Gesture Classification in Robotic Surgery using Recurrent Neural Networks with Kinematic Information

Abstract

In this work we introduce the application of Recurrent
\nNeural Networks (RNNs) on surgical kinematic data,
\nfor the classification of gestures in three fundamental
\nsurgical tasks (suturing, needle passing knot tying). The
\ndeveloped RNN-based classifier achieves close to 60%
\naverage classification accuracy for all three tasks when
\ntrained and tested with dVSS kinematic data from the
\nsame operator. Our preliminary work indicates that this
\ntype of artificial neural networks can be the building
\nblocks in gesture classification systems which can form
\nthe basis for further developing automated skill
\nassessment methods in robotic surgery.