Robust vessel registration and tracking of microscope video images in tumor resection neurosurgery

Abstract

This paper proposes a new method designed to track operative microscope video images recorded during tumor resection neurosurgery. Two steps are involved in this method. The first uses feature vectors constructed from color information of video images and shape information of selected vessels to find homologous points in consecutive frames. The second uses smoothing thin-plate splines (TPS) to interpolate the transformation computed with the vessels over the entire image. This approach only requires several pairs of starting and ending points selected on segments of vessels in the first frame of a video sequence. Then, the proposed method tracks the identified vessels automatically, rapidly, and robustly, even when surgical instruments obscure parts of the image frames.