M. van Stralen

P2A-6 Automatic Segmentation of the Left Ventricle in 3D Echocardiography Using Active Appearance Models

Assessment of left ventricular (LV) functional parameters, such as LV volume, ejection fraction and stroke volume, from real-time 3D echocardiography (RT3DE) is labor intensive and subjective, because in current analyses it requires input from the user. Automating these procedures will save valuable time in the analysis and will remove interobserver variability. We propose a fully automatic segmentation approach for the left ventricle in real-time 3D echocardiography, based on active appearance models (AAMs), using ultrasound specific grey value normalization. We evaluated shape and texture model generalization. Also, automatic segmentation has been preliminarily evaluated on transthoracic, apical acquisitions of 54 patients, acquired with the fast rotating ultrasound (FRU-) transducer (18 patients) and with the Philips Sonos 7500 (36 patients). The evaluations were done in a leave-N-out manner (with N=5). We evaluated point-to-surface (P2S) distances for the segmented endocardial contours to the expert manual contours. The generalization of the shape model was good, but texture model generalization was moderate, hampering the AAM matching We found preliminary segmentation errors (P2S) of 3.9plusmn 1.6 mm (N=54) for detection using AAM matching These results indicate that fully automatic segmentation of the LV in RT3DE using AAMs is feasible.

Segmentation of multiple heart cavities in 3-D transesophageal ultrasound images

Three-dimensional transesophageal echocardiography (TEE) is an excellent modality for real-time visualization of the heart and monitoring of interventions. To improve the usability of 3-D TEE for intervention monitoring and catheter guidance, automated segmentation is desired. However, 3-D TEE segmentation is still a challenging task due to the complex anatomy with multiple cavities, the limited TEE field of view, and typical ultrasound artifacts. We propose to segment all cavities within the TEE view with a multi-cavity active shape model (ASM) in conjunction with a tissue/blood classification based on a gamma mixture model (GMM). 3-D TEE image data of twenty patients were acquired with a Philips X7-2t matrix TEE probe. Tissue probability maps were estimated by a two-class (blood/tissue) GMM. A statistical shape model containing the left ventricle, right ventricle, left atrium, right atrium, and aorta was derived from computed tomography angiography (CTA) segmentations by principal component analysis. ASMs of the whole heart and individual cavities were generated and consecutively fitted to tissue probability maps. First, an average whole-heart model was aligned with the 3-D TEE based on three manually indicated anatomical landmarks. Second, pose and shape of the whole-heart ASM were fitted by a weighted update scheme excluding parts outside of the image sector. Third, pose and shape of ASM for individual heart cavities were initialized by the previous whole heart ASM and updated in a regularized manner to fit the tissue probability maps. The ASM segmentations were validated against manual outlines by two observers and CTA derived segmentations.

Determination of a facial nerve safety zone for navigated temporal bone surgery.

BACKGROUND: Transtemporal approaches require surgeons to drill the temporal bone to expose target lesions while avoiding the critical structures within it, such as the facial nerve and other neurovascular structures. We envision a novel protective neuronavigation system that continuously calculates the drill tip-to-facial nerve distance intraoperatively and produces audiovisual warnings if the surgeon drills too close to the facial nerve. Two major problems need to be solved before such a system can be realized. OBJECTIVE: To solve the problems of (1) facial nerve segmentation and (2) calculating a safety zone around the facial nerve in relation to drill-tip tracking inaccuracies. METHODS: We developed a new algorithm called NerveClick for semiautomatic segmentation of the intratemporal facial nerve centerline from temporal bone computed tomography images. We evaluated NerveClicks accuracy in an experimental setting of neuro-otologic and neurosurgical patients. Three neurosurgeons used it to segment 126 facial nerves, which were compared with the gold standard: manually segmented facial nerve centerlines. The centerlines are used as a central axis around which a tubular safety zone is built. The zones thickness incorporates the drill tip tracking errors. The system will warn when the tracked tip crosses the safety zone. RESULTS: Neurosurgeons using NerveClick could segment facial nerve centerlines with a maximum error of 0.44 ± 0.23 mm (mean ± standard deviation) on average compared with manual segmentations. CONCLUSION: Neurosurgeons using our new NerveClick algorithm can robustly segment facial nerve centerlines to construct a facial nerve safety zone, which potentially allows timely audiovisual warnings during navigated temporal bone drilling despite tracking inaccuracies.

Improved spatiotemporal voxel space interpolation for 3D echocardiography with irregular sampling and multibeat fusion

We developed a novel multi-beat image fusion technique using a special spatiotemporal interpolation for sparse, irregularly sampled data (ISI). It is applied to irregularly distributed 3D cardiac ultrasound data acquired with the fast rotating ultrasound (FRU) transducer developed in our laboratory, a phased array rotating mechanically at very high speed (240-480rpm). High-quality 2D images are acquired at ~100 frames/s over 5-10 seconds. ISI was compared quantitatively to spatiotemporal nearest neighbor interpolation (STNI) on synthetic data and compared qualitatively to classical trilinear voxel interpolation on 10 in-vivo cardiac image sets. ISI showed considerably lower absolute distance errors than STNI. For in-vivo images, ISI voxel sets showed reduced motion artifacts, suppression of noise and interpolation artifacts and better delineation of endocardium. In conclusion, ISI improves the quality of 3D+T images acquired with a fast rotating transducer in simulated and in-vivo data.

