Segmentation of Alexis retractors in the context of augmented reality in video-assisted thoracoscopic surgery

This study takes place in the context of video-assisted thoracic surgery (VATS) for lung nodule resection (wedge surgery). The implementation of an augmented reality solution requires registration between the endoscopic intraoperative video image and a CBCT also acquired in an intraoperative context and used for internal localization of the nodule. To achieve this, we need to automatically segment features common to both modalities, namely the Alexis retractors in the operative scene. In this paper we present how to segment retractors Alexis in video images and in intraoperative CBCT. Thanks to their color, segmentation of Alexis retractors is fairly straightforward in video images. A simple thresholding in the Hue-Saturation-Luminosity (HSL) color space is sufficient. However, the segmentation of Alexis retractors is more complicated on CBCT, as there is no contrast in gray-scale values between them and the surrounding tissues. We had therefore to rely only on their surface properties. So we propose the following framework: 1) extract the mesh of the internal surface of the rib-cage, 2) smooth this mesh; 3) calculate the mean curvatures on the vertexes of this mesh. The vertexes of the Alexis retractors boundaries can then be characterized within a specific range of mean curvature values; 4) automatically separate potential Alexis regions by Gaussian Mixture Model and 5) for each Alexis, fit two concentric ellipses on the vertexes matching the curvatures range. This allows each Alexis to be delimit on the CBCT in the form of an elliptic ring. The future registration can then be carried out by superimposing the Alexis rings extracted from the videos and CBCT.