Research On Regional Localization And Neuron Counting System Of Mouse Brain Slice Based On Unified Modal Transformation

Brain science research often requires accurate detection,localization,and quantitative analysis of the number of cells,molecular expressions,and neuronal activities in different brain regions of experimental animals.The premise of related analysis is to determine the brain region of each site on the brain slice image by referring to the standard Allen Brain Atlas,namely the regional localization of the brain slice.Then,the number of neurons in different brain areas is detected according to the brain location results,to analyze that reflect the dynamic activity of the experimental animals during the specific training process.Due to the different sizes of individual animals and the artificial introduction of distortion and deformation in the preparation and imaging process of the brain slice image,the standard Allen Brain Atlas image cannot be directly fused with the animal brain slice image to complete the regional positioning.At present,the brain slice regional positioning and Neuron counting still require a lot of manpower and time.For this reason,the study of an automated system that can quickly and accurately locate brain slice regions and count neurons has important value and significance for brain science research.In the face of the above problems,this article uses modal transformation technology,image registration technology,and target detection technology in computer vision to complete the automatic processing tasks of mouse brain slice image region positioning and neuron counting.The main research contents are as follows:(1)Because of the large modal difference between the mouse brain slice image and the standard brain atlas image,it is difficult to use the conventional multi-modal image similarity measure,a modal transformation algorithm based on the Joint Enhancement of Multimodal Information is proposed.First,perform data enhancement on the input image to improve the generalization ability and robustness of the network model,and then combine the information of the input image,make full use of the consistency of the key structural features of the image and the context information,and learn the feature map of the input image through the coding neural network.Extract the common features of mouse brain slices and standard brain atlas images,then reverse decoding to classify different features,and finally set appropriate image gain coefficients to map the features to two unified modal images for subsequent image registration.A good foundation is laid for better results.(2)Aiming at the problems of distortion introduced artificially in the preparation of mouse brain slice images,an image registration algorithm with diffeomorphic is proposed.First,the mouse brain slice image is used as a reference image,and the standard Allen Brain Atlas image is used as a floating image,which is jointly sent to the network as an input image.Then,the convolutional neural network is used to predict the mean and covariance variables,and the reparameterization technique is used to convert them to diffeomorphic deformation field,and then deform the floating image,the standard brain atlas is accurately mapped to the mouse brain slice image to complete the regional localization of the mouse brain slice image.(3)Aiming at the problems of missed detection and false detection and slow running speed of labeled neurons on mouse brain slice images when using traditional detection algorithms,a YOLOTB algorithm is proposed.First,the training image is cut into blocks and the mosaic method is used for data enhancement,and then uses the CSP structure and Mish activation function in the Backbone part,the SPP structure and FPN+PAN structure in the Neck part,and the GIOU?Loss loss function in the Prediction part to effectively improve the target detection effect.Finally,accurate and fast neuron detection is realized.Furthermore,the human-computer interaction interface is added to alleviate the problems of missed detection and false detection and improves the accuracy of detection results.