learning can then be used to fine-tune the CD solution . 1 ON CONTRASTIVE DIVERGENCE LEARNING

During the course of this investigation, many observations were made about the estimation of the in vivo power spectrum as it applies to estimating the characteristic size of the tissue microstructure. In this final chapter, the conclusions derived from all of the observations are summarized. Then, future goals for extending the work and applying the conclusions are discussed. 9.1 Conclusions from Current Investigation For the estimation of the characteristic size of the tissue microstructure, the in vivo power spectrum can be determined to sufficient accuracy by estimating the total attenuation along the propagation path over the frequency range used to estimate the scatterer size. Hence, most of the work in this investigation attempted to solve for the scatterer size and total attenuation simultaneously. However, before solving for these two parameters, the effects of focusing on the estimation of the scatterer size were considered while assuming that the attenuation was known so that the results of the investigation could later be applied to clinically relevant fields. By assuming that the velocity potential fields near the focus could be modeled as a three-dimensional Gaussian distribution, focusing could be corrected by using a new generalized attenuation-compensation function that accounted for focusing, windowing, and attenuation. The new generalized attenuation-compensation function was then compared to the traditional attenuation-compensation functions that neglected focusing along the beam axis using computer simulations and phantom experiments. The generalized attenuation-compensation function yielded improvements in size estimation accuracy as high as 100% over the traditional attenuation-compensation functions. However, errors from the 155