Three-dimensional reconstruction data processing method for Golgi stained neurons and application thereof

Abstract

The invention provides a three-dimensional reconstruction data processing method for Golgi dyed neurons, which comprises the following steps: denoising an image data set of dyed neurons based on a gray threshold to obtain a denoised data set; performing point artifact removal on the noise reduction data set based on volume screening to obtain a continuous object data set; removing a high-contrast interferent disconnected from the target neuron from the continuous object data set to obtain a neuron candidate data set; extracting a target single neuron from the neuron candidate data set; and repairing the branch signal interruption of the target single neuron based on the direction consistency and / or gray connectivity of the branch paths. The invention further provides a corresponding computer readable storage medium and a computer program product. According to the three-dimensional reconstruction data processing method for the Golgi stained neurons, background noise reduction, single neuron extraction and branch repair are integrated, the neurons and other stained structures can be effectively distinguished, and reconstruction under the condition that nerve branch signals are interrupted is achieved.

Description

Technical Field

[0001] The present invention relates to the fields of biology and computer, and more particularly to a method for processing three-dimensional reconstruction data of Golgi-stained neurons and its application. Background Art

[0002] Analyzing the structure and function of the brain neural network is the core goal of neuroscience research. Among them, the study of neuron morphology plays an irreplaceable and important role in revealing the brain tissue structure and function. The Golgi staining technique can visually present the fine structures such as the cell body, dendrites, axons, and dendritic spines of neurons by randomly labeling neurons, and has become an important tool for studying neuron morphology. For example, Chailangkarn et al. stained the neurons in the V / VI layer of the postmortem human brain cortex by the Golgi method, and revealed the morphological changes of neurons related to Williams syndrome through optical microscopy imaging; Suzuki et al. found by Golgi staining that after injecting CPTX (a synthetic synaptic histoprotein) into the hippocampal region of 5xFAD mice (an Alzheimer's disease model), the decrease in the density of neuronal dendritic spines was restored under optical microscopy. Hwu et al. stained the whole brain of Drosophila by the Golgi method and obtained the three-dimensional structure of the neurons in the whole brain of Drosophila by combining X-ray imaging, verifying the applicability of the Golgi staining technique in synchrotron X-ray imaging.

[0003] However, Golgi staining samples still face challenges in neuron extraction and three-dimensional reconstruction, especially the widespread interruption of neuron branch signals. This signal interruption is mainly caused by the non-uniformity during the staining process and the discontinuity of dendritic labeling. In addition, the Golgi staining technique randomly labels about 1-3% of the neuron population, but the labeling density is still relatively high, resulting in increased difficulty in neuron extraction. Other factors such as incomplete perfusion of brain tissue, mechanical artifacts during sectioning, and non-specific staining of blood vessels and glial cells further exacerbate the interference of imaging noise and artifacts. These limit the complete reconstruction of neuron structures, may lead to the loss of key information of the neural network, and thus affect the reconstruction and functional analysis of complex brain neural networks.

[0004] To solve the above problems, researchers have developed a variety of neuron 3D reconstruction tools, which are mainly divided into fully automatic and semi-automatic categories. Fully automatic tools (such as NeuroGPS-Tree and NeuronCyto II) are based on computer algorithms and can automatically reconstruct neuron morphology with almost no user intervention. They are suitable for sparsely distributed and simple neuron populations. However, in the case of weak signals or branch signal interruptions, the reconstruction integrity of fully automatic tools is difficult to guarantee. Semi-automatic tools combine user input with algorithm calculations to assist in completing neuron reconstruction tasks. Currently commonly used semi-automatic tools (such as Neuromantic and ManSegTool) perform well in ultra-sparse fluorescent labeled samples, but in Golgi staining samples with complex backgrounds and high labeling density, these tools have difficulty in effectively distinguishing neurons from other stained structures, and their reconstruction capabilities are significantly insufficient in the case of signal interruptions.

[0005] Therefore, developing a technology that can repair signal interruptions and efficiently complete single neuron reconstruction has become an important issue that needs to be solved in this field. Therefore, it is very necessary to establish a data processing method for three-dimensional reconstruction of Golgi-stained neurons. Summary of the invention

[0006] The purpose of the present invention is to provide a three-dimensional reconstruction data processing method for Golgi-stained neurons and its application, so as to solve the problem that the current brain tissue Golgi-stained samples lack the method for repairing signal interruption and efficiently completing single neuron reconstruction.

