Data cleaning method suitable for massive neuron three-dimensional imaging data
By combining segmentation and classification models with attention and gating mechanisms, a system can automatically clean massive amounts of 3D neuron imaging data. This solves the problems of high computational resource consumption and unstable classification results in the processing of massive neuron data, achieving efficient and accurate data cleaning and rapid manual review.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HUST SUZHOU INST FOR BRAINMATICS
- Filing Date
- 2025-02-26
- Publication Date
- 2026-07-24
AI Technical Summary
Existing neuronal data cleaning methods require a large amount of computational and spatial resources when processing massive amounts of neuronal data, especially macaque neuronal data. They are difficult to obtain stable classification results, and manual review is difficult and inefficient.
Attention and gating mechanisms are used to enhance neuronal signals. Segmentation and classification models are combined to automatically clean massive neuronal 3D imaging data. Invalid data is removed and projection blocks are divided by maximum projection map. Cleaning tools are used for re-examination to ensure the stability and accuracy of classification results.
It effectively reduces data processing time, improves the stability and accuracy of neuronal data cleaning, saves computing and storage resources, achieves an efficient data cleaning process, and ensures high recall rate and rapid manual review of classification results.
Smart Images
Figure CN120147529B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of medical image processing, and in particular to a data cleaning method applicable to massive neuronal three-dimensional imaging data. Background Technology
[0002] Neurons, or neuronal cells, are the most basic structural and functional units of the nervous system, playing a crucial role in the study of fundamental brain functions. Through complex network connections, neurons support numerous higher functions such as cognition, emotion, and behavior. The diameter of a neuronal cell body ranges from 10μm to 30μm, and the processes originating from the cell body can reach submicron diameters. To obtain clearer and more complete neuronal morphological information, submicron-level imaging of the entire brain is necessary; however, high-resolution imaging also involves a massive amount of imaging data, with the imaging data of the entire macaque brain reaching hundreds of TB. For sparsely labeled macaque brains, neuronal signals account for less than 5%, and the large amount of redundant data presents challenges in data storage, transmission, and computation. Furthermore, traditional biomedical image processing software such as Amira and Fiji cannot efficiently perform manual review of neuronal data cleaning tasks, requiring significant human resources. Therefore, performing neuronal data cleaning on sparsely labeled macaque brains to extract data blocks containing neurons is crucial for subsequent rapid analysis of neuronal signals.
[0003] Existing neuronal data cleaning methods, such as the invention patent with publication number CN117218144A, disclose a data cleaning method based on neuronal signal recognition. This method reduces manual time consumption, generates fewer intermediate files to achieve good cleaning results, significantly reduces data volume, and effectively avoids unnecessary waste of computational and storage resources. This data cleaning method first acquires sparse labeled data and cellular architecture data of the brain. Then, it divides the sparse labeled data into several data blocks of the same size. Maximum intensity projection is performed on the data blocks within the brain contour, outputting a maximum intensity projection map. Adaptive threshold segmentation and other methods are then used to perform binary classification on all maximum value projections. Finally, manual observation is used to check and fill in any gaps in the binary classification results.
[0004] This data cleaning method is suitable for processing relatively small amounts of data, such as mouse neuron data. However, when processing massive amounts of neuron data (such as macaque neuron data), it still faces the following problems: First, it requires pre-segmentation of the whole brain, which consumes a lot of computing and space resources. Second, when applied to the processing of large and complex data, the above classification method is difficult to obtain stable results on different individuals. Third, the imaging data blocks of macaques are as many as millions of blocks, making it difficult to fill in the gaps in the classification results through manual observation. Summary of the Invention
[0005] Therefore, to solve the above problems, the present invention provides a data cleaning method suitable for massive neuronal three-dimensional imaging data.
