A neural network architecture for automatic and objective surgical skill assessment

Improving surgical skills through various training programs is critical for healthcare. To reach this goal, it is important to increase the reliability of surgical skills assessment. In this paper we propose an objective and automatic procedure to evaluate the performance of trainees on three different tasks: Knot tying, needle passing and suturing. The core of the evaluation procedure is a fully connected deep neural network with three hidden layers and twenty nodes for each layer. The network is trained exclusively on expert database by assuming the tasks being already known and classified. Our framework is then used to evaluate intermediate and novice trainees by providing scores referenced on the performances of the experts. The results show experimentally the ability of our network architecture to automatically evaluate classified surgical tasks.