[0007] In order to solve the above technical problems, the present invention provides a three-dimensional reconstruction data processing method for Golgi-stained neurons, comprising the steps of:

[0008] S1: Perform background noise reduction on the image dataset of stained neurons based on the grayscale threshold to obtain a noise reduction dataset;

[0009] S2: remove point artifacts from the denoised dataset based on volume screening to obtain a continuous object dataset;

[0010] S3: removing high-contrast interference objects disconnected from the target neuron from the continuous object data set to obtain a neuron candidate data set;

[0011] S4: Extract target single neurons from the neuron candidate dataset;

[0012] S5: Repair the branch signal interruption of the target single neuron based on the directional consistency and / or grayscale connectivity of the branch path.

[0013] In the step S1, the image data set comes from one of synchrotron X-ray micron tomography, microscopic optical tomography system, confocal microscope and light sheet microscope.

[0014] In the step S1, the image dataset is an 8-bit or 16-bit grayscale image obtained originally or through conversion; for an 8-bit grayscale image, its threshold is between 160 and 240; for a 16-bit grayscale image, its threshold is between 12,000 and 26,000.

[0015] In the step S2, the threshold for volume screening is between 1 and 3000 μm. 3 。

[0016] In the step S3, the high-contrast interferents include blood vessels, staining precipitates, glial cells, and non-target neurons; the method for removing high-contrast interferents disconnected from the target neurons includes: topologically analyzing whether there is a connection between the high-contrast object and the target neuron, and if not, determining it as a high-contrast interferent and removing it; or, manually judging whether there is a connection between the high-contrast object and the target neuron, and combining the morphological differences between the high-contrast interferent and the target neuron to determine whether it is a high-contrast interferent, and then removing the high-contrast interferent.

[0017] In the step S4, the method for extracting single neurons can be automatic extraction or manual extraction; automatic extraction is to segment regions with similar voxel intensities into single neurons to obtain the target single neurons; manual extraction is to start from the cell body of the target single neuron and extract by observing whether there is a connection between the surrounding branches and the cell body of the target single neuron.

[0018] The interruption of the branch signal includes the interruption of the neuron branch signal in the middle section and the non-recognition of the neuron branch.

[0019] The step S5 specifically includes:

[0020] S51: Set seed points in the cell body of the target single neuron;

[0021] S52: Manually set identification points at the ends of the branches directly connected to the seed points and at the ends of the disconnected neuron branches;

[0022] S53: Repair the interrupted signal of the neuron branch by combining the direction consistency and / or gray-level connectivity of the branch path.

[0023] In the step S51, the seed points are set manually or automatically based on the gray level of the image;

[0024] In the step S53, the direction consistency means calculating the local direction of each voxel / pixel at the identification point at the end of the branch directly connected to the seed point and searching for the identification point of the matching disconnected neuron branch end along this local direction to extend the neuron branch, and using curvature constraint during the searching process to prevent the neuron branch path from deviating suddenly; the gray connectivity means starting from the seed point and searching for the identification point of the matching disconnected neuron branch end along the voxels with similar gray values to extend the neuron branch.

[0025] On the other hand, the present invention provides a computer-readable storage medium, on which computer instructions are stored, and when the computer instructions are executed by a processor, the three-dimensional reconstruction data processing method of the Golgi-stained neurons described above is implemented.

[0026] On the other hand, the present invention provides a computer program product, including computer instructions, and when the computer instructions are executed by a processor, the three-dimensional reconstruction data processing method of the Golgi-stained neurons described above is implemented.

[0027] The three-dimensional reconstruction data processing method of the Golgi-stained neurons of the present invention integrates background noise reduction, single neuron extraction and branch repair, can effectively distinguish neurons from other stained structures, and realizes the reconstruction in the case of interrupted nerve branch signals. The present invention is applicable to the reconstruction of neurons in different brain regions and datasets of different imaging technologies, greatly expanding the application scope and flexibility of the method. Description of the Drawings

[0028] Figure 1 is a flowchart of a three-dimensional reconstruction data processing method of a Golgi-stained neuron according to an embodiment of the present invention.

