10:30
Cardiovascular Mechanics III
10:30
15 mins
|
TOWARDS 3D CAROTID STRAIN IMAGING: A SIMULATION STUDY
Stein Fekkes, Abigail Swillens, Hendrik Hansen, Anne Saris, Maartje Nillesen, Francesco Iannaccone, Patrick Segers, Chris de Korte
Abstract: Atherosclerosis is the primary cause of heart disease and stroke in westernized societies. A substantial part of all strokes is caused by the rupture of a plaque in the carotid artery (CA) which triggers an atherothrombotic reaction leading to stroke or transient ischemic attack. Plaque rupture usually happens at high strain locations. Therefore, accurate assessment of the strain distribution in plaque regions is of high importance. Ultrasound has proven to be a promising imaging modality to assess plaque vulnerability using strain imaging. Non-invasive vascular ultrasound strain imaging has shown to be feasible in the longitudinal and transverse directions to obtain radial, longitudinal and circumferential strains. Since these methods are primarily based on 2D data, the most vulnerable spot of a plaque could be easily missed due to a limited field of view. Insight in the 3D plaque deformation, geometry and location in the CA is of paramount importance for a complete and accurate assessment of the severity of atherosclerosis. In this simulation study we propose a method that uses ultra-fast plane wave imaging to acquire semi 3D radial strain images of the complete plaque region.
A 3D finite element model of a patient-specific, pulsating atherosclerotic carotid artery containing fatty and calcified areas was generated with Abaqus. Plane wave ultrasound data were simulated for a linear array transducer using FIELD II. Cross-sectional radial strain were estimated using the compounding technique developed in our group [1] to provide accurate estimates of the strain tensor. This technique combines displacement estimates obtained by cross-correlation of raw radiofrequency (RF) ultrasound data acquired at three insonification angles at a high pulse repetition frequency of 12 kHz. Finally, the 3D strain volume was composed by stacking the radial strain images for 25 equally spaced (0.5 mm) elevational positions along the carotid artery.
Comparing estimated radial strain values with true strain values originating from the model, demonstrates the feasibility of 3D strain assessment using this multi-slice approach. It also shows correspondence of high and low strain values with lipid rich and calcified regions respectively. Applying this approach to controlled in vitro phantom experiments will be the next validation step. The ongoing development of full 3D plane wave imaging using matrix arrays will allow high frame rate full volume acquisition, which holds the potential to enhance 3D strain assessment of the carotid artery even further.
REFERENCES
[1] H. H. Hansen, A. E. Saris, N. R. Vaka, M. M. Nillesen, and C. L. de Korte, "Ultrafast vascular strain compounding using plane wave transmission," J Biomech, vol. 47, pp. 815-23, Mar 3 2014.
|
10:45
15 mins
|
2D BLOOD VELOCITY IMAGING USING PLANE WAVES
Anne Saris, Rik Hansen, Stein Fekkes, Maartje Nillesen, Marcel Rutten, Chris de Korte
Abstract: Complex flow phenomena are present in our cardiovascular system, with strong spatial and temporal fluctuations in both the velocity magnitude and direction. Conventional Doppler-based blood flow imaging techniques face difficulties capturing the true blood velocity vector, since only the velocity component along the ultrasound beam is estimated, which is projected at an angle defined by the sonographer. A way to circumvent this angular dependence is to estimate the velocity vector by using 2D (speckle) tracking, based on the cross-correlation of small regions of radiofrequency (RF) or envelope data. To capture the rapidly changing dynamics of blood flow in 2-D, a large imaging region needs to be insonified at a high frame rate. This is possible by transmitting plane or diverging waves instead of conventionally used focussed ultrasound pulses. However, due to the lack of focussing in transmit, the image quality is reduced, especially in the off-axis direction, which will hamper the estimation of blood velocities. A way to circumvent the reduced off-axis quality is by using displacement compounding [1], which uses only axial displacement estimates obtained at angled plane waves to derive the 2D displacement and velocity vector. This study compares the performance of 0° and compound 2D vector velocity estimation, using plane wave transmissions.
