Auto-calibration by locally consistent contours for dental CBCT

Abstract

The image quality of dental cone-beam computed tomography (CBCT) is limited by the accuracy of device calibration. Inaccurate calibration introduces errors in the reconstruction process, which may lead to severe artifacts in the reconstructed volume. Patient motion during scan acquisition induces similar effects. This paper introduces a novel auto-calibration approach calculating geometrical projection parameters from unknown patient geometry. We formulate consistency conditions linking the information of consecutive projection images and a regularization technique to prevent overall distortions. Implemented as a global optimization problem we present an efficient greedy optimizer as well. Our strategy turns out to be robust towards inaccurate initialization. As our method does not rely on consistency between projection data and tomography reconstruction it is robust towards reconstruction artifacts such as e.g. truncation. Applying our approach for autocalibration shows a relative improvement of sharpness up to 91% of in a standard dental CBCT setup. Evaluation is performed on digital reconstructed radiographs (DRRs) of a CT head-scan. In particular different motion types are considered and the number of anatomical structures used for calibration is varied to achieve an understanding of the behavior of the approach.