An approach to segment lung pleura from CT data with high precision
Eduard Angelats, Kraisorn Chaisaowong, Achim Knepper, Thomas Kraus, Til Aach · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 2008
A new approach to segment pleurae from CT data with high precision is introduced. This approach is developed in the segmentation's framework of an image analysis system to automatically detect pleural thickenings. The new technique to carry out the 3D segmentation of lung pleura is based on supervised range-constrained thresholding and a Gibbs-Markov random field model. An initial segmentation is done using the 3D histogram by supervised range-constrained thresholding. 3D connected component labelling is then applied to find the thorax. In order to detect and remove trachea and bronchi therein, the 3D histogram of connected pulmonary organs is modelled as a finite mixture of Gaussian distributions. Parameters are estimated using the Expectation-Maximization algorithm, which leads to the classification of that pulmonary region. As consequence left and right lungs are separated. Finally we apply a Gibbs-Markov random field model to our initial segmentation in order to achieve a high accuracy segmentation of lung pleura. The Gibbs- Markov random field is combined with maximum a posteriori estimation to estimate optimal pleural contours. With these procedures, a new segmentation strategy is developed in order to improve the reliability and accuracy of the detection of pleural contours and to achieve a better assessment performance of pleural thickenings.