[0006] This invention is achieved through the following technical solution:
[0007] A data cleaning method suitable for massive neuronal 3D imaging data includes the following steps:
[0008] Acquire two-dimensional brain image sequence data;
[0009] The maximum projection map of the two-dimensional brain image sequence data is obtained. Invalid data located outside the brain contour in each maximum projection map is removed, and the data within the brain contour is divided to obtain projection blocks. The projection blocks are named according to their positions in the current maximum projection map.
[0010] By combining attention and gating mechanisms, the neuronal signals of the projection blocks within the maximum projection map are enhanced while the non-neuronal signals are weakened, so as to highlight the distribution location of the neuronal signals. The projection blocks containing neuronal signals and those without neuronal signals are classified into different projection block types, and the classification results are obtained.
[0011] Based on the classification results, the label information of all projection blocks containing neurons is derived, and the label information includes at least the name of the projection block, generating a cleaned label information set;
[0012] Extract each projection block containing a neuron from the label information set, and then segment and reassemble all projection blocks containing neurons to obtain a three-dimensional data block of the target shape.
[0013] Preferably, after obtaining the classification results, the classification results are reviewed, the incorrectly classified projection blocks are reclassified, and the final classification results are obtained.
[0014] Preferably, the step of "obtaining the maximum value projection map of the two-dimensional brain image sequence data, removing invalid data located outside the brain contour in the maximum value projection map, and dividing the data within the brain contour into blocks" includes the following steps:
[0015] By using sparse labeling with viruses, the brightness of neurons in the imaging data is made greater than that of the background information;
[0016] Using a preset number of image layers as a unit, the maximum value projection is performed sequentially on the two-dimensional brain image sequence data of the whole brain to obtain several maximum value projection maps;
[0017] Each maximum projection image is downsampled sequentially, and then upsampled to the original resolution after brain contour segmentation to obtain the original resolution segmentation image of each maximum projection image;
[0018] The brain contour region of each original resolution segmentation map is divided into multiple projection blocks.
[0019] Preferably, before classifying the projection block types, a segmentation model and a classification model are obtained; wherein, both the segmentation model and the classification model are trained using multiple projection block data containing neurons and multiple projection block data without neurons. The segmentation model is used to enhance the neuron signals of the projection blocks within the maximum projection map and weaken the non-neuron signals to highlight the distribution location of the neuron signals, and the classification model is used to classify the projection blocks containing neuron signals and the projection blocks for which no neuron signals have been extracted.
[0020] Preferably, the segmentation model is the Attention Unet model. The Attention Unet model is trained using the projection block dataset, and the neuron signal is used as the target feature. Each projection block in the projection map of the maximum value to be detected is input into the trained Attention Unet model. The attention mechanism in the model increases the attention of the neuron signal, thereby realizing the extraction of the neuron signal in the projection map of the current maximum value.
[0021] Preferably, the classification model is a ResNet-18 convolutional neural network. The ResNet-18 convolutional neural network is trained using the projection block dataset. Each projection block in the projection map of the maximum value to be detected is input into the trained ResNet-18 convolutional neural network. The neuron signal is used as the classification feature to classify each projection block and the classification result is output.
[0022] Preferably, the phrase "segmenting and reassembling all projection blocks containing neurons to obtain a three-dimensional data block of the target shape" refers to using multi-threaded processing, dividing each projection block containing neurons into several small data blocks, creating a thread for each small data block, and having each thread read its corresponding small data block and write it to the correct position in the merged file, so that the several small data blocks are reassembled to obtain a three-dimensional data block of the target shape.
[0023] Preferably, the small data block is obtained by dividing the projection block by its thickness, and both the length and width are equal.
[0024] Preferably, the "re-inspection of the classification results" is achieved using a cleaning tool, which includes:
[0025] The input module includes a projection block data input port, a label information input port, and a downsampled image input port. The projection block data input port is used to input projection block data located within the brain contour in the maximum projection map, including projection blocks containing neurons and projection blocks without neurons. The label information input port is used to input the cleaned label information set. The downsampled image input port is used to input a 50μm downsampled image of the maximum projection map.
