P. Wochner. Image-Based Tracking of Laparoscopic Instruments. Institute of Biomedical Engineering, Institute for Anthropomatics and Robotics, Karlsruhe Institute of Technology (KIT). Masterarbeit. 2015
Abstract:
Laparoscopic surgery, which offers considerable benefits to patients, has been steadily gaining popularity. In addition to reducing blood loss and post-operative pain, the recovery time can be reduced. However, a laparoscopy is a complex intervention with several disadvantages for the surgeon. The endoscope, for example, covers only a small part of the body inside, resulting in a reduced field of view for the surgeon. The aim of a context- aware assistance system is to overcome some of these drawbacks by providing the surgeon relevant information at the right time. This can be achieved if the current intervention phase is known. To this effect, information about the laparoscopic instruments involved is required. There is already a method that can detect and identify the instruments. However, this method still shows some shortcomings, for example, when two instruments are in close proximity to each other. In this thesis, an image-based method for tracking laparoscopic instruments is developed that both builds on the detector and complements it. The temporal component provides new information that deals more effectively with difficult situations than the detector does. First, features are extracted for each detected instrument. Using the Lucas-Kanade algorithm, the optical flow for these features is determined and used to propagate the instruments in time. The method is evaluated on videos from both in-vivo surgeries and ex-vivo experiments with the DaVinci system for robotic surgery. In addition to the accuracy, the runtime of the method was examined. With an average of 10.1 ms per frame, the tracking can be done in real time and is significantly faster than the detection.