Background: Availability of patient-specific image data, gathered from preoperatively conducted studies, like computed tomography scans and magnetic resonance imaging studies, during a surgical procedure is a key factor for surgical success and patient safety. Several alternative input methods, including recognition of hand gestures, have been proposed for surgeons to interact with medical image viewers during an operation. Previous studies pointed out the need for usability evaluation of these systems.
Objectives: We describe the accuracy and usability of a novel software system, which integrates gesture recognition via machine learning into an established image viewer.
Methods: This pilot study is a prospective, observational trial, which asked surgeons to interact with software to perform two standardized tasks in a sterile environment, modeled closely to a real-life situation in an operating room. To assess usability, the validated "System Usability Scale" (SUS) was used. On a technical level, we also evaluated the accuracy of the underlying neural network.
Results: The neural network reached 98.94% accuracy while predicting the gestures during validation. Eight surgeons with an average of 6.5 years of experience participated in the usability study. The system was rated on average with 80.25 points on the SUS.
Conclusion: The system showed good overall usability; however, additional areas of potential improvement were identified and further usability studies are needed. Because the system uses standard PC hardware, it made for easy integration into the operating room.
Georg Thieme Verlag KG Stuttgart · New York.