Fluid-structure interaction modelling was used to generate realistic 3D flow velocity fields inside a curved carotid artery over the cardiac cycle [2]. Using the velocity fields as input for the ultrasound simulation program Field II, ultrasound data were simulated for a linear array transducer transmitting plane waves at angles of 0°, -20° and 20°. 2D normalized cross-correlation was used to estimate the displacements, rendering the velocity by multiplication with the frame rate. The methods were compared using the root mean squared error (RMSE) between estimated and true velocity components, as derived from the model. Results show the feasibility of plane wave blood velocity imaging. For low blood velocities, the compounding method provides most accurate results. In phases where blood velocities increase, 2D cross-correlation using only 0° PW acquisitions provides more robust velocity estimates.
REFERENCES
[1] H. H. G. Hansen, R. G. P. Lopata, T. Idzenga, and C. L. de Korte, "Full 2D displacement vector and strain tensor estimation for superficial tissue using beam-steered ultrasound imaging," Phys Med Biol, vol. 55, pp. 3201-3218, Jun 7 2010.
[2] B. W. A. M. M. Beulen, M. C. M. Rutten, and F. N. van de Vosse, "A time-periodic approach for fluid-structure interaction in distensible vessels," Journal of Fluids and Structures, vol. 25, pp. 954-966, Jul 2009.
|
11:00
15 mins
|
CARDIAC DEFORMATION ESTIMATION USING HIGH FRAME RATE ULTRASOUND IMAGING
Maartje Nillesen, Anne Saris, Hendrik Hansen, Frebus van Slochteren, Peter Bovendeerd, Chris de Korte
Abstract: Echocardiography is a widely used imaging modality for monitoring the heart function. During the past decades, strain imaging techniques have been developed to estimate the local deformation of the cardiac muscle. Conventional focused ultrasound imaging techniques use line-by-line image acquisition and therefore have limited temporal resolution. In particular in cardiac phases with high deformation rates, ultrafast imaging techniques can be of great advantage for accurate assessment of 3D cardiac deformation. Because of the large imaging depth and large field of view required in cardiac applications, spherical (diverging) waves are designated for ultrafast transthoracic imaging.
Realistic 3D deformation fields of the cardiac muscle were generated using a 3D finite element model describing the mechanics of a healthy left ventricle [1]. These displacement fields served as an input for simulating ultrasound data in Field II for a 2D apical view over the entire cardiac cycle. Five spherical waves were emitted from equally spaced positions behind a phased array transducer at a pulse repetition frequency of 1000 Hz. To improve image quality, these 5 ultrasound transmissions were then combined to a single radiofrequency dataset (frame rate 200 Hz) using so-called coherent compounding. Axial displacement estimations were obtained using 2D cross-correlation techniques [2].
To evaluate the performance of the technique, axial displacement estimates were compared to displacements directly derived from scatterer positions in the myocardial tissue for six characteristic phases of the cardiac cycle: begin-systole (BS): rapid ejection phase, mid-systole (MS): reduced ejection phase, end-systole (ES), begin-diastole (BD): rapid filling phase, mid-diastole (MD): slow filling phase and end-diastole (ED). Good agreement was obtained between the axial displacement estimates and ground truth displacements in all 6 cardiac phases. Root-mean-squared errors between model-based and estimated axial displacements varied from 0.2 mm in rapidly deforming phases (BS and BD) to 0.001 mm in slowly deforming phases. These results indicate that 2D high frame rate cardiac displacement estimation using multiple spherical waves is feasible. The developed technique serves as an important step for high frame rate 3D cardiac deformation imaging.
REFERENCES
[1] P.H.M. Bovendeerd, W. Kroon, T. Delhaas, “Determinants of left ventricular shear strain”, Am J Physiol Heart Circ Physiol, Vol. 297, pp. 1058-1068 (2009).
