INTEGRATING PHYSICS-INFORMED SIMULATION AND OBJECTIVE ASSESSMENT FOR ROBOTIC SURGERY TRAINING

Abstract

Robotic surgery allows high-precision and minimally invasive operations, but it has a steep learning curve. There are also few ways to measure a surgeon’s progress or to practice safely in a realistic environment. These lead to inconsistent performance across surgeons and institutions. This thesis combines two studies that aim to improve robotic surgery training from two directions: how to measure skill and how to model the surgical environment.The first study focuses on assessing surgical skill in Transoral Robotic Surgery (TORS). A detailed task breakdown called a Hierarchical Task Analysis (HTA) was created to describe every step of the operation. From this, new scoring tools called Procedure-Based Assessments (PBA) were developed to rate how well each step was performed. Forty recorded porcine TORS procedures were reviewed, including both novice and experienced surgeons. The results showed that experienced surgeons performed faster, more precisely, and with better results, proving that the assessment method can reliably measure skill and guide training. The second study introduces a physics-informed surrogate model to make soft-tissue simulation faster and more efficient for realistic surgical training environments. A physics-based approach called Extended Position-Based Dynamics (XPBD) was used to generate data on how soft-tissue moves during contact. A machine learning model with a causal Transformer architecture named X-DeltaFormer was then trained to predict these movements. It reproduced realistic tissue behavior with very small errors and nearly constant computation time, even when the model complexity increased. This approach reduces the computational burden of traditional solvers while preserving realism. Together, these two studies support the development of a robotic surgery training system that can both measure performance and simulate realistic operations efficiently. The first provides a structured way to evaluate skills, and the second offers a fast and accurate way to simulate surgical interactions. This combination can help build better training tools for surgeons and improve patient safety.