Online Tracking and Retargeting with Applications to Optical Biopsy in Gastrointestinal Endoscopic Examinations.

Menglong Ye,Stamatia Giannarou,Alexander Meining,Guang-Zhong Yang

DOI: https://doi.org/10.1016/j.media.2015.10.003

IF: 10.9

2015-01-01

Medical Image Analysis

Abstract:With recent advances in biophotonics, techniques such as narrow band imaging, confocal laser endomicroscopy, fluorescence spectroscopy, and optical coherence tomography, can be combined with normal white-light endoscopes to provide in vivo microscopic tissue characterisation, potentially avoiding the need for offline histological analysis. Despite the advantages of these techniques to provide online optical biopsy in situ, it is challenging for gastroenterologists to retarget the optical biopsy sites during endoscopic examinations. This is because optical biopsy does not leave any mark on the tissue. Furthermore, typical endoscopic cameras only have a limited field-of-view and the biopsy sites often enter or exit the camera view as the endoscope moves. In this paper, a framework for online tracking and retargeting is proposed based on the concept of tracking-by-detection. An online detection cascade is proposed where a random binary descriptor using Haar-like features is included as a random forest classifier. For robust retargeting, we have also proposed a RANSAC-based location verification component that incorporates shape context. The proposed detection cascade can be readily integrated with other temporal trackers. Detailed performance evaluation on in vivo gastrointestinal video sequences demonstrates the performance advantage of the proposed method over the current state-of-the-art.