[2] R.G.P. Lopata, M.M. Nillesen, H.H.G. Hansen, I.H. Gerrits, J.M. Thijssen and C.L. de Korte. "Performance of two dimensional displacement and strain estimation techniques using a phased array transducer", Ultrasound Med Biol., Vol. 35, pp. 2031-2041, (2009).
|
11:15
15 mins
|
ESTIMATION OF LEFT-VENTRICULAR FUNCTION IN LVAD-SUPPORTED HEARTS
Kim Pennings, Frans van de Vosse, Bas de Mol, Marcel Rutten
Abstract: see word document, since this textblock does not allow the right format (subscription, supscription).
|
11:30
15 mins
|
NON-CONTACT MONITORING OF VITAL SIGNS INDEPENDENT OF AMBIENT LIGHT AND SKIN PIGMENTATION
John Klaessens, Rudolf Verdaasdonk, Albert van der Veen, Frank van den Dungen
Abstract: Non-contact monitoring of vital signs of patients with video cameras is a promising technique in different clinical applications. The heart rate can be derived from the small variation in skin tone during the heart cycle using ordinary cameras and even smart phones apps. The respiration rate can be derived from minuscule motion. However this will work only in light environments and for persons with less pigmentation. For reliable clinical use, non-contact monitoring (photoplethysmography) was improved for both day and night use and for all skin types using near infrared (NIR) illumination and thermography in view of monitoring neonates.
The video stream from a NIR camera was analysed by defining a region of interest on the skin, calculating the mean pixel value and performing a Fourier transformation using a band-pass filter revealing the main frequency component representing the heart rate. A clinical pulse oximeter was used as reference. From temperature variations near the nostrils imaged with a thermo camera (Xenics Gobi 384) the respiration rate was derived.
In a population of 20 volunteers equally divided along the Fitzpatrick skin tone scale (1-6), the heart rate could be derived for all skin tones within 3 % accuracy compared to the pulse oximeter using NIR illumination. However, the HR could only be derived in 45 % of dark skin tones using white light illumination. In a preliminary clinical study with 10 neonates the HR was accurate within 3%.
Using NIR light, non-contact monitoring of heart and respiration rate is possible, independent of light conditions and skin tone and has potential for many applications e.g. baby monitoring, patients in MRI. Other vital parameters like oxygenation can be added. The technique has to be developed into a real time application and validated in clinical studies.
|
11:45
15 mins
|
HEART BEAT DETECTION IN MULTIMODAL DATA USING AUTOMATIC SIGNAL TYPE RECOGNITION
Thomas De Cooman, Griet Goovaerts, Carolina Varon, Devy Widjaja, Tim Willemen, Sabine Van Huffel
Abstract: Introduction. Heart beat locations can be found by multiple types of signals. Typically the electrocardiogram (ECG) is used for this, but also blood pressure and stroke volume signals can for example help to detect them. This makes it possible to use all these signals in order to find the heart beat locations in multimodal datasets. Labeling of the different signals is however required then, which is not always done correctly or sometimes not done at all. Therefore a method is described here that is able to detect these signals of interest automatically, so no signal labeling is required.
Data. The algorithm was submitted as an entry for the 2014 Computing in Cardiology (CinC) challenge. For this challenge, a multimodal training dataset consisting of 100 records of 10 minutes was available. The performance of the algorithm is tested by using a hidden test of 300 records.
Method. First, a noise-free training period is extracted from the ECG using an autocorrelation similarity matrix. Next, signals with the same periodic behavior as the ECG (like blood pressure and stroke volume) are found by analyzing the correlation coefficients between their power spectral density and that of the ECG. Once these signals of interest are detected, two different peak detection algorithms are used: a specific R peak detector for the known ECG signal and a general peak detection algorithm for the other signals. The beat labels coming from the different signals are combined in two phases. First, majority voting is applied in order to find a first list of heart beat locations. Afterwards, if too large RR intervals are found in this list, the beat labels from 1 signal in this RR interval are added to this list. This selection is done by analyzing the mean square error between the beat labels from 1 signal and the estimated heart beat locations.
Results. All useful periodic signals in the training set were detected with this approach. The algorithm got an overall performance of 99.95% on the available training set and 85.62% on the hidden test set. The third place in the 2014 CinC challenge was obtained with this approach.
Conclusion. We conclude that this method is able to detect the heart beat locations robustly in multimodal datasets with noisy or missing signal labeling.
|
|
