability in the diagnosis of myocardial ischemia. Therefore, we developed a three‐dimensional (3D) analysis tool that makes it possible to anatomically align 3D rest and stress data systematically, to generate optimal, nonforeshortened standard anatomical cross sections and to analyse the images synchronized and side‐by‐side. Aim of the study: To investigate whether this 3D analysis tool could improve interobserver agreement on myocardial ischemia during 3D DSE. Methods: The study comprised 34 consecutive patients with stable chest pain who underwent both noncontrast and contrast 3D DSE. Two observers scored segmental wall motion using a conventional analysis and the novel analysis with the new 3D tool. Results: The two observers agreed on the presence or absence of myocardial ischemia in 81 of 102 coronary territories (agreement 79%, kappa (κ) 0.28) during noncontrast 3D imaging and 92 of 102 coronary territories (agreement 90%, kappa 0.65) during contrast‐enhanced 3D imaging. With the new 3D analysis software these numbers improved to 98 of 102 coronary territories (agreement 96%, kappa 0.69) during noncontrast 3D imaging and 98 of 102 coronary territories (agreement 96%, kappa 0.82) during contrast‐enhanced 3D imaging. Conclusion: The use of a 3D DSE analysis tool improves interobserver agreement for myocardial ischemia both for noncontrast and contrast images.

Model driven quantification of left ventricular function from sparse single-beat 3D echocardiography

This paper presents a novel model based segmentation technique for quantification of left ventricular (LV) function from sparse single-beat 3D echocardiographic data acquired with a fast rotating ultrasound (FRU) transducer. This transducer captures cardiac anatomy in a sparse set of radially sampled, curved cross-sections within a single cardiac cycle. The method employs a 3D Active Shape Model of the left ventricle (LV) in combination with local appearance models as prior knowledge to steer the segmentation. A set of local appearance patches generate the model update points for fitting the model to the LV in the curved FRU cross-sections. Updates are then propagated over the dense 3D model mesh to overcome correspondence problems due to the data sparsity, whereas the 3D Active Shape Model serves to retain the plausibility of the generated shape. Leave-one-out cross-validation was carried out on single-beat FRU data from 28 patients suffering from various cardiac pathologies. Detection succeeded in 24 cases, and failed in 4 cases due to large dropouts in echo signal. For the successful 24 cases, detection yielded Point to Point errors of 3.1+/-1.1mm, Point to Surface errors of 1.7+/-0.9mm and an EF error of 7.3+/-4.9%. Comparison of fitting on single-beat versus denser multi-beat data showed a similar performance for both types of data irrespective of frame angles of the intersections. Robustness tests with respect to different model initializations showed acceptable performance for initial positions within a range of 26mm for displacement and 12 degrees for orientation. Furthermore, a comparison study between the proposed method and global LV function measured from MR studies of the same patients showed an underestimation of volumes estimated from echocardiographic data compared to MR derived volumes, similar to other results reported in literature. All experiments demonstrate that the proposed method combines robustness with respect to initialization with an acceptable accuracy, while using sparse single-beat FRU data.

Probabilistic framework for tracking in artifact-prone 3D echocardiograms

The analysis of echocardiograms, whether visual or automated, is often hampered by ultrasound artifacts which obscure the moving myocardial wall. In this study, a probabilistic framework for tracking the endocardial surface in 3D ultrasound images is proposed, which distinguishes between visible and artifact-obscured myocardium. Motion estimation of visible myocardium relies more on a local, data-driven tracker, whereas tracking of obscured myocardium is assisted by a global, statistical model of cardiac motion. To make this distinction, the expectation-maximization algorithm is applied in a stationary and dynamic frame-of-reference. Evaluation on 35 three-dimensional echocardiographic sequences shows that this artifact-aware tracker gives better results than when no distinction is made. In conclusion, the proposed tracker is able to reduce the influence of artifacts, potentially improving quantitative analysis of clinical quality echocardiograms.

Left Ventricle Segmentation from Contrast Enhanced Fast Rotating Ultrasound Images Using Three Dimensional Active Shape Models

In this paper we propose a novel segmentation technique for quantification of sparsely sampled single-beat 3D contrast enhanced echocardiographic data acquired with a Fast Rotating Ultrasound transducer (FRU). The method uses a 3D Active Shape Model of the Left Ventricle (LV) in combination with local appearance models as prior knowledge to steer the segmentation. From a set of semi-manually delineated contours, 3D meshes of the LV endocardium are constructed for different cardiac phases. Mesh surfaces are partitioned into a fixed number of regions, each of which is modeled by a local image appearance. During segmentation, model update points are generated based on similarity matches with these local appearance models in multiple curved 2D cross-sections, which are then propagated over a dense 3D mesh. The Active Shape Model effectively constrains the shape of the 3D mesh to a statistically plausible cardiac shape. Leave-one-out cross validation was carried out on single-beat contrast enhanced FRU data from 18 patients suffering from various cardiac pathologies. Experiments show that the proposed method generates segmentation results that agree with the ground truth contours with average Point to Point (P2P) error of 4.1±2.0 mm and average Point to Surface (P2S) error of 2.4±2.1mm. Convergence tests show that the proposed method is capable of producing acceptable segmentation results (with less than 1.5X error compared to favorable initialization) within the range of 18~22 mm of in-plane displacement and 12~14 degrees of long-axial orientation error.

