Towards a 3D ultrasound imaging system for needle-guided minimally invasive procedures

Needle-based minimally invasive procedures such as percutaneous biopsies, regional anesthesia, and peripheral vascular interventions are ubiquitous in healthcare. During these procedures, accurate real-time needle localization is vital to minimize risk and improve efficacy. For this purpose, image guidance is utilized, and ultrasound is the desirable modality because of its ease of use, low-cost, non-ionizing radiation and real-time imaging capabilities. However, with ultrasound, needle visualization is inhibited by non-axial specular reflection, hyperechoic artifacts and noise. This dissertation describes the development of accurate, automatic, real-time and robust enhancement and localization techniques for the needle shaft and tip in challenging ultrasound-guided procedures. The developed methods employ both traditional computer vision and advanced machine learning approaches for modeling ultrasound signal transmission, feature detection and pixel-wise classification of needles from 2D and 3D ultrasound data. In extensive ex vivo imaging studies on realistic phantoms across a broad range of imaging settings and scenarios, the methods accurately localize needles that have low contrast or are imperceptible to the naked eye. The results of these studies will inform future clinical trials for evaluating the feasibility of our methods. Once translated, this work will provide means for a new ultrasound imaging platform to support real-time enhancement and localization of needles and will be applicable to commercially available 2D/3D cart-based and portable ultrasound systems thus benefitting clinical practice, research and industry.Ph.D.Includes bibliographical reference