This research seeks to achieve a deeper understanding of the critical computer vision components and techniques needed to develop a new generation of image guided surgery systems. Such systems are important to the successful use of the emerging class of minimally invasive surgical procedures. Effective use of image guidance will improve the accuracy of such procedures, broaden the range of applicable procedures, and enable interactive patient-specific anatomical modeling. The key computer vision issues to be addressed are segmentation, data registration, instrument tracking and flexible modeling of structures. This research will develop methods for utilizing anatomical atlases and other anatomic knowledge, to provide structure-specific segmentation templates. Such atlases will need to be registered to the patient's data set, requiring new registration algorithms, that deal with flexible and deformable structures. Motion tracking algorithms are needed to monitor the position of surgical instruments within the operating field. By utilizing Mutual Information based registration algorithms, this proposal will develop tracking methods that match models of the surgical instruments to passive views of the operating field, enabling 3D tracking of those instruments. Finally, the tracked position of the surgical instruments, coupled with detected variations in images of the surgical field itself, the 3D models of the patient's anatomy will be automatically updated to reflect removal of tissue and changes in position of anatomical structures.
