Open Collections

Schmidt, Adam

University of British Columbia

This thesis addresses the problem of tracking surgical tissue using camera data in robotically assisted minimally invasive surgery. Millions of robotically assisted surgeries are performed yearly. These surgeries are performed by a surgeon who uses a teleoperation console. This console gives the surgeon a 3D view of their environment as they perform surgery. If we would like to overlay preoperative scans onto the surgeon's field of view using augmented reality, we must understand where the tissue is and where it is moving. Additionally, to enable automation of tasks, or remembering measurements made at tissue locations, such as a biopsy, we require robust tracking. In this thesis, we first perform an in-depth review of the field. We then propose multiple interlocking pieces to enable tissue tracking. To enable tracking salient features, we design a real-time keypoint descriptor that is trained in an unsupervised manner, demonstrating improved performance over classical methods. Afterwards, we propose a novel method that uses these keypoints to parameterize motion in 2D space using graph neural networks. We demonstrate the efficiency of this graph interpolation method. We then incorporate a temporal model into this graph interpolation paradigm. Finally, we extend our graph interpolation algorithm into 3D. In developing our methodology, we realized that there is an acute need for more datasets for quantification. We develop a novel and complete dataset for tissue tracking and mapping and release our dataset for public use by researchers. We close with a summary of important future work.

Text

2024-03-18

Attribution-NonCommercial-ShareAlike 4.0 International

http://hdl.handle.net/2429/87592

Electrical and Computer Engineering

University of British Columbia

UBCV

http://creativecommons.org/licenses/by-nc-sa/4.0/