|M.Sc Student||Kerner Yair|
|Subject||Ultrasound Image Enhancement Based on Image Compounding|
|Department||Department of Electrical Engineering||Supervisor||Professor Moshe Porat|
We developed a new method to exploit pairs of ultrasound scans of the same image plane with the aim of enhancing the quality of ultrasound imaging. Each pair is assumed to consist of two co-registered images with 90° separation between the two insonification directions. The motivation for such spatial compounding is the resolution in medical ultrasound imaging, which is significantly worse in the lateral direction compared to the axial direction.
The proposed method seeks a Maximum-Likelihood solution for the identification of the system response and the noise variance through the Expectation-Maximization (EM) technique, similar to the approach of multi-channel image restoration. The result is an iterative algorithm, where the recent parameter estimates are used to calculate conditional expectation and variance of the tissue reflectivity and these are used in turn to update the parameter estimates.
For the following step of image reconstruction and compounding we take into account the non-linear operations that are required for display. We show that the best ability to separate close small objects is achieved when the compound image is produced through first using separate Wiener filtering for each RF image, and afterwards performing envelope-detection and averaging. On the other hand, applying envelope-detection to the estimated tissue reflectivity from the EM algorithm produces the greatest suppression of noise and speckle.
Our experiments show that the parameter
identification is robust to changes in the relative level of speckle or white noise
as well as in the spectral system response. Our conclusion is that the proposed
approach to image restoration can enhance the quality of the compound image
compared to presently available methods.