A Reconstruction Method For Bioluminescence Tomography Based On Adaptive Hexahedral Voxel Refinement And Atlas Registration

In recent years,bioluminescence tomography?BLT?has become a hot research topic due to its high sensitivity,no radiation and low cost.However,BLT reconstruction is a seriously ill-posed problem.A common way to solve this problem is to increase a prior knowledge,for example,by collecting multispectral data to increase the dimension of measurements,combining structural information to improve the accuracy of modeling,or specifying a permissible source region to reduce the dimension of reconstruction domain.The structral information of a small animal is usually obtained by using a micro computed tomography?Micro-CT?or magnetic resonance imaging?MRI?system.To enhance the contrast of soft tissues,it is necessary to inject contrast agent into the mouse,which may cause accidental death.Organ segmentation is a necessary procedure for obtaining structural information of small animal.Because of the poor contrast of soft tissues,manual segmentation is usually needed.It is not only time-consuming and laborious,but also error-prone.Usually,in the BLT reconstruction process,the tetrahedral mesh is used to numerically discritize.However,the tetrahedral mesh produces a large number of units that need huge storage space.The reconstruction precision is not so good as well.In this thesis,we first propose a BLT reconstruction method based on adaptive hexahedral voxel discretization and atlas registration,and then verify its performance on Micro-CT and MRI data.First,in order to avoid the problem caused by manual segmentation,we develop a atlas registration based segmentation method.Fifty groups of Micro-CT data of mouse were manually segmented by Amira^TM to prepare the single atlas and multi-atlas,which forms the single atlas registration-based and multi-atlas registration-based segmentation methods.Then we used these two segmentation methods to segment the internal organs of the mouse,and quantitatively analyzed the results based on the Dice coefficient and the skin surface distance.The results showed that the atlas registration based segmentation method not only can avoid the problems caused by the manual segmentation and the use of contrast agent,but also provide good accuracy for organ segmentation.Furthermore,the multi-atlas registration based segmentation is proved to be better than the single atlas registration based method.Second,we selected one group from 6 groups of Micro-CT data and 3 groups of MRI data respectively,which has the best registration and segmentation accuracy.We then designed five groups comparative experiments to performe BLT reconstruction,including homogeneous reconstruction,heterogeneous reconstruction,single atlas registration based reconstruction,multi-atlas registration based reconstrution,and adaptive voxel refinement and multi-atlas registration based reconstruction.Therein,the first four groups of reconstructions were performed on tetrahedral mesh,and the last group was on hexahedral voxel.By calculating the position deviation of the reconstructed light source and true light source,we analyzed the reconstruction results quantitatively.The reconstruction results proved that:?1?the accuracy of heterogeneous reconstruction is higher than that of homogeneous method;?2?the accuracy of multi-atlas registration based reconstruction is higher than that of single atlas based method;?3?adaptive voxel refinement and multi-atlas registration based reconstruction method provides better performance than the tetrahedral and multi-atlas registration based method.