[0026] The display module is used to display the projection map of the maximum value being operated on and the selected area in the current maximum value projection map. The visualization range of the selected area can be adjusted by pressing the buttons.
[0027] The adjustment module is used to display a magnified view of the selected area and all the projection blocks within the magnified view. The incorrectly classified projection blocks can be reclassified by the mouse to obtain the final classification result after re-inspection.
[0028] The export module is used to export the final classification result as a txt file, which contains the names of all neuron projection blocks.
[0029] The beneficial effects of the technical solution of this invention are mainly reflected in:
[0030] 1. This method first obtains the maximum projection map of two-dimensional brain image sequence data, and after data preprocessing to remove invalid data and automatic data cleaning, it distinguishes the projection blocks containing neurons. It only needs to perform three-dimensional segmentation and stitching on the projection block data containing neurons, saving a lot of data processing time. When cleaning neuron data, it adopts a combination of segmentation and classification to effectively suppress the influence of noise. It is convenient to complete the cleaning of neuron data of the whole brain of macaques in a very short time while ensuring the stability and accuracy of classification results.
[0031] 2. Before cleaning the neuron data, the segmentation model and classification model are trained with large-scale data to improve the robustness of the model. The segmentation model combines attention and gating mechanisms to enhance the neuron signals of the projection blocks in the maximum projection map and weaken the non-neuron signals, so as to highlight the distribution location of the neuron signals, which facilitates the classification model to classify quickly and improves work efficiency.
[0032] 3. This method provides a data cleaning tool. The tool allows users to adjust the range of the visualization area using buttons and add projection blocks that the user judges to contain neurons or remove projection blocks that do not contain neurons by clicking with the mouse. This allows misclassified projection blocks to be reclassified, facilitating rapid manual review, ensuring a high recall rate for the data cleaning results, and finally exporting the final classification results in txt format. Attached Figure Description
[0033] Figure 1 This is a flowchart of a data cleaning method applicable to massive neuronal 3D imaging data;
[0034] Figure 2 This is a schematic diagram of a data cleaning method applicable to massive neuronal 3D imaging data (including a re-inspection step);
[0035] Figure 3 This is a schematic diagram of the data preprocessing steps (including maximum projection, removal of data outside the brain contour, and segmentation of data inside the brain contour) in a data cleaning method applicable to massive neuronal three-dimensional imaging data;
[0036] Figure 4 This is a schematic diagram of the segmentation and classification steps (automatic cleaning) in a data cleaning method applicable to massive neuronal 3D imaging data;
[0037] Figure 5 This is a schematic diagram of the cleaning tool interface;
[0038] Figure 6 This is a schematic diagram of the segmentation and reassembly steps in a data cleaning method applicable to massive neuronal 3D imaging data. Detailed Implementation
[0039] To make the objectives, advantages, and features of the present invention clearer and more detailed, the following non-limiting description of preferred embodiments will be illustrated and explained. These embodiments are merely typical examples of applying the technical solutions of the present invention; all technical solutions formed by equivalent substitutions or equivalent transformations fall within the scope of protection claimed by the present invention. It should also be noted that in the description of the solutions, the terms "multiple" or "several" in the present invention mean two or more, and unless otherwise explicitly defined, should not be construed as limiting the present invention.
[0040] This invention discloses a data cleaning method suitable for massive neuronal three-dimensional imaging data, such as... Figure 1 , Figure 2 As shown, the process, starting from the original two-dimensional sequence data and ending with the acquisition of all three-dimensional data blocks containing neurons, specifically includes the following steps:
[0041] Acquire two-dimensional brain image sequence data;
[0042] The maximum projection map of the two-dimensional brain image sequence data is obtained. Invalid data located outside the brain contour in each maximum projection map is removed, and the data within the brain contour is divided to obtain projection blocks. The projection blocks are named according to their positions in the current maximum projection map.