[0029] Figure 2a is an image after noise reduction of the visual cortex in a micro-optical tomography dataset;

[0030] Figure 2b is the extraction result of single neurons in the visual cortex in a micro-optical tomography dataset;

[0031] Figure 2c is the result after repairing single neurons in the visual cortex in a micro-optical tomography dataset;

[0032] Figure 3a and Figure 3b are respectively images after noise reduction of the visual cortex and the sensory cortex in a micro-optical tomography dataset;

[0033] Figure 3c and Figure 3d are the extraction result diagrams of single neurons in the visual cortex and the sensory cortex in a micro-optical tomography dataset;

[0034] Figure 3e and Figure 3f are the post-repair result diagrams of single neurons in the visual cortex and sensory cortex in the microscopic optical tomography dataset;

[0035] Figure 3g and Figure 3h are the statistical diagrams of the total number of neuron branches and the total branch length before and after the repair of single neurons in the visual cortex and sensory cortex in the microscopic optical tomography dataset;

[0036] Figure 4a and Figure 4b are the images after noise reduction of the sensory cortex in the synchrotron radiation X-ray microtomography and microscopic optical tomography datasets;

[0037] Figure 4c and Figure 4d are the extraction result diagrams of single neurons in the sensory cortex in the synchrotron radiation X-ray microtomography and microscopic optical tomography datasets;

[0038] Figure 4e and Figure 4f are the post-repair result diagrams of single neurons in the sensory cortex in the synchrotron radiation X-ray microtomography and microscopic optical tomography datasets;

[0039] Figure 4g and Figure 4h are the statistical diagrams of the total number of neuron branches and the total branch length before and after the repair of single neurons in the sensory cortex in the synchrotron radiation X-ray microtomography and microscopic optical tomography datasets. Detailed implementation manners

[0040] The following further describes the present invention in conjunction with specific embodiments. It should be understood that the following embodiments are only used to illustrate the present invention and not to limit the scope of the present invention. For the experimental methods without specific conditions in the following embodiments, they are carried out according to conventional methods and conditions, or selected according to the product specifications. The reagents and raw materials used in the present invention are all commercially available.

[0041] As Figure 1 shown is a three-dimensional reconstruction data processing method for Golgi-stained neurons according to an embodiment of the present invention, which includes the following steps:

[0042] Step S1: Perform background noise reduction on the image dataset of stained neurons based on a gray threshold to obtain a noise-reduced dataset;

[0043] Step S2: Remove dot-like artifacts from the noise-reduced dataset based on volume screening to obtain a continuous object dataset;

[0044] Step S3: Remove high-contrast interferents (such as blood vessels, etc.) disconnected from the target neurons from the continuous object dataset to obtain a neuron candidate dataset;

[0045] Step S4: Extract the target single neuron from the neuron candidate data set;

[0046] Step S5: Repair the interruption of the branch signal of the target single neuron based on the direction consistency and / or gray-level connectivity of the branch path.

[0047] Considering that some steps require manual operation, a three-dimensional reconstruction data processing method for Golgi-stained neurons provided by the present invention is a semi-automatic method. The working principle of a three-dimensional reconstruction data processing method for Golgi-stained neurons of the present invention is as follows: Based on the gray-level difference between the Golgi-stained neurons and the tissue background, background noise reduction can be performed by setting a gray-level threshold to partially remove the tissue background; based on the volume of the dot-like artifacts, a volume threshold can be set to remove the dot-like artifacts; based on the fact that there is no connection between the high-contrast interfering objects (such as blood vessels, etc.) and the target neurons, the high-contrast interfering objects can be removed; based on the direction consistency and gray-level connectivity, the branches where the neuron branch signals are interrupted can be repaired.

[0048] Among them, in step S1, the image data set can be from one of a variety of imaging devices, and the imaging devices include: synchrotron radiation X-ray microtomography, micro-optical tomography system, confocal microscope, and light sheet microscope. Among them, the synchrotron radiation X-ray microtomography and the micro-optical tomography system have the best effects.

