Visual modelling and evaluation of surgical skill

Abstract

For many surgical procedures, computer-based training has become an increasingly attractive alternative to traditional training methods. One of the key problems in computer-based surgical training is automatic skill evaluation, which in turn requires skill modelling. To model and evaluate human skills it is necessary to provide the means by which the skills can be measured and interpreted by computers. This paper proposes a new approach for observing continuous, long sequence of hand movements in surgical operation, and then modelling and evaluating the skill demonstrated in the observation. This involves a video-based technique for tracking the hand during a surgical exercise. Because of the non-contact nature of the tracking technique, there is minimal interference with the skill execution, unlike other methods that employ instrumented gloves. To increase the robustness of hand tracking, a Kalman filter (see Appendix A for the system and measurement model) is employed together with a set of coloured markers on the surgical glove. For modelling the surgical skill, a stochastic approach is proposed that uses Hidden Markov Models (HMMs). Using this technique, person-independent models can be developed through human demonstration of particular surgical skills. To automatically evaluate a person’s skill, an objective evaluation criterion is proposed that is based on the log probability of an observation sequence for the given HMM. The probability measures the stochastic similarity between the performance of the observation sequence and the performance represented by the model. This paper also describes an implemented prototype system and experiments that establish the feasibility of the proposed approach for surgical skill modelling and evaluation.