Elaborate Structure Analysis Of Brain Neurons In Alzheimer’s Disease Mice And Establishment Of Soma Automatic Tracking Model

Alzheimer’s disease(AD),as a neurodegenerative disease,is the main cause of dementia in the elderly.Its development process is closely related to changes in the shape and number of neurons.Micro-Optical Sectioning Tomography(MOST)is a whole brain scanning stereo imaging technology developed in recent years,which can provide high-resolution,high-definition neuronal images.However,due to the acquisition of a large amount of data,the speed of bioinformatics analysis and processing is far from meeting the requirements.The morphology of a large number of brain neurons and their relationship with AD have not been mined and used.The purpose of this study is to analyze the morphological characteristics of AD mouse brain neurons,construct an automatic tracking network model of AD neuronal soma,and analyze the changes in neurons during AD development.In this paper,MOST technology was used to perform a whole-brain scan on eight triple transgenic AD mice(3 × Tg-AD)and wild-type control(WT)mice to obtain 8 TB brain coronal image data of each mouse.Based on these data,we selected 4 brain regions that are closely related to the pathological features of AD,including hippocampal CA3,medial entorhinal cortex,lateral entorhinal cortex,and presubiculum region.These data were used for three-dimentional reconstruction to analyze the complexity of neuronal morphology.Comprehensive analysis methods were used to analyze neuron morphology,including basic index analysis,Sholl analysis,artificial soma number analysis,multi-index analysis.In view of the fact that manually tracking neurons faces a large number of data processing inefficiencies,time-consuming and difficult to analyzing whole brain data,convolutional neural network technology was applied in this study to build neuron automatic tracking network models,including soma automatic tracking network models and neuron fiber automatic tracking network model.The main procedures of soma automatic tracking network model is as follows.Firstly,some parts of the image data in the hippocampus and entorhinal cortex were cropped to perform data enhancement and manual annotation.Three types of database were constructed,including soma,blank and fiber.Secondly,neuronal soma automatic tracking network model were built and trained with database.Finally,the established network model was used to calculate the number of neuron soma in different brain regions.The results were compared with those of manual analysis of soma number and Neuron GPS analysis of soma number in order to estimate the accuracy of the convolutional neural network soma automatic tracking model in this paper.We applied the model to count soma number in different brain regions including hippocampal CA1,medial entorhinal cortex,lateral entorhinal cortex,and presubiculum region.Using the above method,the morphological changes of neurons were analyzed in different brain regions from the MOST data of whole brain from AD and WT mice.It was found that during AD progression,the morphological complexity of neuronal dendrites varied in different brain regions.The dendritic complexity of AD mice did not change significantly in the presubiculum region.However,it decreased significantly in the lateral entorhinal cortex,increased significantly in the medial entorhinal cortex,and decreased significantly in the hippocampal CA3 region.Using the multi-index analysis,a difference was found in the value of total area and the mean value of terminal fiber length in the lateral entorhinal cortex between AD and WT mice.In hippocampal CA3,the number of bifurcation points,average fiber length,average non-terminal fiber length,average non-terminal fiber length / total area value showed differences between AD and WT.It shown how neurons change in different brain regions.Next,we built neuron database and convolutional neural network soma automatic tracking model for soma positioning and counting.Comparing the results of calculating the number of soma by the three methods,we found that the number of soma counted by the model was more accurate and close to the result of manual soma number analysis(p>0.05).The model was applied to count the number of soma in different brain regions.We found that the number of neurons in AD mice decreased significantly in hippocampal CA1(p=0.02).In the lateral entorhinal cortex,medial entorhinal cortex,and presubiculum region,no significant difference was found in the number of soma between AD and WT.Summarily,this study shows that neuron complexity in different brain regions varied with the progression of AD pathology.The convolutional neural network soma automatic tracking model we built can better counts the number of soma in the brain region with higher accuracy and speed.It provides research methods and basic data for our next work of automatic tracking neuronal fibers and automatic analysis of neurons.