Real-time acoustic sensing and artificial intelligence for error prevention in orthopedic surgery
Matthias Seibold, Steven M. Maurer, Armando Hoch, Patrick O. Zingg, Mazda Farshad, Nassir Navab, Philipp Fürnstahl
Abstract
In this work, we developed and validated a computer method capable of robustly detecting drill breakthrough events and show the potential of deep learning-based acoustic sensing for surgical error prevention. Bone drilling is an essential part of orthopedic surgery and has a high risk of injuring vital structures when over-drilling into adjacent soft tissue. We acquired a dataset consisting of structure-borne audio recordings of drill breakthrough sequences with custom piezo contact microphones in an experimental setup using six human cadaveric hip specimens. In the following step, we developed a deep learning-based method for the automated detection of drill breakthrough events in a fast and accurate fashion. We evaluated the proposed network regarding breakthrough detection sensitivity and latency. The best performing variant yields a sensitivity of [Formula: see text]% for drill breakthrough detection in a total execution time of 139.29[Formula: see text]. The validation and performance evaluation of our solution demonstrates promising results for surgical error prevention by automated acoustic-based drill breakthrough detection in a realistic experiment while being multiple times faster than a surgeon's reaction time. Furthermore, our proposed method represents an important step for the translation of acoustic-based breakthrough detection towards surgical use.