Freehand ultrasound imaging : efficient simulation, volume reconstruction and application to augmented reality

Overview & Synopsis

This thesis focuses on the relationship between a known 3-D volume and the ultrasound images obtained from this medium. This manuscript explores two main axes related to this question: Ultrasound image simulation, which enables the creation of ultrasound images from any 3-D medium information; and 3-D Ultrasound volume reconstruction, which aims to retrieve this 3-D information from a collection of ultrasound images. Additionally, these approaches are examined in the context of their use with augmented reality, which introduces new challenges and promising applications. First, a method to simulate ultrasound images from any position and orientation, from any given 3-D medium in real time is explored. This method is based on the standard scatterer paradigm for 2-D ultrasound simulation and is extended to a 3-D framework with a particular interest in time and memory performances, enabling the creation of physically plausible scatterers and by extension simulations multiple times by seconds. This method finds a direct application in augmented reality, where it can be used to create realistic datasets or training programs without any need for an ultrasound system. From these datasets, or from a real ultrasound acquisition, 3-D volume reconstruction enables the retrieval of the medium that has been imaged. The proposed method generates a smooth volume in less than 30 seconds by leveraging the properties of neural implicit representation. This is achieved by training a network for each scene, which then provides a continuous representation of the scene that can be saved and transmitted for further use. Potential uses include creating a 3-D model, treating the volume, or directly displaying it. A straightforward application to augmented reality is a display of the volume right after an acquisition by a practitioner, which allows for a more detailed visualization of the medium.