Ultrasound simulations using computed tomography images as priors

Abstract

Medical ultrasound system is one of the most ubiquitous imaging modalities in the world. In addition to the noninvasive nature of it, being low-cost and portable compared to other modalities makes it wide-spread in hospitals. Such common usage of the system needs more ultrasound technicians, or sonographers, and consequently training of new ones which requires at least a patient and an expert and almost a year to be a sonographer at last. In this phase, ultrasound simulators act a key role which reduces the need of human and cost factor significantly. The need for such real-time simulators has caused various works suggested in this area. They basically either interpolates pre-recorded ultrasound images (interpolative methods) or generates ultrasound images using other modalities i.e, computed tomography, magnetic resonance imaging (generative methods). The drawbacks of these methods respectively are incapability of simulating with various probe positions (view-dependent) and lacking realism due to not solving the wave physics exactly. In this thesis, a real-time ray-based generative ultrasound simulator using the computed tomography images will be presented. The novalties in this work are emanating multi-rays from each transducer element instead of single ray transmission by each of the element and two speckle models based on a local entropy map of computer tomography images and a machine learning algorithm in which some pairs of computer tomography images and ultrasound images of the same body slice are used.