A Segmentation Method Of Mouse Organs In Micro-CT Image Based On Atlas

Fluorescence molecular tomography is a kind of in-vivo molecular imgaing technology which can realize the localization and quantification of fluorophore. And it has been widely used in disease diagnosis, disease treatment, drug development and other fundamental researches. However, due to the limitation of the reconstruction algorithm, the reconstruction quality and the resolution are not ideal. By coupling with CT imaging, structure priori information can be introduced into reconstruction algorithm, which can improve the accuracy of the forward problem and restrained the inverse problem. As a result, the reconstruction quality will be improved. Due to the low contrast of CT images, only high contrast organs can be segmented without auxiliary. In this paper, a highly automated segmentation method based on mouse atlas is presented which can also obtain the structure information of the low contrast organs simultaneously. First, the high contrast organs are segmented. Then the bones and lungs of atlas domain and CT domain are registered by establishing a hierarchical registration model. After that, corresponding landmarker sets on bones and lungs are obtained based on the previous registration, so as to a sparse corresponding landmarker set on skins. Then the landmarks on skins are extended to the whole skins by local geodesic distance distribution and normalized ratio. At last, the low contrast organs are mapped from atlas domain to CT domain by the thin plate spline interpolation algorithm based on all the corresponding landmarkers. We compare the results using manual segmentation after contrast enhancement. The Dice similarity coefficient of bone, lung, heart, liver and kidney obtained were from 0.55±0.08 to 0.78±0.04 and the average Euclidean distance between the organs' surface obtained were from 0.36±0.01 mm to 1.13±0.23 mm. The accuracy of this paper's segmentation method is equivalent to or even better than the similar methods based on individual atlas and is worse than the similar methods based on statistical atlas.