| The early diagnosis is of paramount importance in treating cancer, but it still remains a difficult problem. This thesis investigates quantitative tissue characterization and image deconvolution as steps towards improving ultrasound based early detection of liver tumors. The goal of tissue characterization is to develop objective tissue signatures that assume distinct forms for normal and diseased tissues; the goal of deconvolution is to compensate for distortions associated with the imaging process.;In this thesis, tissue is modeled as a collection of point scatterers embedded in a uniform media, and higher-order statistics (HOS) of the scatterer spacing distribution is estimated from RF-echoes and used in obtaining tissue signatures. Schemes for the estimation of resolvable periodicity, as captured in the signature Mean-scatterer-spacing (MSS) as well as higher-order correlation structure of scatterers, as captured in the novel signature color of the tissue response, are developed. Working on simulated data, it is shown that the proposed periodicity estimation technique outperforms the power cepstrum techniques in terms of the accuracy of estimations. Working on clinical data, it is shown that the MSS stays fairly constant over healthy livers, and deviate from the value for healthy livers, inside some tumors. The color of the tissue response estimated inside and outside tumors showed a consistent difference.;This thesis introduces HOS based methods for the in-vivo identification of mixed-phase imaging distortions associated with clinical ultrasound images. The mathematical formulation based on HOS of the RF-echo is described, and the distortion kernels are estimated. Axial distortion kernels estimated through the technique are experimentally verified. Image deconvolution was shown to consistently lead to resolution gains as indicated by image auto-covariance functions and by the size of speckle spots.;Both tissue characterization and image deconvolution approaches depend on the theory of non-minimum phase blind system identification from the output. In this thesis, novel techniques for system reconstruction from higher-order output cumulants are proposed. The performance of the proposed techniques are similar to those of existing higher-order cepstra based ones, as revealed by Monte Carlo experiments. In contrast, the computational complexity of the proposed techniques is dramatically lower and similar to those of second-order correlation based approaches, independent of the order of the cumulant estimates. |
