Video Scene Tagging from Awake Craniotomy Sound Recodes

Abstract

 Medical technology has been highly developed, however advanced medical technique depends on expert surgeons. In our study, we aim to analyze surgeon’s knowledge and judgment process from surgical video records. In general, surgical video records are objective data, so they are widely used for such as detection of important scene and knowledge discovery. However, the length of surgical video records is more than 10 hours, so automatic tagging method is required for detecting such as significant scene and occurred events labeling.
 In this paper, we propose automatic tagging using audio information for awake craniotomy videos recorded in Tokyo Women’s Medical University Hospital. The timing of electrical stimulation for brain cortex is tagged from recorded sound. After that, we construct speaker model by Gaussian Mixture Model (GMM) and to index speaker from surgery sounds. In the experiments, we validated electrical stimulation timing detection and speaker labeling. The results show that awake craniotomy videos are able to tag by audio information.