[0049] The image data set in step S1 should be an original 8-bit or 16-bit grayscale image; if the original image is in a format such as RGB, the image data set should be a converted 8-bit or 16-bit grayscale image. For images with different bit numbers, different gray-level thresholds can be selected for background noise reduction. For an 8-bit grayscale image, its threshold is between 160 - 240; for a 16-bit grayscale image, its threshold is between 12000 - 26000.

[0050] In addition, for the image data sets of different imaging techniques, the optimal gray-level thresholds are different. For the image data set of synchrotron radiation X-ray microtomography imaging, the optimal gray-level thresholds corresponding to 8-bit or 16-bit grayscale images are 220 and 23000 respectively. For the image data set of micro-optical tomography system imaging, the optimal gray-level thresholds corresponding to 8-bit or 16-bit grayscale images are 215 and 24000 respectively.

[0051] In step S2, the threshold for volume screening is between 1 - 3000 μm 3 , and the optimal volume threshold is 1000 μm 3 .

[0052] In step S3, the high-contrast interferents include blood vessels, staining precipitates, glial cells, and non-target neurons. The method for removing high-contrast interferents disconnected from the target neurons can be topological analysis and manual removal; among them, topological analysis is to analyze whether there is a connection between the high-contrast object and the target neuron. If not, it is determined as a high-contrast interferent and removed; manual removal is to manually judge whether there is a connection between the high-contrast object and the target neuron, and combine the morphological differences between the high-contrast interferent and the neuron to determine whether it is a high-contrast interferent, and then remove the high-contrast interferent. The best is manual removal.

[0053] In the said step S4, the method for extracting single neurons can be automatic extraction or manual extraction. Among them, automatic extraction is to divide the regions with similar voxel intensities into single neurons to obtain the target single neurons; manual extraction is to start from the cell body of the target single neuron and extract by observing whether there is a connection between the surrounding branches and the cell body of the target single neuron. The best method for extracting single neurons is manual extraction.

[0054] In the said step S5, there are two types of branch signal interruptions, namely neuron branch signal mid-interruption and neuron branch cannot be recognized.

[0055] The said step S5 specifically includes:

[0056] Step S51: Set seed points in the cell body of the target single neuron.

[0057] Among them, the seed points can be set manually or automatically based on the gray scale of the image (that is, find the point with the highest local brightness and automatically set it as the seed point). The best way is to set the seed points manually.

[0058] Step S52: Manually set identification points at the ends of the branches directly connected to the seed points and at the ends of the disconnected neuron branches;

[0059] Step S53: Combine the direction consistency and / or gray-scale connectivity of the branch path (the best is to combine both) to repair the interrupted signal of the neuron branch.

[0060] Among them, direction consistency means calculating the local direction of each voxel / pixel at the identification point at the end of the branch directly connected to the seed point and searching for the identification point at the end of the disconnected neuron branch that matches along this local direction to extend the neuron branch. During the search, curvature constraints are used to prevent the neuron branch path from suddenly deviating; among them, the curvature constraint specifically includes calculating the curvature and the change rate of the curvature, and realizing the smoothness of the path by minimizing the change rate of the curvature to avoid unnatural broken lines or sharp turns.

[0061] Gray connectivity means starting from a seed point, searching for the identification points at the ends of the disconnected neuron branches that match along the voxels with similar gray values to extend the neuron branches, thereby ensuring the continuity of nerve fibers and avoiding incorrect tracking caused by noise or low-contrast regions.

[0062] That is to say, the basis for repairing the interruption signal of neuron branches can be specifically divided into two cases: (1) Only the signal at the interruption is weak. At this time, the signal region is still stronger than the background signal, and it can be directly determined that the gray values are similar, and the interruption signal of the neuron branches is repaired according to gray connectivity; (2) There is no signal at the interruption. Since identification points are set, it can be determined that this neuron is continuous through direction consistency, and then the two broken branches at both ends are directly connected.

[0063] Repairing the interruption signal of neuron branches specifically includes: generating a smooth fiber structure image at the signal interruption between the identification point at the end of the branch directly connected to the seed point and the identification point at the end of the matching disconnected neuron branch. Specifically, the two broken identification points at both ends can be directly connected through the shortest path algorithm such as Dijkstra.

