Research And System Design On Deformation Algorithm Of Registration Of 3D Whole Mouse Brains

Brain science is increasingly becoming the focus of science and technology strategy around the world.Such as brain imaging,neural tracking and analysis are the key cornerstones of research in the field of brain science,and have very forward-looking significance for the prevention of brain diseases and artificial intelligence-oriented brain-like research.At present,most of the research on the internal mechanism of the brain is based on the mouse.In order to study the morphology and brain area projection of the neurons in the mouse brain,it is necessary to register the mouse brain image to the standard template mouse brain.And use the same transformation parameters to map neurons to the template mouse brain for unified analysis.Although traditional deformation field can simulate a certain degree of distortion in the image,it is not completely applicable to the three-dimensional mouse brain data set.This article focuses on the study of the deformation field method in rat brain image registration,and explores the effects of different deformation algorithm models for 3D rat brain images on its registration accuracy and speed,and applies it to neuron mapping.The main contents of this paper are as follows:(1)Large difference in image deformation of mouse brain,a discrete point multi-level Bspline deformation is introduced.Use the anatomically feature points in the mouse brain to optimize the grid node offset,and add cascade to make it both smooth and accurate.Compared with the traditional B-spline,increased the average DICE of each brain area by 4.82% after registering 31 rat brain images.And the time is only 27.63 seconds.On this basis,the smooth thin plate spline deformation model is introduced to further improve the registration accuracy.(2)Aiming at the out-of-bounds problem of the STPS deformation model mapping neurons to the standard template mouse brain,the application of the L-BFGS optimization algorithm to iteratively search for the best mapping point of the neuron from the same direction of registration is studied.(3)An interactive visual registration prototype system is designed,which can select the deformation model according to the actual data during registration,and can intuitively display the registration effect.