Spatial Compounding and Segmentation of Volumetric Ultrasound Data Sets for Generating Interactive Anatomic Models

author:Jeremy Wynne Cannon
adviser:Robert D. Howe, Derek Rowell
degree: S.M.
institution: Massachusetts Institute of Technology

Surgical procedures guided by non-optical imaging represent a growing proportion of medical interventions. The goal of this type of approach is to reduce the invasiveness of current procedures or to treat disease using new procedures which would not possible without the assistance of medical imaging. The field of cardiac surgery represents a medical subspecialty where image-guided interventions could markedly improve current treatment methods. In particular, all surgical procedures performed inside the heart require the use of cardiopulmonary bypass and cardiac arrest even when performing these procedures using "minimally invasive" techniques. Side effects of this bypass process range from a mild inflammatory response to multi-organ dysfunction.
This thesis investigates the use of real time 3 D ultrasound (US) for possibly guiding surgical procedures inside the heart without using cardiopulmonary bypass. In a series of in vivo studies in an animal model and an in vitro tank study, the limitations of both 2 D US and 3 D US for guiding surgical tasks were systematically identified. These initial studies confirmed the inadequacy of 2 D US for guiding complicated surgical maneuvers and verified the utility of real time 3 D US for efficiently guiding both basic and more complex surgical tasks. However, several important problems with using US for guiding surgical procedures were identified.
First, because the surgeon is entirely dependent upon the US image for instrument positioning, additional safety measures are required to prevent inadvertent injuries. In addressing this problem, a computationally efficient active contour segmentation model was applied to the volumetric US images. This algorithm (developed recently by Perrin at the University of Minnesota) proved sufficiently fast to keep pace with the volumetric US data stream with good accuracy. Thus, this approach can be applied to establish a "virtual fixture" inside the heart wall during these surgical procedures to prevent injuries.
Another significant problem present during US guided procedures (2 D and 3 D) involves maintaining spatial orientation. With 3 D US-guided procedures, the image affords sufficient spatial cues to guide the procedure once the tools are within the imaging field of view. However, guiding the tools into view and keeping them there proves quite challenging. In addition, maintaining orientation relative to known anatomic landmarks during the procedure also proves challenging. To address these problems, a spatially compounded US image of the heart was produced and a graphical model was registered to this data set. Such a registered graphical model could be used as a navigational aid for the surgeon during image-guided procedures inside the beating heart.

Harvard BioRobotics Laboratory Home