|M.Sc Student||Goldman Avi|
|Subject||On Model-Based Ultrasound Imaging|
|Department||Department of Electrical and Computer Engineering||Supervisors||ASSOCIATE PROF. Moshe Porat|
|DR. Zvi Friedman|
|Full Thesis text|
Ultrasound images are often contaminated with interference noise and acoustic clutter, which obscures image details of interest, thus leading to potentially inaccurate clinical diagnosis. In order to address this problem, a model-based image reconstruction approach is proposed using the individually stored channel data of the ultrasound transducer elements, at all image grid points.
This approach is challenged by the enormous amounts of data involved. The data is acquired at the front-end of the system and transferred to the back-end for processing.
In order to address this challenge, we developed a method for simultaneously compressing the individual element RF data and reducing the clutter noise. This study performed data compression based on JPEG technique. Analysis of the data allows the description of the image as consisting of coherent strong reflectors, speckled tissue, and clutter noise, which can be mostly rejected according to its inter-element and inter-transmit second-order statistics. Accordingly, it is shown that this de-noised reconstruction becomes a spectral estimation task.
A model for the transthoracic cardiac-image individual element pixel data has been conceived following a detailed analysis of the data. We have observed that the data can be relatively accurately decomposed into several main components: a strong 'coherent' component originating from strong specular reflectors, signals from weaker reflectors and noise comprising of 3 main components: interference noise from strong off-axis components, clutter and thermal noise. A coherent strong reflector (either on or off-axis) is detected and accurately located from the properly time-delayed base-band complex element data. We have shown that in this case the complex element data follows a first order auto-regressive model. The phase of the auto-regressive coefficient is proportional to the angular off-axis position of the reflector. We have further shown that for relatively narrow transmit focusing, the single element complex base-band data follows the far-field single frequency DOA (direction of arrival) classical model. The image could thus be formed using methods 'borrowed' from Radar and radio-astronomy. The simplest method follows the well-known 'CLEAN' algorithm. The amplitude of the coherent strong reflector is determined by adjusting the delays so that the receive beam is directed toward the strong reflector, computing and subtracting its contribution to the individual elements signals and then using these signals to compute the residual reflectivity function. A similar concept is also employed in radar systems in order to reduce perturbations from off-axis jammers.
As in radar systems, we made use of the overlap between signals reflected in consecutive transmit events. We have used this coherence for improving the estimation of the directions of arrival, effectively narrowing the transmit focal zone as well as for noise reduction.
Our conclusion is that the proposed model-based approach could be applied to echocardiographic scanners yielding high quality de-noised images with limited computational and power resources.