13:30   Image Analysis 1: Ultrasound Applications
Chair: Christiaan van Swol
15 mins
Paulo Rodrigues, Anna Vilanova, Thorsten Twellmann, Bart ter Haar Romeny
Abstract: In segmentation techniques for Diffusion Tensor Imaging (DTI) data, the similarity of diffusion tensors must be assessed for partitioning data into regions which are homogeneous in terms of tensor characteristics. Various distance measures have been proposed in literature for analysing the similarity of diffusion tensors (DTs), but selecting a measure suitable for the task at hand is difficult and often done by trial-and-error. We propose a novel approach to semiautomatically define the similarity measure or combination of measures that better suit the data. We use a linear combination of known distance measures, jointly capturing multiple aspects of tensor characteristics, for comparing DTs with the purpose of image segmentation. The parameters of our adaptive distance measure are tuned for each individual segmentation task on the basis of user-selected ROIs using the concept of Kernel Target Alignment. Experimental results support the validity of the proposed method.
15 mins
Martin Verweij, Koos Huijssen, Koen van Dongen, Nico de Jong
Abstract: In the recent decade, the Tissue Harmonic Imaging (THI) echography modality has gained rapid popularity because of its superior image quality as compared to traditional Fundamental Imaging (FI). THI benefits from the second harmonic field component that is caused by the nonlinear acoustic behavior of biomedical tissue. A further improvement of image quality is expected from SuperHarmonic Imaging (SHI), which is an echography modality that is currently under development and that employs the third, fourth and fifth harmonic field components. To accurately predict the higher harmonics and to optimize the design of novel SHI ultrasound transducers, a new simulation method for nonlinear acoustic wave fields is needed. This method should be capable of simulating the pulsed acoustic wave field from a phased array transducer in a three-dimensional, large-scale configuration that shows nonlinearity and frequency power law attenuation. The current paper focuses on the newly developed Iterative Nonlinear Contrast Source (INCS) method, which is specifically designed for this purpose. The INCS method generates a solution of the nonlinear acoustic wave equation by treating the nonlinear term as a contrast source. Initially, the linear field solution is used to obtain an estimate of this contrast source. Then, a first approximation to the nonlinear wave field is obtained by solving the linear wave problem for this contrast source. Iteration of this procedure leads to a scheme that converges to the full nonlinear wave field. The linear substeps in the iterative scheme are performed by a convolution of the background Green’s function with the relevant sources. Spatiotemporal filtering of the convolution factors enables the use of a coarse computational grid with only two points per wavelength. The numerical evaluation of the convolution sum is performed efficiently with the aid of FFT’s. The INCS method is implemented in Fortran on a single-processor machine as well as on a multi-processor computer system. Large-scale, nonlinear field profiles have been computed for transducers with cylindrical, rectangular and phased array geometries, excited with a pulsed waveform having a center frequency of 1-2 MHz. Comparison with results obtained from models of reduced complexity and with measurements shows that the INCS method accurately predicts the nonlinear wave field in all situations that were considered.
15 mins
Gert Weijers, Johan Thijssen, Alexander Starke, Alois Haudum, Kathrin Herzog, Jurgen Rehage, Chris de Korte
Abstract: In this study, a Computer-Aided UltraSound (CAUS) diagnostic method has been developed for the detection and staging of hepatic steatosis (non-alcoholic fatty liver disease, NAFLD). Hepatic steatosis induces a risk of various kinds of co-morbidity. Assessment of liver fat content is generally done by taking a biopsy. The authors’ goal was to estimate the liver fat content non-invasively. CAUS was developed for calibration and objective estimation of ultrasound parameters from conventional B-mode images and was validated in a bovine model. Transcutaneous and intra-operative echographic B-mode images were obtained postpartum from dairy cows (n=151) using a convex array transducer (fc=4.2 MHz). During surgery, a biopsy was taken from the right lobe to assess the triglyceride (TG) content in the liver and to perform a histopathological examination. First, the gray levels of the images were transformed to dB relative to the background echo level in a tissue mimicking phantom. Then the sector-shaped images were transformed to a rectangular format (back-scan conversion). A control group of healthy cows (n=12) was used from which Automatic Gain Correction curves (AGC) were derived and applied to all images. Furthermore, combined skin and subcutaneous layer attenuation was estimated and corrected for. Estimated echo characteristics were: mean echo level, residual attenuation coefficient and mean axial and lateral speckle size. These parameters were significantly correlated to the TG. Multivariate analysis yielded a linear regression formula. This formula was used to predict the TG content for each animal. Finally ROC curves were estimated for different thresholds of TG content. High correlations with fat content were obtained for the residual attenuation (r=0.80), mean echo level (r=0.59) and fat layer thickness (r=0.46). Step-wise multiple linear regression revealed a high correlation (r=0.83) of the estimated to the true TG content. Residual attenuation and axial and lateral speckle size were incorporated in the regression formula. ROC curve analysis show promising results for sensitivity (0.93), specificity (0.86) and area under the curve (0.93) in distinguishing fatty livers from healthy livers. This study showed the feasibility of generic computer-aided ultrasound for non-invasively diagnosing, maybe even screening, of liver steatosis.