[0043] In some embodiments, this step specifically includes:
[0044] By using sparse viral labeling, the brightness of neurons in the imaging data is made greater than the background information. Assuming the acquired monkey brain 2D image sequence data is 100000*80000*20000 (i.e., 20000 images of 100000*80000 pixels each), the imaging resolution is 0.65*0.65*3μm. With sparse viral labeling, the brightness of neurons in the imaging data is typically greater than the background information; therefore, the morphological information of neurons can be significantly observed through maximum projection.
[0045] Using a preset number of image layers as a unit, the maximum value projection is performed sequentially on the two-dimensional brain image sequence data of the whole brain to obtain several maximum value projection maps; in one embodiment, the maximum value projection is performed on the whole brain of the macaque monkey using 360 layers as a unit, and finally 56 maximum value projection maps are obtained.
[0046] Each maximum projection image is downsampled sequentially, and then upsampled to the original resolution after brain contour segmentation to obtain the original resolution segmentation image of each maximum projection image. Specifically, in the maximum projection image, some tissue debris with autofluorescence properties remains outside the brain contour. This debris will affect the accuracy of neuron recognition. Therefore, it is necessary to remove invalid data outside the brain contour before segmenting the maximum projection image. The brain contour segmentation and data removal outside the brain contour both adopt existing segmentation and removal methods, which will not be elaborated here. Specifically, the projection image at the original resolution is first downsampled to 50μm, and the 50μm image is segmented and then upsampled to the original resolution. Downsampling reduces the image complexity by reducing the resolution or size of the image. Its purpose is to reduce the number of pixels in the image, thereby reducing the impact of noise on the results, reducing the amount of computation, and increasing the receptive field size of the convolution kernel.
[0047] The brain contour region of each original resolution segmentation image is divided into multiple projection blocks. In a preferred embodiment, the projection block is a small square block with uniform thickness and side length. In conjunction with the above embodiment, the thickness and side length of the projection block are both 360 pixels.
[0048] Specifically, "naming the projection block according to its position in the current maximum value projection map" means that after dividing the projection blocks, each projection block is named, the position of the current projection block in the current maximum value projection map is located by the name of the projection block, and label information for each projection block is generated and the name of the projection block is stored in the corresponding label information.
[0049] In the above embodiment, the thickness and side length of the projection block are both 360 pixels. Therefore, its name can display its coordinate information. For example, a data block named "59_60_27" has the coordinates of its upper left corner [59*360, 60*360, 27*360] and its lower right corner [27*360].
[0050] [60*360, 61*360, 28*360], the location of the data block can be accurately located based on the coordinates of the upper left and lower right corners of the data block.
[0051] After obtaining the label information of each projection block, the neuronal signals of the projection blocks in the maximum projection map are enhanced and the non-neuronal signals are weakened by combining the attention mechanism and the gating mechanism, so as to highlight the distribution position of the neuronal signals. The projection blocks containing neuronal signals and the projection blocks without neuronal signals are classified into different projection block types and the classification results are obtained.
[0052] In some embodiments, before classifying the projection patch types, a segmentation model and a classification model are obtained; wherein the segmentation model and the classification model are both trained using multiple projection patch data containing neurons and multiple projection patch data without neurons. In a preferred embodiment, before training the model, a projection patch dataset consisting of 8,000 projection patches containing neurons and 100,000 projection patches without neurons is first constructed, and the segmentation model and the classification model are trained using the projection patch dataset, thereby effectively improving the robustness of the model.