Rapid 3D Transesophageal Echocardiography using a fast-rotating multiplane transducer

3D transesophageal echocardiography (3D TEE) with acquisition gating for electrocardiogram (ECG) and respiration is slow, cumbersome for the patient and prone to motion artifacts. We realized a rapid 3D TEE solution based on a standard multiplane TEE probe, extended with a fast-rotating transducer array (FR-TEE). The fast left-right rotation allows acquisition of sufficient image data from the entire rotation range for the full heart cycle within one breath-hold. No ECG- or respiration-gating is applied. In normal mode, the probe has uncompromised optimal 2D quality. 10 seconds of image data with ECG and angle values are recorded and post-processed with specially developed 4D reconstruction software based on normalized convolution interpolation. High quality 3D images of phantoms were acquired, accurately depicting the imaged objects. Sequences of reconstructed 3D volumes of a cyclic moving (4D) balloon phantom show only minimal temporal artifacts. Preliminary results on 5 open-chest pigs and 3 humans showed the overall anatomy as well as valvular details with good diagnostic accuracy and high temporal and spatial resolution. A bicuspid aortic valve was diagnosed from the 3D reconstructions and confirmed by a separate 2D exam, proving the 3D diagnostic capabilities of FR-TEE.

Spatiotemporal interpolation by normalized convolution for 4D transesophageal echocardiography

For interventional monitoring, we aim at 4D ultrasound reconstructions of structures in the beating heart from 2D transesophageal echo images by fast scan plane rotation, unsynchronized to the heart rate. For such sparsely and irregularly sampled 2D images, a special spatiotemporal interpolation approach is desired. We have previously shown the potential of spatiotemporal interpolation by normalized convolution (NC). In this work we optimized NC for our application and compared it to nearest neighbor interpolation (NN) and to temporal binning followed by linear spatial interpolation (LTB). The test datasets consisted of 600, 1350, and 1800 2D images and were derived by slicing a 4D echocardiography data sets at random rotation angle (θ, range: 0-180°) and random normalized cardiac phase (τ, range: 0-1). A Gaussian kernel was used for NC and optimal kernel sizes (σ τ and σ θ) 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 σ θ=3.24°/ σ τ=0.048, σ θ=2.34°/σ τ=0.026, and σ θ=1.89°/σ τ=0.023 for 600, 1350, and 1800 input images, respectively. The minimum RMSE for NC was 13.8, 10.4, and 9.4 for 600, 1350, and 1800 input images, respectively. The NN/LTB reconstruction had an RMSE of 17.8/16.4, 13.9/15.1, and 12.0/14.7 for 600, 1350, and 1800 2D 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.

P2A-8 Fully Automatic Detection of Left Ventricular Long Axis and Mitral Valve Plane in 3D Echocardiography

Automated segmentation approaches for the left ventricle (LV) in real-time 3D echocardiography (RT3DE) often rely on manual initialization. So far, little effort has been put in 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. It assigns probabilities to candidate LV center points through a Hough transform for circles, and detects the LV LAX by combining dynamic programming detections on these probabilities in 3D 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 (11 patients) and with the Philips Sonos 7500 ultrasound system with the X4 matrix transducer (14 patients). For the FRU data, the LAX was estimated with a distance of 2.85 plusmn 1.70 mm (Av plusmn SD) and an angle of 5.25 plusmn 3.17 degrees; the mitral valve plane was estimated with a distance of -1.54 plusmn 4.31 mm. For the matrix data these distances were 1.96 plusmn 1.30 mm with an angle error of 5.95 plusmn 2.11 and -1.66 plusmn 5.27 mm for the mitral valve plane. These results confirm that the method is very well suitable for automatic detection of the LV LAX and MVP. It provides a basis for further automatic exploration of the LV and could therefore replace manual initialization of 3D segmentation approaches.

PS-8 Sparse Appearance Model Based Registration and Segmentation of 3D Echocardiographic Images

In this study, active appearance model segmentation and intensity-based image registration were combined for detecting contours in three-dimensional (3D) echocardiograms fully automatically. An appearance model was built in 3D sparsely, consisting of the anatomical 4-chamber, 2-chamber, and short-axis views, which were extracted from end-diastolic 3D data sets. The model was used to segment an unseen image in a registration framework, by optimizing appearance and pose parameters simultaneously. Encouraging results were obtained with leave-one-out experiments on 16 patient data sets. The method may help inter- and intrapatient comparison of images, and is intended as an initialization for a complete 3D segmentation

Automatic segmentation of puborectalis muscle on three-dimensional transperineal ultrasound: Automatic puborectalis segmentation

The introduction of three‐dimensional (3D) analysis of the puborectalis muscle (PRM) for diagnostic purposes into daily practice is hindered by the need for appropriate training of observers. Automatic segmentation of the PRM on 3D transperineal ultrasound may aid its integration into clinical practice. The aims of this study were to present and assess a protocol for manual 3D segmentation of the PRM on 3D transperineal ultrasound, and to use this for training of automatic 3D segmentation method of the PRM.

Automatic segmentation of puborectalis muscle on three-dimensional transperineal ultrasound

The introduction of three‐dimensional (3D) analysis of the puborectalis muscle (PRM) for diagnostic purposes into daily practice is hindered by the need for appropriate training of observers. Automatic segmentation of the PRM on 3D transperineal ultrasound may aid its integration into clinical practice. The aims of this study were to present and assess a protocol for manual 3D segmentation of the PRM on 3D transperineal ultrasound, and to use this for training of automatic 3D segmentation method of the PRM.