Robotic motion compensation for 3D ultrasound-guided beating heart surgery

Beating heart surgeries offer significant health benefits to patients by removing the need for the heart-lung machine and its attendant side effects. These surgeries are challenging to perform and only feasible in certain types of procedures because of the rapid movement of the heart. Equipping the surgeon with fast, actuated, and intelligent surgical instruments that automatically compensate for heart motion could facilitate the execution of existing beating heart procedures and enable the development of new procedures that are currently not possible. These tools have particular promise for intracardiac beating heart procedures, where passive tissue stabilization techniques are not available; however, achieving motion compensation in this setting is challenging because of the sensing and space restrictions imposed from working inside of the beating heart. This thesis investigates 3D ultrasound-guided robotic motion compensation as an assistive technology to intracardiac beating heart surgery. A number of engineering challenges are addressed to develop a viable system for in vivo experimentation: heart motion prediction to counter time delays in 3D ultrasound imaging and image processing, real-time tracking of surgical targets in noisy 3D ultrasound images, and safe force control schemes for the manipulation of tissue without exciting vibratory modes in the robot. Solutions are provided in the form of a quasiperiodic extended Kalman filter, a synergistic "flashlight" tissue tracker, and a force controller with feed-forward target motion information, respectively. Integrating these components into a system, motion compensation within the beating heart is not only shown to be feasible under in vivo conditions, but also to provide significant performance advantages in beating heart tasks. Motion and force tracking accuracies of 1.0 mm and 0.11 N are obtained in in vivo surgical tasks with the system, constituting a 70% and 75% reduction in error when compared to human performance in the same tasks.