A data glove for a new surgical training tool

AIM The understanding of surgical gesture, by means of a measuring apparatus, can play a key role in the evaluation of surgical performance. To this aim, a neural network classification algorithm can be helpful, since it combines good generalization performances along with a parsimonious architecture when dealing with high dimensional classification problems. We present its use as a surgical training tool for surgery, a field of research highly underrepresented in the surgical teaching scenario. We operated a bounding box decomposition of surgeon’s hand movements analysis and gesture recognition during training of novice surgeons. This feature was applied to analyze trajectories of surgeon’s wrist and finger postures, so to recognize different hand gestures. METHODS Dataset of surgical gestures: a team composed by expert surgeons and attending surgeons performed exercises focused on basic surgical technical skills (interrupted and running suture) Gesture measurement: we developed a data glove on the basis of acquired experiences. This glove is provided with bending sensors capable to measure movements of distal interphalangeal, proximal interphalangeal, metacarpo phalangeal finger joints and inertial sensors to measure wrist posture. Trajectories of surgeon’s wrist and fingers were recorded and we analyzed the dataset of surgical gestures to evaluate parameters as execution time and repeatability of the gesture. Gesture classification: in order to classify each gesture, we focused on the synthesis of an algorithm that automatically assigns each gesture to a predefined class: master, resident or attending surgeons. RESULTS Operator’s training: Currently, mentors transfer their expertise to trainee via practical demonstrations and oral instructions. With recorded data of measures it is possible to reproduce such movements via avatar representation on a PC screen. It gets the important aspect that the same gesture can be represented several times always in the same manner and that it is possible to look at the gesture from all possible points of view, just rotating, translating, zooming the avatar. Furthermore, we intend to develop a graphical interface capable to superimpose a “ghost” avatar of the learner upon the “guide” avatar of the expert. In this manner the trainee will be capable to easily auto-evaluate her/his performance with instinctive ability. CONCLUSIONS This work, still in progress, would be an innovative, accurate and non invasive method to measure and evaluate surgical gestures. It will be useful to accelerate the in-training surgeon’s learning curve who can compare the basic level of his expertise with master surgeon’s level and verify step by step his improvement.