Surgical video retrieval using deep neural networks
Christos Varytimidis, Konstantinos Rapantzikos, Constantinos Loukas, Stefanos Kollias · Lincoln Repository (University of Lincoln) · 2016
Although the amount of raw surgical videos, namely videoscaptured during surgical interventions, is growing fast, automatic retrievaland search remains a challenge. This is mainly due to the natureof the content, i.e. visually non-consistent tissue, diversity of internal organs,abrupt viewpoint changes and illumination variation. We proposea framework for retrieving surgical videos and a protocol for evaluatingthe results. The method is composed of temporal shot segmentation andrepresentation based on deep features, and the protocol introduces novelcriteria to the field. The experimental results prove the superiority ofthe proposed method and highlight the path towards a more effectiveprotocol for evaluating surgical videos.