15 mins
Jithin Jose
Abstract: Photoacoustic imaging is a recently developed imaging modality, which can be used for small animal imaging. It relies on irradiating the tissue surface with nanosecond pulses of visible, or deeply penetrating near-infrared (NIR) laser light. Optical absorption in the tissue causes a thermoelastic expansion, which produces broadband pulses (MHz) of acoustic energy. These pulses propagate to the tissue surface and are detected by an array of ultrasound transducers. With the knowledge of the speed of sound in the tissue, the acoustic signal can be backprojected in 3D to reconstruct a volumetric image of the internally distributed photoacoustic source. Photoacoustic imaging does not suffer from the limited resolution faced by optical imaging, while still using light as the probing energy. This allows combining the investigation of optical absorption contrast with high resolution of ultrasound imaging. In this work we present a curvilinear array based computed tomography (CT) photoacoustic imaging system optimized for rapid, high resolution imaging of small animals. The system features a 32 element ultrasonic transducer array shaped to 85 degree of a circle of 40 mm radius. The system operates at a central frequency of 6.25 MHz with a fractional frequency bandwidth of 80%. A Q-switched Nd:YAG laser, equipped with an OPO, delivering 5 ns pulses at 800 nm is used to produce photoacoustic signals from the subject under investigation. The subject under investigation is held stationary in an imaging tank with water. The detector and laser beam are coupled to a rotary mechanism which allows a CT type measurement of the object to be performed. At all angles around the object, the laser illumination is arranged to make both forward- and backward-mode measurements possible. Measurement of a single slice takes less than 1 minute. Image reconstruction is performed using a modified acoustic backprojection algorithm. We present the system overview and imaging results on phantoms and sacrificed mice.
15 mins
Rene Willemink, Srirang Manohar, Kees Slump, Ferdi van der Heijden, Ton van Leeuwen
Abstract: In ultrasound transmission tomography, images are calculated which represent distributions of ultrasound attenuation and speed of sound in objects. For the reconstruction of these images, measurements of projections of ultrasound propagation parameters are required. We have developed several new methods to obtain accurate measurements of these projections. There are existing algorithms, which can estimate ultrasound attenuation. However, these algorithms have limitations in the sense of an unnecessarily low signal to noise ratio or strict assumptions on the source signal and kind of object to use. In general the ultrasound propagation parameters can be divided into parameters representing the attenuation and parameters representing the time delay. For most materials, the attenuation can be described with a function which is related to the frequency by a power law. In the case of soft tissue, this function is almost a linear function of frequency. Directly related to such an attenuation function is a phase function, representing the speed of sound of the material. The attenuation and phase functions are related to eachother by the Kramers Kronig relations. We have developed ultrasound parameter estimators that take this relation into account so that a better signal to noise ratio can be obtained. The developed estimators also take into account the noise distribution of the input measurements so that maximum likelihood esimators are obtained. The developed estimators are applicable to ultrasound transmission mode measurements, meaning that a sending ultrasound source should be placed on one side of the object and a receiving ultrasound transducer at the other side. The inputs to the estimators are obtained by measuring the source with the object in place and with the object removed. The source signal is ideally a broadband pulse so that a wide range of the spectrum can be measured. We compare our newly developed estimators with two existing algorithms, the frequency shift method and the log spectral difference method. The performance is measured in terms of bias and standard deviation of the estimators. To evaluate this performance, we have used both computer simulations and ultrasound phantoms. The results show that our estimators perform better than both existing methods.