[0064] On the other hand, the present invention provides a computer-readable storage medium, on which computer instructions are stored, and when the computer instructions are executed by a processor, the three-dimensional reconstruction data processing method of the Golgi-stained neurons described above is implemented.

[0065] On the other hand, the present invention provides a computer program product, including computer instructions, and when the computer instructions are executed by a processor, the three-dimensional reconstruction data processing method of the Golgi-stained neurons described above is implemented.

[0066] Experimental results:

[0067] In the following multiple experiments, mouse brain tissues are selected, and the imaging methods are synchrotron radiation X-ray microtomography and microscopic optical tomography systems to specifically illustrate the implementation effects of the present invention.

[0068] Experiment 1: Three-dimensional reconstruction data processing method of Golgi-stained neurons implemented based on mouse brain neurons

[0069] Staining and imaging of mouse brain. After the brain tissue is dissected from the mouse brain, the mouse brain neurons are stained by Golgi staining, followed by reduction development and resin embedding, and finally imaged under a microscopic optical tomography system with a voxel resolution of 1 μm × 1 μm × 2.5 μm.

[0070] Three-dimensional reconstruction data processing method of Golgi-stained neurons in mouse brain neurons: Select the image dataset of the visual cortex. First, perform background noise reduction on the data, set the gray threshold to 220, and obtain the image after background noise reduction (seeFigure 2a )。Then, dot-like artifacts were removed from the data. Through volume screening, dot-like artifacts smaller than 1000μm 3 were removed, and high-contrast interferents such as blood vessels were removed. Subsequently, single neurons were extracted to obtain images of single neurons (see Figure 2b ). Finally, seed points were set at the neuron cell bodies, and marker points were manually set at the ends of neuron branches. Combining the direction consistency and gray-level connectivity of the branch paths, the interrupted signals of neuron branches were repaired to obtain the repaired images of single neurons (see Figure 2c ). The visualization results before and after repair showed that after adopting the three-dimensional reconstruction data processing method for Golgi-stained neurons of the present invention, the integrity of neuron reconstruction was greatly improved.

[0071] Experiment 2: Comparison of the repair effects of single neurons in different brain regions

[0072] Staining and imaging of mouse brains. After the brain tissues were dissected from the mouse brains, the neurons in the mouse brains were stained by Golgi staining, followed by reduction development and resin embedding. Finally, imaging was performed under a microscopic optical tomography system with a voxel resolution of 1 μm × 1 μm × 2.5 μm.

[0073] Three-dimensional reconstruction data processing method for Golgi-stained neurons in mouse brains: Image data sets of two brain regions, the visual cortex and the sensory cortex, were selected. First, the data was subjected to background noise reduction, and the gray-level threshold was set to 220 to obtain the image after background noise reduction (see the images of the visual cortex and the sensory cortex in Figure 3a 、 Figure 3b ). Then, dot-like artifacts were removed from the data. Through volume screening, dot-like artifacts smaller than 1000μm 3 were removed, and high-contrast interferents such as blood vessels were removed. Subsequently, single neurons were extracted to obtain images of single neurons (see the images of the visual cortex and the sensory cortex in Figure 3c 、 Figure 3d ). Finally, seed points were set at the neuron cell bodies, and marker points were manually set at the ends of neuron branches. Combining the direction consistency and gray-level connectivity of the branch paths, the interrupted signals of neuron branches were repaired to obtain the repaired images of single neurons (see the images of the visual cortex and the sensory cortex in Figure 3e 、 Figure 3f ). The visualization results before and after repair showed that after adopting the three-dimensional reconstruction data processing method for Golgi-stained neurons of the present invention, the integrity of neuron reconstruction was greatly improved.

[0074] Comparison of repair effects. We conducted a quantitative analysis of the morphological characteristics of neurons before and after repair, including the total number of branches and the total length of neurons. The results showed that the total number of branches of neurons in the visual cortex and the sensory cortex increased from 19 and 23 before repair to 30 and 30 respectively, an increase of 58% and 30% respectively. The total branch length of neurons increased from 1553 and 1473 before repair to 2896 and 2595 respectively, an increase of 86% and 76% (for the statistical results of the visual cortex and the sensory cortex, see Figure 3g , Figure 3h ). The above results indicate that the three-dimensional reconstruction data processing method for Golgi-stained neurons provided by the present invention can be widely applied to the extraction and reconstruction of neurons in different brain regions, demonstrating excellent applicability and reconstruction effects.

