PhD Oral Exam - Francis Boabang, Information System Engineering

When studying for a doctoral degree (PhD), candidates submit a thesis that provides a critical review of the current state of knowledge of the thesis subject as well as the student’s own contributions to the subject. The distinguishing criterion of doctoral graduate research is a significant and original contribution to knowledge.

Once accepted, the candidate presents the thesis orally. This oral exam is open to the public.

Abstract

Robotic surgery offers a multitude of benefits to patients, including quicker recovery, reduced bleeding, minimized discomfort, and heightened precision. The emergence of remote robotic surgery has further transformed the medical landscape by enabling surgeons to perform procedures from distant locations. This capability not only enhances access to specialized care but also has the potential to enhance surgeons' expertise. Nevertheless, the critical nature of remote robotic surgery underscores the paramount importance of system reliability. Reliability in this context encompasses seamless communication between the surgeon and the patient domain, ensuring prompt execution of commands and accurate relay of feedback. However, challenges such as communication disruptions, such as packet loss and delays, can hinder this essential exchange of information, potentially compromising patient outcomes.

From an algorithmic perspective, reliability also extends to the robot's ability to respond effectively to unforeseen circumstances, such as instrument slippage or excessive tissue force. Overcoming these challenges requires innovative solutions, with predictive models emerging as a promising avenue for improving reliability. Machine learning emerges as a powerful tool for improving the reliability of remote robotic surgery. However, leveraging modern machine learning methods in this field presents its own set of obstacles. Primarily, conventional optimization algorithms employed in machine learning frequently are often ill-suited for remote robotic surgery. To address reliability concerns in remote robotic surgery, concerted efforts are needed to refine and optimize various components of machine learning training. By enhancing optimization schemes, including loss functions, optimizers, and model compression techniques, we can enhance the resilience and effectiveness of remote robotic surgical systems. This advancement ultimately translates into improved patient care and surgical outcomes.

The first segment of the thesis introduces an innovative low-rank matrix factorization scheme aimed at enhancing the scalability of machine learning and bolstering reliability in remote robotic surgery. Specifically, our proposal entails employing a low-rank matrix factorization approach to address issues related to packet loss and delay in remote robotic surgery. We advocate for a nonconvex formulation of a low-rank matrix factorization (SRLSMF) with convex formulation of a low-rank matrix factorization initialization (SRLSMF) to scale GPR. Thus, by employing convex relaxed low rank matrix factorization initialization to scale a given the GPR model, the model can avoid local minima and converge to a solution with smaller recovery residuals. Also, the running time of convex nonconvex low rank matrix factorization is expected to be smaller than that of applying nonconvex low rank matrix factorization alone under the same stopping criterion. It should be noted that we are the first to exploit nonconvex formulation of a low-rank matrix factorization (SRLSMF) with convex formulation of a low-rank matrix factorization initialization (SRLSMF) to scale machine learning in machine learning domain. Recognizing the cost-prohibitive nature of standard eigen decomposition for online GPR covariance update, we implement incremental eigen decomposition within the SRLSMF and SRLSMF GPR methodologies. Finally, an illustration of the potential applications in suturing, knot-tying and needle passing task using kinematic dataset is provided.

In the latter part of the thesis, we present a novel approach called adaptive stochastic gradient descent (ASGD) method, which leverages the non-uniform p-norm concept to train machine learning. The proposed ASGD assigns distinct categories of coordinate values with varying base learning rates, thereby enabling the training of machine learning models to provide reliable predictions to real-world applications including remote robotic surgery. Additionally, we provide theoretical guarantees for the efficacy of the proposed ASGD method in convex and nonconvex settings. We demonstrate the ability of our proposed ASGD approach to detect suturing gestures within the remote surgical gesture recognition task.

Finally, we wrap up this thesis by outlining potential avenues for future research.