0
喜欢
0
书签
声明论文
Evaluation of four 3D non rigid registration methods applied to early zebrafish development sequences   
摘  要:   Abstract—Cell image,registration can be used to answer,the question ”where and,when,do cells move?”. It is a crucial issue to describe vertebrate development,and explore the relations be- tween cell mechanical,activities and the formation,of macroscopic structures. Thus motion,estimation,using registration is a first step towards,the understanding,of cells and tissues biomechanics. The present study aims at characterizing,registration algorithm ability to capture,cells and tissues displacements. An evaluation protocol,has,been,set up,in order,to measure,the error pro- duced by a given algorithm. Evaluation relies on a reference, a gold standard,of cells trajectories. The difference between,this reference,and,the cell trajectories built by,registration is the error measurement.,Four non-rigid registration methods,used in the biomedical field have been tested: Bspline elastic registration, MIA fluid registration, ITK demons optical flow and ITK levelset optical flow. Two artificial datasets were,registered with these methods. We measured,errors and,their distributions with our gold standard,to assess registration quality. Results show,that all methods,are well suited to achieve,nuclei registration with an accumulated,error remaining,below,the voxel size for 80% of the cells. BSpline gave the best registration under,five time steps of error accumulation,whereas,ITK demons,was the best for longer accumulation. The same,method,were,used on zebrafish development,3D data. Inaccuracy in the gold standard,produced by automated,tracking,method,compromised,the accuracy,of our error measurements.,We however,concluded,that the perfor- mances,of the four tested methods,are in accordance,with the

论文统计图