Representation of shape in ultrasonic images with a physically-based image model

In contrast to the emphasis on shape structures that is characteristic of most image analysis research, we have emphasized the imaging physics in developing a model for representing ultrasonic images in terms of the underlying tissue shape. Ultrasound has been a major focus in image-guided surgery research for over a decade, but limited accuracy and ease-of-use issues remain prohibitive given current methodology. Toward a robust method for understanding as well as representing ultrasonic images, we have developed an image model framework that is based on a physical description of image formation. Our framework incorporates the gross surface shape, its scattering microstructure and the 3D point-spread function for the imaging system. The framework permits simulation of images, characterization of tissue via statistical images, and construction of a data likelihood for statistical inference of underlying model components. We have tested the model with cadaveric vertebrae by direct, registered comparison of simulated and actual images and by using the data likelihood to infer vertebral pose for registration. Because of its physical basis, the model could be adapted to account for other effects relevant to image-guided surgery such as attenuation- and speed-of-sound-based image distortion. As a framework, the physical description and image model are fully extendable to other applications, including adaptive image formation incorporating tissue shape