• Home
• Preliminary Program
• News
• Speaker/Poster Information
• Conference Topics
• Programme Information
• Research Market / Beer & Wine party
• Awards
• Company Market
• Invited Speakers
• Important Dates
• Registration and Venue
• Authors Instructions
• Organizing Board
• Contact

10:40   Imaging - General Chair: Jan de Munck 10:40 15 mins PARALLELIZING 2D ULTRASOUND STRAIN IMAGING Tim Idzenga, Evghenii Gaburov, Jan Menssen, Chris de Korte Abstract: Deformation of tissue can be accurately estimated from radiofrequency ultrasound data using 2-dimensional Normalized Cross Correlation (NCC) [1]. These deformations can be used to identify pathological tissues, e.g. tumors in breast tissue or vulnerable plaques in the arteries. Strain estimation, however, is very computation time-consuming and could therefore benefit tremendously from a speedup. Parallelizing the NCC computations could lead to a major time-reduction. For this we have investigated the use of a Graphics Processing Unit (GPU) and compared the results with 2 CPU-based approaches. The latter two were the MATLAB build-in function (running on a single core) and an optimized C-based function (running on a single core and an 8-core node using OpenMP). The GPU-implementation was written using Compute Unified Device Architecture (CUDA). The performance of the three implementations was expressed in the number of floating-point operations per second (FLOPS) that could be achieved. On an in vivo ultrasound data set of the common carotid artery during one cardiac cycle we compared the performance and computation time for the three implementations. All computations were executed on the LISA GPU-cluster (www.sara.nl) which consisted of nodes equipped with 2 quad CPU processors: Intel(R) Xeon(R) CPU L5420 (20 GFLOPS/core single precision). Each node was equipped with a Tesla M1060 (933 GFLOPS, Nvidia, Santa Clara, CA). On NCC-computation the MATLAB-function achieved a maximum performance of 2.3 GFLOPS (11.5% of peak performance), whereas the C-implementation achieved a maximum performance of 5.4 GFLOPS on the single core (27% of peak performance) and 40.5 GFLOPS (25% of peak performance) on the 8-core node. The GPU-implementation achieved the highest performance of 225 GFLOPS (24% of peak performance). The mean execution time of the strain estimation algorithm on the in vivo data using the MATLAB NCC-function was 144 mins. Using the C-function on a single core it was 25 mins and 8 mins on the 8-core. Using the CUDA-function the mean execution time was 5 mins. Neither of the implementations resulted in a significant loss in elastogram image quality. Parallelization of the NCC calculations in strain estimation using a GPU significantly reduces the computation time, CUDA achieves the largest performance. The gain in computation time is less than expected from the performance. This is probably due to overhead code that is executed in MATLAB. Nevertheless, the time required for estimating deformation in tissue is reduced from hours to minutes. This opens the door to real-time ultrasound strain estimation. REFERENCES [1] R.G.P. Lopata, M.M. Nillesen, H.H.G. Hansen et al., “Performance evaluation of methods for two-dimensional displacement and strain estimation using ultrasound radio frequency data,” Ultrasound Med.Biol., vol. 35, no. 5, pp. 796-812, 2009. 10:55 15 mins OPTIMAL SELECTION OF IMAGING WAVELENGTHS FOR SPECTRAL UNMIXING Martijn van de Giessen, Boudewijn Lelieveldt, Jouke Dijkstra Abstract: Real-time multispectral intra-operative imaging of fluorescent probes imposes constraints on the imaging hardware that are not present in pre-clinical imaging. First, all spectral bands must be imaged simultaneously. This constraint, together with the small form factor of the camera limits the number of imaged wavelengths. In this work a method is proposed to determine the optimal measurement wavelengths based on measured spectra. Assuming a linear mixing model, the measured light intensity in a single spectral band to a mix of probes can be described by a row in a mixing matrix. Similar rows (i.e. measurements) add little information about the ratio between imaged probes, contrary to very different rows. The condition number of this matrix is small for spectra that are linearly independent (the desired situation) [1]. Therefore, we propose to select wavelengths such that the condition number is minimal. Unmixing results using 'optimal' wavelengths are compared to unmixing using 'peak' wavelengths of the probes for simulated multispectral images with increasing noise levels. Probe spectra are measured with the Maestro system between 630nm and 850nm (interval 20nm). The experiment is repeated for all combinations of 2 to 5 probes: AF680, QD700, AF750, QD800 and skin auto-fluorescence. For in-vitro multi-spectral luminescence data from the IVIS Spectrum system, the unmixing results using optimized and peak wavelengths are compared to unmixing based on all acquired wavelengths (500nm to 700nm, interval 20nm) in in-vitro experiments with wells of two mixed probes: CBG99 (540nm) and PpYRE8 (620nm). In simulations the 'peak' unmixing errors were on average 2.86 times larger than the 'optimal' unmixing errors. Selecting the optimal wavelengths was especially beneficial for 3 or more probes. The IVIS in-vitro measurements showed that unmixing with the 'optimal' wavelengths was 8.9% more precise than using the 'peak' wavelengths. Qualitative assessment showed clearer visibility of small structures, e.g. vasculature in mouse brains, when using the ‘optimal’ wavelengths. The results clearly showed that selecting the optimal wavelengths as proposed gives almost 3 times more accurate estimates than measuring at the spectral peaks of the imaged probes, both in simulations as in real measurements. This research was supported by the Center for Translational Molecular Medicine (MUSiS). [1] Silvan-Cardenas and Wang, Fully Constrained Linear Spectral Unmixing: Analytic Solution Using Fuzzy Sets, IEEE Trans. Geoscience and Remote Sensing, 2010 11:10 15 mins MEMBRANE DESIGN OF AN ALL-OPTICAL ULTRASOUND RECEIVER IN SILICON ON INSULTAOR TECHNOLOGY Suzanne Leinders, Wouter Westerveld, Jose Pozo, Mirvais Yousefi, Paul Urbach, Koen van Dongen Abstract: Ultrasonography is a medical diagnostic tool utilized for the imaging of subcutaneous body structures, i.e. tendons, muscles, joints, vessels and internal organs for possible pathology or lesions. Lately, it has been recommended as an effective diagnostic tool for the diagnostic of atherosclerosis. By bringing the ultrasonic transducer, mounted on the tip of a catheter, into the artery an image of the vessel wall could be obtained. However, respiratory motion can displace the catheter tip as much as 6 mm, resulting in serious deterioration of images. To improve the image quality, it is advantageous to use an array of many transducers in the arterial direction [1,2]. Unfortunately, traditional piezo-electric receivers are relatively large (500 µm) and must be wired individually, which highly limits the amount of transducers that could be used to build an array. Therefore, alternative transducers or methods are needed. We propose a novel type of ultrasound receiver array based on integrated photonic resonators, in which photons rather than electrons, carry information. This technology allows for small footprint (~50 µm per sensor) and hence, for maximizing the amount of sensors in the catheter. One sensor consists of an optical resonator on a silicon substrate with an acoustical membrane that can be efficiently excited by an ultrasonic wave field. The deformation of the membrane caused by ultrasonic waves shifts the optical resonance[3], which can be monitored by an external interrogator system. Finite element modeling (COMSOL Multiphysics) has been performed to optimize the acoustical sensitivity of the receiver. Insight about the influence on the acoustical behavior of the size, thickness and shape of the membrane were obtained using both static and time domain analysis. The model predicts that for a 2.5 µm thick SiO2 circular membrane with 30 µm radius, 0.9 MHz resonance occurs when the receiver is submerged in water. The