Tracking left ventricular borders in 3D echocardiographic sequences using motion-guided optical flow

For obtaining quantitative and objective functional parameters from three-dimensional (3D) echocardiographic sequences, automated segmentation methods may be preferable to cumbersome manual delineation of 3D borders. In this study, a novel optical-flow based tracking method is proposed for propagating 3D endocardial contours of the left ventricle throughout the cardiac cycle. To take full advantage of the time-continuous nature of cardiac motion, a statistical motion model was explicitly embedded in the optical flow solution. The cardiac motion was modeled as frame-to-frame affine transforms, which were extracted using Procrustes analysis on a set of training contours. Principal component analysis was applied to obtain a compact model of cardiac motion throughout the whole cardiac cycle. The parameters of this model were resolved in an optical flow manner, via spatial and temporal gradients in image intensity. The algorithm was tested on 36 noncontrast and 28 contrast enhanced 3D echocardiographic sequences in a leave-one-out manner. Good results were obtained using a combination of the proposed motion-guided method and a purely data-driven optical flow approach. The improvement was particularly noticeable in areas where the LV wall was obscured by image artifacts. In conclusion, the results show the applicability of the proposed method in clinical quality echocardiograms.

Sparse appearance model based registration of 3D ultrasound images

In this paper, we propose a sparse appearance model based registration algorithm for segmenting 3D echocardiograms. The end-diastolic model is built in 3D sparsely on 2D planes, representing the anatomical 4-chamber, 2-chamber, and short-axis views. Ultrasound specific intensity normalization and shape-based intensity modeling are employed. The model is matched in an intensity-based registration approach, by perturbing appearance and pose parameters simultaneously. Leave-one-out experiments on 10 patients reveal significant improvement in the segmentation using the normalized cross-correlation metric. The registration method will allow fully automatic extraction of the standard views as used in echocardiography. This will aid in the selection of images for inter- and intra-patient comparison and may provide an alternative for a complete 3D AAM.

Comparison of spatiotemporal interpolators for 4D image reconstruction from 2D transesophageal ultrasound

°For electrophysiology intervention monitoring, we intend to reconstruct 4D ultrasound (US) of structures in the beating heart from 2D transesophageal US by scanplane rotation. The image acquisition is continuous but unsynchronized to the heart rate, which results in a sparsely and irregularly sampled dataset and a spatiotemporal interpolation method is desired. Previously, we showed the potential of normalized convolution (NC) for interpolating such datasets. We explored 4D interpolation by 3 different methods: NC, nearest neighbor (NN), and temporal binning followed by linear interpolation (LTB). The test datasets were derived by slicing three 4D echocardiography datasets at random rotation angles (θ, range: 0-180) and random normalized cardiac phase (τ, range: 0-1). Four different distributions of rotated 2D images with 600, 900, 1350, and 1800 2D input images were created from all TEE sets. A 2D Gaussian kernel was used for NC and optimal kernel sizes (σθ and στ) were found by performing an exhaustive search. The RMS gray value error (RMSE) of the reconstructed images was computed for all interpolation methods. The estimated optimal kernels were in the range of σθ = 3.24 - 3.69°/ στ = 0.045 - 0.048, σθ = 2.79°/ στ = 0.031 - 0.038, σθ = 2.34°/ στ = 0.023 - 0.026, and σθ = 1.89°/ στ = 0.021 - 0.023 for 600, 900, 1350, and 1800 input images respectively. We showed that NC outperforms NN and LTB. For a small number of input images the advantage of NC is more pronounced.

Model Driven Quantification of Left Ventricular Function from Sparse Single-beat 3D Echocardiography

This paper presents a novel model based segmentation technique for quantification of Left Ventricular (LV) function from sparse single-beat 3D echocardiographic data acquired with a Fast Rotating Ultrasound (FRU) transducer. This transducer captures cardiac anatomy in a sparse set of radially sampled, curved cross sections within a single cardiac cycle. The method employs a 3D Active Shape Model of the Left Ventricle (LV) in combination with local appearance models as prior knowledge to steer the segmentation. A set of local appearance patches generate the model update points for fitting the model to the LV in the curved FRU cross-sections. Updates are then propagated over the dense 3D model mesh to overcome correspondence problems due to the data sparsity, whereas the 3D Active Shape Model serves to retain the plausibility of the generated shape. Leave-one-out cross validation was carried out on single-beat FRU data from 28 patients suffering from various cardiac pathologies. Error measurements against expert-annotated contours yielded an average point-to-point distance of around 3.8 ± 2.4 mm and point-to-surface distance of 2.0 ± 1.8 mm and average volume estimation error of around 9 ± 7%. Robustness tests with respect to different model initializations showed acceptable performance for initial positions within a range of 22 mm for displacement and 12° for orientation. This demonstrates that the method combines robustness with respect to initialization with an acceptable accuracy, while using sparse single-beat FRU data.