Vinci system, robotic minimally invasive surgery (RMIS) has crossed
the threshold from the laboratory to the real world. As a consequence,
there is now an emerging interest in methods for evaluating and
teaching RMIS surgical technique. Our group has been developing
statistical methods for modeling RMIS using techniques borrowed from
speech and language. We consider surgery to be composed of a set of
identifiable tasks which themselves are composed of a small set of
reusable motion units that we call “surgemes.” We thus speak of a “Language of Surgery.”
In this talk, I will describe our progress at developing techniques
for recognizing surgemes in continuous motion recorded during benchtop
models of RMIS tasks. We have demonstrated that it is possible to
recognize surgemes reliably in a diverse corpus of user motion and
video data. Further, we have begun to develop methods for comparing
the performances of subjects to evaluate skill at both the task and
surgeme level. In particular, we now have strong evidence suggesting
that a simple notion of string distance on the output of a language
model is a natural way of measuring similarity in the space of
surgical skill. These models lead naturally to a set of methods for
effective training of RMIS using automatically learned models of
expertise, and toward methods for automating component actions in
surgery.