[0075] Experiment 3: Comparison of the repair effects of single neurons in datasets of different imaging techniques

[0076] Staining and imaging of the mouse brain. After the brain tissue was dissected from the mouse brain, the neurons in the mouse brain were stained by Golgi staining, followed by reduction development. Then, paraffin embedding and resin embedding were performed respectively, and finally, imaging was performed under a synchrotron radiation X-ray microtomography system and a microscopic optical tomography system, with voxel resolutions of 0.65 μm * 0.65 μm * 0.65 μm and 1 μm * 1 μm * 2.5 μm respectively.

[0077] Three-dimensional reconstruction data processing method for Golgi-stained neurons in the mouse brain: In both datasets, the image dataset of the sensory cortex was selected. First, the data was denoised for the background. For synchrotron radiation X-ray microtomography, the gray threshold was set to 215, and for the microscopic optical tomography system, the gray threshold was set to 220, to obtain the image after background denoising (for the images of synchrotron radiation X-ray microtomography and the microscopic optical tomography system, see Figure 4a , Figure 4b ). Then, dot artifacts were removed from the data. Through volume screening, dot artifacts smaller than 1000 μm 3 were removed, and high-contrast interferents such as blood vessels were removed. Subsequently, single neurons were extracted to obtain the images of single neurons (for the images of synchrotron radiation X-ray microtomography and the microscopic optical tomography system, see Figure 4c , Figure 4d ). Finally, we set seed points at the neuron cell bodies, manually set the identification points at the ends of neuron branches, and combined the direction consistency and gray connectivity of the branch paths to repair the interrupted signals of neuron branches, obtaining the images of repaired single neurons (for the images of synchrotron radiation X-ray microtomography and the microscopic optical tomography system, see Figure 4e , Figure 4f ). The visualization results before and after repair indicate that the integrity of neuron reconstruction has been greatly improved.

[0078] Comparison of repair effects. We performed a quantitative analysis of the morphological characteristics of neurons before and after repair, including the total number of branches and the total length of neurons. The results showed that for synchrotron X-ray microtomography and microscopic optical tomography systems, the total number of branches of neurons in the sensory cortex increased from 12 and 23 before repair to 15 and 30, respectively, an increase of 25% and 30%. The total branch lengths of neurons increased from 1960 and 1473 before repair to 2726 and 2595, respectively, an increase of 39% and 76% (see the statistical results of synchrotron X-ray microtomography and microscopic optical tomography systems in Figure 4g , Figure 4h ). The above results indicate that the three-dimensional reconstruction data processing method for Golgi-stained neurons provided by the present invention can be widely applied to the extraction and reconstruction of neurons in different imaging technologies, demonstrating excellent applicability and reconstruction effects.

[0079] Currently, single-neuron reconstruction tools developed by researchers can segment sparsely distributed and morphologically simple neuron populations. However, in Golgi-stained brain tissue samples with complex backgrounds and high labeling densities, these tools are difficult to effectively distinguish neurons from other stained structures, and their reconstruction ability is significantly insufficient in the case of signal interruption. The three-dimensional reconstruction data processing method for Golgi-stained neurons of the present invention integrates background noise reduction, single-neuron extraction, and branch repair, can effectively distinguish neurons from other stained structures, and realizes reconstruction in the case of interrupted nerve branch signals.

[0080] The method of the present invention solves the technical challenges of segmentation difficulties and interrupted branch signals faced in the three-dimensional reconstruction of neurons in Golgi-stained samples, and can be widely applied to the reconstruction of neurons in a variety of imaging technologies and different brain regions. By improving the integrity and complexity of neuron morphology, this method has important application potential in neuroscience research.

[0081] In summary, the present invention provides a three-dimensional reconstruction data processing method for Golgi-stained neurons and its application, which has good application prospects in neuroscience.

