| An ideal intelligent surgical assistant system should have task models of surgical procedures, patient-specific anatomical models, and the ability to use this information to predict surgeons' intentions. It should have the ability to provide the surgeon with (1) augmented visual and haptic sensory information about the procedure being performed, (2) augmented ability to position and manipulate surgical instruments, and (3) the ability to perform semi-automated tasks and/or provide virtual fixtures for safe navigation and tissue manipulation.; Endoscopic sinus surgery presents a constrained working environment and limited scene view for the surgeon. At their current state of development, surgical assistants are able to provide limited, though still useful assistances to surgeons. We have used our development of an intelligent robotic surgical assistant system for endoscopic sinus surgery as a motivating example for development of more broadly applicable techniques. In this dissertation, we outline our work on several directions to improve the capabilities of surgical assistance systems. We have created a virtual fixture library of task primitives. This library provides a link between surgeon-understandable task behaviors and low level control for surgical robots. We have automatically generated virtual fixtures for robot-assisted sinus surgery from registered CT models. The virtual fixture generation method enables a real-time task-based control of surgical robot in a precise interactive minimally invasive surgical environment. We have applied an HMM-based method to intelligently recognize operator motions, and choose appropriate virtual fixtures in real-time based on the recognition. This method provides a key element in intelligent user interface for surgical assisted robots. We have developed a scale-invariant registration algorithm to register the scaled structure that reconstructed from monocular endoscope images to pre-operative CT scan. This approach directly registers surgical endoscopic visualization with navigation and robotics, and enables real-time registration without external navigation. |
