Learning Invariant Representation of Tasks for Robust Surgical State Estimation

Surgical state estimators in robot-assisted surgery (RAS)-especially those trained via learning techniques-rely heavily on datasets that capture surgeon actions laboratory or real-world surgical tasks. Real-world RAS are costly to acquire, obtained from multiple surgeons who may use different strategies, and recorded under uncontrolled conditions highly complex environments. The combination of high diversity limited data calls for new methods robust invariant operating techniques. We propose StiseNet, a Task Invariance State Estimation Network with an invariance induction framework minimizes the effects variations technique environments inherent datasets. StiseNet's adversarial architecture learns separate nuisance factors information needed estimation. StiseNet is shown outperform state-of-the-art estimation three (including dataset: HERNIA-20).