Synchronized brain dynamics and eye movement trajectory for objective evaluation of robot-assisted surgical skills

Complicated and costly robot assisted surgery (RAS) training results in less frequent use of this technology in
several complex areas of surgery, and consequently ends up in harm. RAS requires a unique blend of skills in
addition to manual competence with human-machine interaction skills, while operating remotely from patient
with no tactile feedback. To address this challenge, numerous studies have focused on simulation-based
robotic training curricula, like Fundamental Skills of Robotic Surgery (FSRS), to develop and assess the
performance level of the surgeon operator. However, such training tools were developed based on metrics
measured by performance on a simulator and other subjectively evaluated metrics. The goal of this research
proposal is to develop a tool for objective RAS skill assessment and a model for performance
monitoring. We hypothesize that brain dynamics - Electroencephalogram (EEG) - and eye movement
behavior are able to detect change of skill level and the level of surgeon’s performance. To validate this
hypothesis, we will record EEG signals and eye movement time series from subjects with different RAS
expertise levels. Ten novices, 5 beginners, 5 advanced beginners, and 5 expert surgeons will be included in
the study and continuously perform four levels of designed RAS training tasks on surgical robot simulator, dry
lab, and animal lab during one year; (1) performing six basic tasks on surgical simulator. All subjects will
practice these tasks during two weekly sessions and each practice session takes 2 hours. (2) Subjects will
practice 3 tasks of peg transfer, pattern cutting, and suturing on dry lab. (3) Subjects will practice 2 tasks
(anastomosis and dissection) on animal tissue and also on plastic models. (4) Subjects will practice two
operations of nephrectomy and hysterectomy on animal lab, 2 operations in each session, and each session
takes 3 hours and occurs every other week. Two master surgeons will subjectively evaluate performance of
subjects (all 25 subjects; Score scale: 1-20) and expertise level (four categories) in performing the designed
tasks, every practice session. Master surgeons evaluate surgeon’s skill and performance throughout task and
notify change of skill level and performance through time.
We will then develop a ‘deep convolutional neural network’ algorithm trained by EEG and eye movement time
series through running windows with equal size, to classify subject skill level into four categories of a novice,
beginner, advanced beginner, and expert. We will also use network neuroscience techniques to extract
engineered features from EEG and eye movement data and use them for training a regression algorithm to
develop a model for performance level prediction. Ultimately, the developed objective skill evaluation tool and
performance monitoring model will make RAS training more efficient by providing feedback to the trainee
regarding his/her skills and directing him/her to focus on skills needed improvement. These improvements will
result in more frequent use of RAS in complex surgical areas and ultimately lead to patient safety.

Grant Number: 5R01EB029398-04
NIH Institute/Center: NIH
Principal Investigator: Somayeh Besharat Shafiei