Research On The Algorithm For Constructing Magnetic Resonance Diffusion Tensor Image Templat

Diffusion tensor imaging(DTI)is a novel and non-invasive in vivo brain imaging technique that enriches the representation of microscopic anatomical and functional features of the human brain.DTI atlas plays an important role in brain science research.High-quality DTI atlas can greatly facilitate not only the study of brain structure-function relationship,clinical medical research and neuroanatomy research,but also the development of artificial intelligence and intelligent medicine.However,traditional diffusion tensor atlas construction algorithms are timeconsuming and the constructed atlas can hardly capture the variability of the global data set,and there are still few reports on drug addiction-based DTI atlas.To address this situation,this paper investigates the diffusion tensor image alignment algorithm based on deep learning,and then designs a diffusion tensor atlas construction algorithm based on this algorithm,and uses the DTI atlas construction algorithm to construct a DTI atlas to study the effects of drug addiction on brain organization structure.The specific work is as follows:(1)A cascade diffeomorphic network for DTI registration(CDnet-DTIR)is proposed,which is capable of fast and accurate DTI registration in 3D human brain.The model is mainly composed of two registration sub-networks: fractional anisotropy(FA)parametric map registration network and mean diffusivity(MD)parametric map registration network.In order to ensure that the fiber orientation and the tissue structure can be consistent after the DTI spatial transformation,the logarithmic Euclidean Riemann interpolation layer and the reorientation layer are designed in the registration model,both of which can ensure the gradient back propagation during the network training.In addition,in order to improve the DTI registration accuracy,this paper uses various macro similarity measures and micro similarity measures as loss functions to guide the network training.The experimental results show that the CDnet-DTIR model can achieve global anatomical structure and fiber direction alignment of DTI with high registration realtime,which meets the demand of clinical real-time registration.(2)Considering DTI atlas construction as a probabilistic estimation model,a convolutional neural network-based DTI atlas construction framework is proposed on the basis of the CDnet-DTIR model.The framework not only generates DTI atlas with global variability,but also aligns diffusion tensor images with the atlas quickly and precisely.Compared with the traditional DTI atlas construction algorithm,the proposed framework is more efficient and the constructed atlas is more central.In addition,the DTI atlas was created using laboratory data collected from healthy and drug-addicted individuals within the framework of the DTI atlas construction proposed in this paper,and the whole brain voxel-based analysis(VBA)method and tract-based spatial statistic(TBSS)method were applied to study the effects of drug addiction on brain tissue structure based on the FA parameter maps of the atlas.The results showed that there were significant differences between drug addicts and healthy subjects in the occipital forceps,corpus callosum,and frontooccipital fasciculus.