Research On Segmentation And Reconstruction Algorithms Of Cell Bodies And Blood Vessels In Images Of Mouse Brain Regions Based On 3D Neural Networks

The brain,as one of the most important organs in the human body,dominates the activities of human thought,language,emotions,and memory.The most basic constituent unit in the brain is neurons,and the distribution patterns of different types of neurons in the brain are also different.Quantitative identification and analysis of neurons are important means for studying the distribution of neurons and correctly understanding the working mechanism of the brain.In addition,relevant medical studies have shown that abnormalities in cerebral blood vessels are closely related to the severity of cerebrovascular diseases.In some cases,lesions in the cerebral blood vessels may have appeared before these diseases occur.Therefore,the recognition and reconstruction of whole cerebral vascular images are of great significance for predicting,diagnosing,and treating cardiovascular diseases.However,there is still room for further research on the segmentation and reconstruction of blood vessels and cell bodies in the Acridine Orange(AO)stained images of mouse brain regions based on fluorescence microscopy optical tomography technology,as well as the recognition and counting of cell bodies in the neuronal distribution map of transgenic hybrid mice.On the one hand,it is because there are multiple complex structures present in AO stained fluorescence images of the whole brain,which makes the already challenging fluorescence image processing even more difficult.However,deep learning,with its powerful ability to automatically extract features and its adaptability to complex data,provides a new breakthrough for solving the aforementioned problems.In summary,this article has done the following work on the segmentation and recognition of blood vessels and cell bodies,as well as impurity removal in mouse brain region images:(1)Because there are multiple complex structures present in AO stained fluorescence images of the whole brain,which makes the already challenging fluorescence image processing even more difficult.Therefore,this article constructs a3 D multi-scale and multi-channel neural network convolution module based on the morphological features of cell bodies and blood vessels in mouse brain regions stained with AO fluorescence,and verifies the effectiveness of this module through ablation experiments.And based on this multi-scale and multi-channel 3D convolution module,a U-shaped network for vessel and cell segmentation was constructed.(2)By constructing a multi-scale 3D neural network,an automated cell and vascular reconstruction method was designed,and its effectiveness was verified through experiments.For cell segmentation,the Dice value was 0.9081,F1 score was0.8979,vascular Dice value was 0.9658,and F1 score was 0.9418.It can recognize and reconstruct the three-dimensional morphological structure of cell bodies and blood vessels in complex backgrounds.This method can accurately obtain relevant information about blood vessels and cell bodies in mouse brain regions,which helps biologists obtain the morphological characteristics of cell bodies and the connection information of blood vessels.(3)There is interference from some non-target cells in the neuronal distribution map data of transgenic hybrid mice.In neuronal counting,we collectively refer to these non-target cells as impurities.These impurities have a significant impact on neuronal cell body counting across the entire brain,resulting in significant deviation in cell body counting results.For the distribution map data of mouse brain regions,this article constructs a set of fast methods for removing impurities from the distribution map and counting cell bodies based on an improved 3D neural network.The impurity removal rate of this method reached 0.9443,which can effectively remove impurities from the distribution map data.