[0082] The above-described are only the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various changes can be made to the above embodiments of the present invention. All simple, equivalent changes and modifications made according to the claims and the content of the specification of the present invention application fall within the scope of the claims of the present invention patent. What the present invention does not describe in detail is conventional technical content.

Claims

1. A method for processing data of three-dimensional reconstruction of Golgi-stained neurons, characterized in that: include: Step S1: performing background noise reduction on the image dataset of stained neurons based on the grayscale threshold to obtain a noise reduction dataset; Step S2: removing point artifacts from the denoised data set based on volume screening to obtain a continuous object data set; Step S3: removing high-contrast interferers disconnected from the target neuron from the continuous object data set to obtain a neuron candidate data set; Step S4: extracting target single neurons from the neuron candidate data set; Step S5: Based on the directional consistency and / or grayscale connectivity of the branch path, the branch signal interruption of the target single neuron is repaired.

2. The three-dimensional reconstruction data processing method according to claim 1, characterized in that: In the step S1, the image data set comes from one of synchrotron X-ray micron tomography, microscopic optical tomography system, confocal microscope and light sheet microscope.

3. The three-dimensional reconstruction data processing method according to claim 1, characterized in that: In step S1, the image data set is an original or converted 8-bit or 16-bit grayscale image; for an 8-bit grayscale image, the threshold is between 160 and 240; for a 16-bit grayscale image, the threshold is between 12000 and 26000.

4. The three-dimensional reconstruction data processing method according to claim 1, characterized in that: In step S2, the threshold of volume screening is 1-3000 μm 3 .

5. The three-dimensional reconstruction data processing method according to claim 1, characterized in that: In step S3, the high-contrast interferences include blood vessels, stained precipitates, glial cells and non-target neurons; The method for removing high-contrast interference objects that are disconnected from target neurons includes: topologically analyzing whether there is a connection between the high-contrast object and the target neuron, and if not, determining it as a high-contrast interference object and removing it; Alternatively, it is manually determined whether there is a connection between the high-contrast object and the target neuron, and the morphological difference between the high-contrast distractor and the target neuron is combined to determine whether it is a high-contrast distractor, and then the high-contrast distractor is removed.

6. The three-dimensional reconstruction data processing method according to claim 1, characterized in that: In step S4, the method of extracting single neurons includes automatic extraction or manual extraction; automatic extraction is to segment the region with similar voxel intensity into single neurons to obtain the target single neuron; manual extraction is to extract by taking the cell body of the target single neuron as the starting point and observing whether there is a connection between the surrounding branches and the cell body of the target single neuron; and / or The branch signal interruption includes the interruption of the middle section of the neuron branch signal and the inability to identify the neuron branch.

7. The three-dimensional reconstruction data processing method according to claim 1, characterized in that: The step S5 specifically includes: Step S51: setting a seed point in the cell body of the target single neuron; Step S52: manually setting identification points at the ends of branches directly connected to the seed point and at the ends of disconnected neuron branches; Step S53: Repair the interruption signal of the neuron branch by combining the directional consistency and / or grayscale connectivity of the branch path.

8. The three-dimensional reconstruction data processing method according to claim 7, characterized in that: In the step S51, the seed point is manually set or automatically set based on the grayscale of the image; and / or In step S53, directional consistency refers to calculating the local direction of each voxel / pixel at the identification point of the branch end directly connected to the seed point and searching for the matching identification point of the disconnected neuron branch end along the local direction to expand the neuron branch. During the search process, curvature constraints are used to prevent sudden deviations of the neuron branch path. Grayscale connectivity refers to starting from the seed point, searching for the matching identification point of the disconnected neuron branch end along the voxels with similar grayscale values ​​to expand the neuron branch.

9. A computer-readable storage medium having computer instructions stored thereon, characterized in that: When the computer instruction is executed by a processor, the three-dimensional reconstruction data processing method of Golgi-stained neurons described in any one of claims 1 to 8 is implemented.

10. A computer program product comprising computer instructions, characterized in that: When the computer instruction is executed by a processor, the three-dimensional reconstruction data processing method of Golgi-stained neurons described in any one of claims 1 to 8 is implemented.

Images