[0053] In this embodiment, the segmentation model described above is used to enhance the neuronal signals of the projection blocks within the maximum value projection map and weaken the non-neuronal signals to highlight the distribution location of the neuronal signals. In a preferred embodiment, the segmentation model is the Attention Unet model. The Attention Unet model introduces an attention mechanism on the basis of the traditional Unet architecture, which can improve the performance of image segmentation tasks. By dynamically focusing on and selecting image regions of interest, it can improve the accuracy and precision of segmentation and effectively suppress incoherent regions. The Attention Unet model is trained using the projection block dataset, and the neuronal signals are used as target features. Each projection block in the maximum value projection map to be detected is input into the trained Attention Unet model. Through the attention mechanism in the model, different convolution operations are used to generate and adjust the attention coefficients, thereby focusing on the neuronal signals and realizing the extraction of neuronal signals within the current maximum value projection map.
[0054] In addition, a classification model is used to classify projection blocks containing neuron signals from projection blocks without extracted neuron signals. In a preferred embodiment, the classification model is a ResNet-18 convolutional neural network. The ResNet-18 convolutional neural network is trained using the projection block dataset. Each projection block in the maximum value projection map to be detected is input into the trained ResNet-18 convolutional neural network. The neuron signals are used as classification features to classify each projection block, and the classification result is output.
[0055] In some embodiments, after obtaining the classification result, the classification result is reviewed, the incorrectly classified projection blocks are reclassified, and the final classification result is obtained; and after obtaining the classification result, the label information of all projection blocks containing neurons is derived according to the classification result to generate a cleaned label information set.
[0056] In a preferred embodiment of the present invention, the "re-inspection of the classification results" is achieved by a cleaning tool, which includes:
[0057] The input module includes a projection block data input port, a label information input port, and a downsampled image input port. The projection block data input port is used to input projection block data located within the brain contour in the maximum projection map, including projection blocks containing neurons and projection blocks without neurons. The label information input port is used to input the cleaned label information set. The downsampled image input port is used to input a 50μm downsampled image of the maximum projection map.
[0058] The display module is used to display the projection map of the maximum value being operated on and the selected area in the current maximum value projection map. The visualization range of the selected area can be adjusted by pressing the buttons.
[0059] The adjustment module is used to display a magnified view of the selected area and all the projection blocks within the magnified view. The incorrectly classified projection blocks can be reclassified by the mouse to obtain the final classification result after re-inspection.
[0060] The export module is used to export the final classification result as a txt file, which contains the names of all neuron projection blocks. This cleaning tool enables rapid semi-manual re-examination, thereby improving the efficiency of manual re-examination.
[0061] After exporting the final classification results, each projection block containing a neuron is extracted from the label information set, and all projection blocks containing neurons are segmented and reassembled to obtain a three-dimensional data block of the target shape.
[0062] Specifically, the label information of each projection block containing neurons is extracted from the label information set, and these projection blocks are located by name. Then, these projection blocks containing neurons are segmented and reassembled. In a preferred embodiment, "segmenting and reassembling all projection blocks containing neurons to obtain a three-dimensional data block of the target shape" refers to using multi-threaded processing. Each projection block containing neurons is divided into several small data blocks. A thread is created for each small data block, and each thread reads its corresponding small data block and writes it to the correct position in the merged file. This allows the several small data blocks to be reassembled to obtain the three-dimensional data block of the target shape. According to the block, the small data block is obtained by dividing the thickness of the projection block, and both the length and width are equal. In conjunction with the above embodiment, when the thickness and side length of each projection block are 360 pixels, each projection block is first cut into small data blocks with a thickness of about 10 pixels and other side lengths equal to the size of the projection block (i.e., the length and width are both 360 pixels). Then, multiple small data blocks are spliced into a large data block of the target shape through multi-threading. The target shape of the large data block can be set according to the requirements, which will not be elaborated here. This process only cuts the data blocks containing neurons in the original data, which greatly saves computing and storage resources compared with the existing technical process.
[0063] To verify the effectiveness of the data cleaning method disclosed in this invention for massive neuronal three-dimensional imaging data, whole-brain images of five macaques were provided, and the following experiments were conducted:
[0064] Experiment 1: Projection blocks were prepared from a portion of the whole brains (Brain1-Brain5) of 5 macaques. Data cleaning methods suitable for massive neuronal 3D imaging data were used to classify the projection blocks using segmentation and classification models. The data cleaning test results are shown in Table 1.
