An automated 3D-atlas-based registration towards the anatomical segmentation of pulmonary pleural surface

摘要

A computer-assisted diagnosis system to automatically detect pleural thickening has been developed. The CAD system also carries out the assessment of each detected pleural thickening by specifying its size and location. Moreover, anatomical details i.e. part of pleura (pars costalis, medistinalis, diaphragmatica, spinalis), lung side and position will be automatically determined. For this purpose, a 3D atlas of lung surface has been manually constructed. This paper describes the overall registration chain of patient lung to the 3D atlas which consists of three sub-steps. First, a rigid registration aims a translational alignment of both 3D data. A distance map leads then to an affine registration in the second step. The last final step is a non-rigid registration to assure a better registration quality. After the alignment, each detected pleural thickening will be labeled to describe its anatomical characters which will yield a higher precision and reproducible assessment of pleural thickenings.