Human robot interface and control methods for steerable catheter procedures in neurosurgery

Abstract

This thesis explores novel human-machine interface hardware, methods and algorithms for robotic assisted surgical needle steering, including the design and validation of a novel neurosurgical robotic platform which is part of the Enhanced Delivery Ecosystem for Neurosurgery in 2020 (EDEN2020) European Research and Innovation Action. At the heart of this project is the bio-inspired Programmable Bevel-tip needle (PBN), which is a soft, steerable, flexible, multi-segment catheter that is able to follow curvilinear 3D paths in soft tissue. An electro-mechanical design that addresses aspects of standards applicable to medical device design, was completed and integrated in order to drive a 4-segment PBN. This system has been used as the basis for the research presented in this thesis, in understanding the optimal control modalities to move the catheter through soft tissue, and in designing an intuitive human machine interface that is appropriate for clinical use.

The PBN design and actuation is inspired by how some inspects can penetrate and steer through a medium in 3D with their ovipositor in order to lay eggs. The motion is achieved by simultaneously extending and retracting different sections of the ovipositor, and using the forces from the medium to push (or steer) the tip of a section in a particular direction. This reciprocating motion minimises the net pushing force, and in doing so, decreases the displacement and strain applied on the medium. This same technique is applied when actuating a 4-segment PBN, and low level control modalities can exploit the kinematics of the catheter in following clinically advantageous motion profiles that may reduce damage to the tissue along the insertion tract. The aim of this research is to optimise a controller that can simultaneously ensure accurate path following and target reaching performance for the catheter, while minimising the displacement, hence possible damage, to the surrounding tissue.

A cyclic motion controller was compared against a direct push controller in order to understand the performance of the catheter when following a path and reaching a target over 3D trajectories. An expert user was able to achieve a target position error of $0.58 \pm 0.68$ mm for the direct controller, and $1.45 \pm 1.41$ mm for the cyclic controller, both clinically viable results. The difference in these metrics was found to be due to the cyclic controller under-steering, as it takes more time for the desired offset between the catheter segments to be reached compared to the direct push controller. As such, a hybrid controller that uses both cyclic and direct motion profiles during an insertion was implemented, and an expert user achieved a target position error of $0.7 \pm 0.69$ mm. In this way, the accuracy of the direct push controller can be maintained, while a patient can still benefit from the reduced tissue strain resulting from the cyclic profiles.

As the system is designed for human-in-the-loop control, the Human Machine Interface (HMI) is vital in allowing users to intuitively, safely and accurately use the system. The HMI proposed here consists of the visual interface the surgeon is shown during the neurosurgical procedure and the haptic interface they interact with via a control joystick. Having designed and implemented appropriate hardware for the task, a further objective of this research was to assess the impact different interfaces have on the performance of the system, in order to identify the best combination of feedback via human studies trials.

A standard neurosurgical visual interface depicts slices in the Axial, Coronal and Sagittal planes of the brain as well as a 3D volume based on preoperative imaging. It is easy for the surgeon to visualise the 2D path a needle would take through the brain, as they can extrapolate the needle tip position easily from one slice to the next. However, the same method does not work for 3D needle insertions. In this thesis, a novel visual interface was developed that combines both first person and third person viewpoints, so the surgeon can navigate with the catheter as if they were driving a vehicle, and still see the full brain volume perspective. Overlays are used to show the desired path the surgeon should follow, as well as waypoints for them to navigate through. The visual interface was evaluated by user trials \textit{in vitro} and in simulation and by \textit{ex vivo} assessments during an ovine trial. The best \textit{ex vivo} results had a target position error of $0.29$ mm, and an orientation error of $9.38$ degrees.

A haptic interface included in the HMI employs a 2 degree of freedom haptic joystick to render forces to the user's hand that can be used as navigational cues. A user study evaluated if haptic feedback increased the user's performance, however the results concluded that haptic feedback did not significantly increase the performance if visual feedback was also present.

A discussion of the results and their implications has led to suggestions for future work.