[0065] Table 1: Quantitative evaluation of the cleaning results of samples Brain1-Brain5;
[0066]
[0067]
[0068] The test results in Table 1 show that the automatic cleaning model of this invention has high recall and precision on different individuals, which indicates that the segmentation model and classification model used in this invention have high robustness.
[0069] Experiment 2: Whole-brain two-dimensional sequence data from 5 macaques were classified using a data cleaning method suitable for massive neuronal three-dimensional imaging data, and then re-examined using a cleaning tool. The cleaning results are shown in Table 2:
[0070] Table 2: Results of whole-brain data cleaning after manual re-examination;
[0071]
[0072] As shown in Table 2 above, data within the brain contour accounts for an average of about 30% of the entire brain. Therefore, removing the brain contour can greatly improve the efficiency of data cleaning. Meanwhile, neuron data blocks typically account for less than 1% of the entire brain. This indicates that neuron data cleaning can significantly reduce content unrelated to neurons in the original data, providing a more efficient data foundation for subsequent rapid neuron analysis.
[0073] Experiment 3: The data cleaning method for massive neuronal 3D imaging data disclosed in this invention and existing data cleaning methods were used to clean the whole brain neuronal data of a macaque. In this experiment, as... Figure 2 As shown, in the rapid data cleaning method of the present invention, data preprocessing includes: maximum value projection, removal of data outside the brain contour, and segmentation of data within the brain contour; the automatic cleaning refers to classifying the projection blocks using segmentation and classification models; simultaneously, data cleaning tools are used for review; existing data cleaning methods directly segment the whole brain data into blocks, perform maximum value projection on the data within the brain contour, and use a threshold segmentation method to perform binary classification on all maximum value projections, and manually observe the binary classification results to check for omissions; the cleaning results of each method are as follows:
[0074] Table 3: Data cleaning time consumption of a single rhesus monkey whole brain neuron in this invention;
[0075]
[0076] Table 4: Data cleaning time for single rhesus monkey whole brain neurons using traditional methods;
[0077]
[0078] Based on the above cleaning results, it can be seen that the data cleaning method for massive neuronal three-dimensional imaging data disclosed in this invention requires only about 20 hours of manual intervention, and the total time to clean one macaque is only about 5 days. In contrast, the manual intervention time (i.e., the time spent on re-examination) and the total cleaning time of existing data cleaning methods are many times that of the method of this invention. This shows that this invention can greatly improve the efficiency of neuronal data cleaning in the whole brain of macaques and effectively reduce the degree of manual intervention and improve the degree of automation.
[0079] This invention has many other embodiments, and all technical solutions formed by equivalent transformation or equivalent transformation fall within the protection scope of this invention.
Claims
1. A data cleaning method applicable to massive neuronal three-dimensional imaging data, characterized in that: Includes the following steps: Acquire two-dimensional brain image sequence data; The maximum projection map of the two-dimensional brain image sequence data is obtained. Invalid data located outside the brain contour in each maximum projection map is removed, and the data within the brain contour is divided to obtain projection blocks. The projection blocks are named according to their positions in the current maximum projection map. By combining attention and gating mechanisms, the neuronal signals of the projection blocks within the maximum projection map are enhanced while the non-neuronal signals are weakened, so as to highlight the distribution location of the neuronal signals. The projection blocks containing neuronal signals and those without neuronal signals are classified into different projection block types, and the classification results are obtained. Based on the classification results, the label information of all projection blocks containing neurons is derived, and the label information includes at least the name of the projection block, generating a cleaned label information set; Extract each projection block containing a neuron from the label information set, and then segment and reassemble all the projection blocks containing neurons to obtain a three-dimensional data block of the target shape. The process of "obtaining the maximum value projection map of two-dimensional brain image sequence data, removing invalid data located outside the brain contour in each maximum value projection map, and dividing the data within the brain contour to obtain projection blocks" includes the following steps: By using sparse labeling with viruses, the brightness of neurons in the imaging data is made greater than that of the background information; Using a preset number of image layers as a unit, the maximum value projection is performed sequentially on the two-dimensional brain image sequence data of the whole brain to obtain several maximum value projection maps; Each maximum projection image is downsampled sequentially, and then upsampled to the original resolution after brain contour segmentation to obtain the original resolution segmentation image of each maximum projection image; The brain contour region of each original resolution segmentation map is divided into multiple projection blocks.
