Workflow Recognition from Knee Surgical Videos: Role of Deep Neural Networks

Abstract

The Orthopedic surgery differs from its counterpart surgery, such as laparoscopic and laparotomy, because of a large variety of surgical techniques. Furthermore, the procedures are complicated, and many types of equipment's have been using in the Orthopedic surgery. So, nurses who deliver surgical instruments to surgeon are supposed to be forced to incur a heavy burden. In our previous work, the navigation system for assisting operating room nurse in Orthopedic surgery-Unicompartmental knee arthroplasty (UKA)-was proposed, and we achieved satisfactory accuracy for operation and out-of-operation phase detection. In this work, we propose method for improvement of recognizing Orthopaedics procedures from video images of UKA and TKA (Total knee arthroplasty) by deep neural network. Firstly, we construct the recognition model and then evaluate the recognition of procedures using convolutional neural network, and finally investigated the role of deep and densely connected neural nets for improving the recognition accuracy. The outcome confirmed the improved recognition rate with deep and dense layers, however, further improvement is yet to discover in future.