Evaluating the Mental Workload During Robot-Assisted Surgery Utilizing Network Flexibility of Human Brain

Abstract

Mental Workload (MWL) is traditionally evaluated by psychophysiological signals using spectral analysis and event-related potentials. Robot-assisted Surgery (RAS) is a complex task that involves human-robot interaction, multitasking, quick and appropriate reactions to various stimuli and unforeseen circumstances, as well as frequent switches between surgical subtasks. There is a lack of standardized methodology for objectively monitoring a surgeon's MWL during RAS. In this study, we propose an innovative framework, using dynamic functional brain network measurements and a deep convolutional neural network, to assess MWL. A model was developed and validated using Electroencephalogram (EEG) data from 22 trainees who performed basic surgical tasks, as well as four surgical fellows and an expert surgeon who carried out cystectomies and prostatectomies. The resulting accuracies of the MWL classification into low, intermediate and high were 93%, 89%, and 91% respectively. The proposed method can be used for continually monitoring mental workload levels in an objective fashion.