promising results shown by the simulations, lead to the conclusion that this technology is a suitable candidate for miniaturized ultrasound sensors. REFERENCES [1] R.S.C. Cobbold, Foundations of biomedical ultrasound, 2007. [2] E.J. Alles, et al. “An axial array for three-dimensional intravasular ultrasound”, IEEE International Ultrasonics Symposium, Dresden Germany (2012) [3] Wouter J. Westerveld, et al. "Characterization of a photonic strain sensor in silicon-on-insulator technology," Opt. Lett. 37, 479-481 (2012) 11:25 15 mins FAST EIGENVECTOR CENTRALITY MAPPING OF VOXEL-WISE CONNECTIVITY IN FUNCTIONAL MRI: IMPLEMENTATION, VALIDATION AND INTERPRETATION Alle Meije Wink, Jan de Munck, Frederik Barkhof Abstract: Eigenvector centrality mapping (ECM) has recently emerged as a measure to spatially characterise connectivity in functional brain imaging by attributing network properties to voxels [1]. The main obstacle for widespread use of ECM in functional MRI (fMRI) is the cost of computing and storing the connectivity matrix. We present fast ECM (fECM), an efficient algorithm to estimate voxel-wise eigenvector centralities from fMRI time series [2]. Instead of explicitly storing the connectivity matrix, fECM computes matrix-vector products directly from the data, achieving high accelerations for computing voxel-wise centralities in fMRI at standard resolutions for multivariate analyses, and enabling high-resolution analyses performed on standard hardware. We demonstrate the validity of fECM at cluster and voxel levels, using synthetic and in vivo data. Results from synthetic data are compared to the theoretical 'gold standard', and local centrality changes in fMRI data are measured after experimental intervention. A simple scheme is presented to generate time series with prescribed covariances that represent a connectivity matrix. These time series are used to construct a 4D data set, whose volumes consist of separate regions with known intra- and interregional connectivities. The fECM method is tested and validated on these synthetic data. We demonstrate fECM on real fMRI data, first in healthy subjects scanned after real vs. sham repetitive transcranial magnetic stimulation (rTMS), an also in patients with multiple sclerosis (MS) patients compared to controls and in patients with Alzheimer's Disease (AD) compared to controls. Its computational speed and statistical sensitivity make fECM a good candidate for connectivity analyses of multi-modality and high-resolution functional neuroimaging data. 11:40 15 mins BIOMECHANICAL CHARACTERISTICS OF AORTIC ANEURYSMS GENERATED IN FIBULIN-4 DEFICIENT MICE Lambert Speelman, Natasja Ramnath, Kim van der Heiden, Ton van der Steen, Jeroen Essers, Jolanda Wentzel, Frank Gijsen Abstract: Aortic aneurysms affect approximately 5% of the elderly population and are responsible for a significant number of deaths in the western world, due to aneurysm rupture. Fibulin-4 is a secreted glycoprotein, which is expressed in medial layers of blood vessels and a critical component for the structural integrity and elasticity of the aortic wall [1]. Mice with reduced levels of Fibulin-4 develop aortic abnormalities similar to Fibulin-4 patients, such as dilation of the ascending aorta. A 4-fold reduction of Fibulin-4 expression (fib-4R/R) causes a severe dilation, while a mice with a 2-fold reduction (fib-4+/R) show an onset of aneurysm formation, comparable with the development of an aneurysm in aging humans [2]. In this study we evaluate the biomechanical characteristics of the arterial wall of fib-4+/+ (with normal Fibulin-4 expression), fib-4+/R, and fib-4R/R mice, using high-frequency ultrasound, two-photon scanning laser microscopy (TPSLM), and histology. Ultrasound B-mode imaging of the aortic arch was followed by M-mode imaging at standardized locations in the aorta for diameter measurements with a VisualSonics Vevo2100 system equipped with a linear array probe (MS-550D, frequency 22-55 MHz). After sacrificing the mice, carotid arteries were excised and placed on glass cannulas in a vessel perfusion system. The