Study On Compression Strategy Of Whole Brain Threedimensional Image Data For Neuron Morphology Reconstruction

Neuron reconstruction refers to the digitization of neuron morphology and structure through specific labeling technique,imaging technology,and specific tracking algorithms or tools.The digitized neuron information shows the distribution and specific morphological structure of neurons in the brain,and is used in the research of neuron classification,neuron projection,and neural circuits.It is necessary to perform sub-micron imaging of model animals such as mouse in the whole brain for neuron reconstruction.The amount of whole brain data used for neuron reconstruction is as high as 10 TB due to the extremely high voxel resolution and large sample size,which brings huge storage and transmission pressure to neuron reconstruction.The simplest,safe and effective mitigation method is to compress the data used for neuron reconstruction.Whole brain data is essentially a massive amount of three-dimensional image data,but the existing compression methods for threedimensional image data have the following shortcomings: one is to save a large amount of data without neuron signal,and the other is that it does not use data features to adaptively select and compress parameter.Therefore,these methods generally have the problem of low compression rate.For the problem of saving a large amount of data without neuron signal.First,the location information of the block data in the mouse brain contour is obtained by segmentation of the mouse brain contour,and then the detection and classification are carried out according to whether there are neuron signals in the block data,and finally the data is cleaned according to the segmentation and classification results.The proposed fast contour segmentation method can complete the fast segmentation of down-sampled data within 1 to 3 minutes,and the accuracy,recall and F1-measure of the segmentation results can reach more than 0.96.The classification network designed based on Dense Net-169 can use the projection images of the three directions of the data block to classify whether there is a neuron signal.The recall of the classification results is above 0.93,and most of the block data with neuron signals can be detected.For the problem that the compression parameters are not adaptively selected according to the data characteristics,an adaptive compression test scheme is proposed.This scheme can perform compression tests of a variety of video encodings in multiple modes.It can not only select the best video encoding for different whole brain data used for neuron reconstruction,but also can be used to optimize fixed compression parameters and adaptive compression parameters.Based on the above study,a compression strategy for the whole brain data for neuron reconstruction is constructed.The whole brain data can be compressed within 1 to 1.5 hours through this compression strategy.The average compression ratio has reached 32.06,and the compressed file size is about three percent of the uncompressed data,when the constant rate factor is 25.The final compressed file size is only one-thousandth to four-thousandths of the uncompressed data after clearing the invalid data.In summary,the whole brain data compression strategy proposed in this study can quickly complete the compression of the whole brain data,and can retain most of the neuron signals while ensuring a high compression rate.It can effectively alleviate the transmission and storage pressure caused by massive data,and improve the efficiency of neuron reconstruction.