2. The data cleaning method for massive neuronal three-dimensional imaging data according to claim 1, characterized in that: After obtaining the classification results, the classification results are reviewed, the incorrectly classified projection blocks are reclassified, and the final classification results are obtained.
3. The data cleaning method for massive neuronal three-dimensional imaging data according to claim 1, characterized in that: Before classifying the projection block types, a segmentation model and a classification model are obtained. The segmentation model and the classification model are trained using multiple projection block data containing neurons and multiple projection block data without neurons. The segmentation model enhances the neuron signals of the projection blocks within the maximum projection map and weakens the non-neuron signals to highlight the distribution location of the neuron signals. The classification model is used to classify the projection blocks containing neuron signals and the projection blocks without extracted neuron signals.
4. The data cleaning method for massive neuronal three-dimensional imaging data according to claim 3, characterized in that: The segmentation model is the Attention Unet model. The Attention Unet model is trained using the projection block dataset, and the neuron signal is used as the target feature. Each projection block in the projection map of the maximum value to be detected is input into the trained Attention Unet model. The attention mechanism in the model increases the attention of the neuron signal, thereby realizing the extraction of the neuron signal in the projection map of the current maximum value.
5. The data cleaning method for massive neuronal three-dimensional imaging data according to claim 3, characterized in that: The classification model is a ResNet-18 convolutional neural network. The ResNet-18 convolutional neural network is trained using the projection block dataset. Each projection block in the projection map of the maximum value to be detected is input into the trained ResNet-18 convolutional neural network. The neuron signal is used as the classification feature to classify each projection block and the classification result is output.
6. The data cleaning method for massive neuronal three-dimensional imaging data according to claim 1, characterized in that: The phrase "segmenting and reassembling all projection blocks containing neurons to obtain a three-dimensional data block of the target shape" refers to using multi-threaded processing. Each projection block containing neurons is divided into several small data blocks. A thread is created for each small data block. Each thread reads its corresponding small data block and writes it to the correct position in the merged file, so that the several small data blocks are reassembled to obtain a three-dimensional data block of the target shape.
7. The data cleaning method for massive neuronal three-dimensional imaging data according to claim 6, characterized in that: The small data block is obtained by dividing the thickness of the projection block, and both have the same length and width.
8. The data cleaning method for massive neuronal three-dimensional imaging data according to claim 2, characterized in that: The "re-inspection of classification results" is achieved through a cleaning tool, which includes: The input module includes a projection block data input port, a label information input port, and a downsampled image input port. The projection block data input port is used to input projection block data located within the brain contour in the maximum projection map, including projection blocks containing neurons and projection blocks without neurons. The label information input port is used to input the cleaned label information set. The downsampled image input port is used to input a 50μm downsampled image of the maximum projection map. The display module is used to display the projection map of the maximum value being operated on and the selected area in the current maximum value projection map. The visualization range of the selected area can be adjusted by pressing the buttons. The adjustment module is used to display a magnified view of the selected area and all the projection blocks within the magnified view. The incorrectly classified projection blocks can be reclassified by the mouse to obtain the final classification result after re-inspection. The export module is used to export the final classification result as a txt file, which contains the names of all neuron projection blocks.