arteries were preconditioned between 80 and 120 mmHg and imaged with a TPSLM with a resolution of 0.2x0.2x0.2 μm. A 2D FFT analysis was performed to determine the predominant fiber angle and the alignment index of the collagen fibers [3]. The architecture of the ascending thoracic aorta was evaluated with microscopic histological analysis with a Verhoef-Van Giesson staining for the elastin architecture of the vessel. Compared to the fib-4+/+ and the fib-4+/R mice, the fib-4R/R mice showed a larger diameter and lower distensibility of the aortic arch. This loss of ‘elastic’ behavior can be attributed to fragmentation of elastin and loss of smooth muscle cells in the medial layer. The TPSLM data on collagen distribution in the adventitia confirm this: fibers are more aligned and stretched in the fib-4+/R and fib-4R/R mice, even at lower pressure. The loss of structural elastin and decreased distensibility is also found in the aortas of patients with aortic aneurysms. This mouse model therefore shows to be an excellent model to study the interplay between elastin degradation and collagen remodeling and might be a promising model to study aneurysm formation and progression. REFERENCES [1] Lederle FA, et al. Ann. Intern. Med. 126(6):441-9, 1997. [2] Hanada K, et al. Circ. Res. 100(5):738-46. 2007. [3] Timmens et al. Am. J. Physiol. Heart CircPhysiol. 2010. 11:55 15 mins VOLUME DENSITY OF INTRAMURAL CORONARY BRANCH SEGMENTS DIFFERS IN RELATION TO MYOCARDIAL DEPTH Pepijn van Horssen, Jeroen van den Wijngaard, Martin Brandt, Imo Hoefer, Jos Spaan, Maria Siebes Abstract: Background: An ordered description of the intramural coronary arterial network is essential for vascular tree generation using scaling laws. Knowledge of the intramural vasculature also plays an important role for modeling the transmural blood flow distribution. We investigated the detailed structural organization of coronary arteries and microvessels using an imaging cryomicrotome. Methods: High-resolution registered 3D image stacks of coronary vessels in 8 canine hearts were acquired at 40 µm slice thickness. After skeletonization of the 3D vascular network [1], all branch segments were labeled and connectivity with neighboring segments was determined. Starting from the epicardium, the sub-tree (crown) belonging to each penetrating artery (stem) was identified and the associated perfusion territory was demarcated by 3D Voronoi tessellation, with the endpoints of terminal segments as Voronoi cell centers. The transmural location of the centroid of the Voronoi cell cluster per stem served to classify the perfused crown territory as endocardial, mid-myocardial or epicardial. Vascular volume density was expressed as fraction of vascular volume, comprised of small arteries <400 µm diameter, per total tissue volume within a perfusion territory. Results: Endocardial territories were in the order of 0.57 ml and comprised about 50 per heart, while volumes of the epicardial territories had a median value of 0.029 ml and numbered about 600 per heart. The vascular volume density was 3.2% at the endocardium compared to 0.8% in the epicardium. Accordingly, different scaling law properties for vascular volume vs. tissue mass were derived for these territories. Conclusions: A four times higher volume density of small arteries in endocardial territories may compensate for the stronger flow impediment by extravascular forces due to cardiac contraction, but also entails a larger area at risk for ischemia in case of a proximal stenosis. The resulting different scaling law properties of vascular volume vs. tissue mass per territory type should be taken into account when patient-specific computational models are used to predict coronary perfusion. This work was funded by NHS grant 2006B226 and by FP7-ICT-2007-224495: euHeart. Reference [1] P. van Horssen, J.P.H.M. van den Wijngaard, F. Nolte, I.E. Hoefer, R. Haverslag, J.A.E. Spaan and M. Siebes (2009). Extraction of coronary vascular tree and myocardial perfusion data from stacks of cryomicrotome images. Functional Imaging and Modeling of the Heart: 486-494.