Many medical interventions require inserting a needle towards the desired site. Robotic needle insertion can potentially provide reliable needle access independent of practitioner experience and availability. However, when the needle is inserted into heterogeneous tissue, the complicated needle-tissue interaction introduces multiple challenges to automatic needle steering using 2D ultrasound as image feedback: (1) curved and partially visible needle segmentation; (2) replanning the needle steering path in dynamic environments.
This thesis presents novel methods targeting the above challenges. Our first work focuses on tracking a partially visible curved needle. The second work uses information fusion to utilize both computer vision and robot kinematics to track a straight needle robustly. We then present two mechanical model-based steering replanning algorithms to robotically insert an ordinary clinical needle to reach an internal moving target in the presence of tissue motion and needle bending.
John Galeotti (Advisor)
Zoom Participation. See announcement.