A digital reconstruction method and system for neuron shape

Abstract

The present invention discloses a method and system for digital reconstruction of neuron shape. Aiming at the morphological data of neuron protrusions distributed throughout the brain, especially the data lacking path node radius information, a method for optimizing protrusion paths and automatically estimating path node radius that can be used for neuron shape reconstruction is designed in combination with a sub-micron resolution image data set obtained by fiber optical imaging technology. The present invention can correct the neuron protrusion path of existing neuron protrusions, improve the degree of fit between the neuron protrusion path and the center line of the neuron protrusion in the image, and eliminate the jagged phenomenon of the neuron protrusion path. At the same time, the radius of each neuron protrusion path node is automatically estimated in combination with the neuron image and the optimized neuron protrusion path, thereby improving the accuracy and calculation efficiency of the existing neuron protrusion path radius.

Description

Technical Field

[0001] The invention relates to the field of biomedical graphics processing, and in particular to a method and system for digitally reconstructing neuron shapes. Background Art

[0002] The brain is the most important organ in the human body. The normal operation of various life activities of the body cannot be separated from the control and regulation of the brain. At the same time, the brain is also the most complex organ in the human body. The brain contains a large number of neurons and glial cells. These cells form an intricate vascular neural network that is interconnected with the abundant blood vessels in the brain. Neurons are the basic units that make up the structure of the nervous system in the brain and produce brain functions. They are composed of cell bodies and cell processes. Among them, cell processes are divided into two types: axons and dendrites according to differences in morphology and function. The complex functions of the brain depend on a large number of neurons and the neural circuits formed by the interconnection between neurons. The morphological changes of neuronal processes affect the activities and functions of neurons. There are obvious differences in the morphology of axons and dendrites of different neurons. Therefore, it is of great significance to study neuronal morphology.

[0003] The acquisition of neuronal morphological data requires the combination of multiple technologies. By imaging brain tissue with sparse labeling technology and optical imaging technology, neuronal image data with submicron resolution can be obtained. To obtain vectorized neuronal morphological data from these image data, neuronal morphological reconstruction technology is required. Neuronal morphological reconstruction mainly includes three parts: image segmentation of neuronal cell bodies, reconstruction of neuronal processes, and identification of fine structures inside neuronal processes (such as axonal boutons and dendritic spines). Among them, neuronal process shape reconstruction is the basic but difficult content in neuronal morphological reconstruction.

[0004] At present, the method of neuronal process reconstruction is mainly to analyze the image characteristics of neuronal images, and then design or select corresponding mathematical models to analyze the images based on the characteristics of these images, and then extract the topological structure of neuronal processes by connecting points and lines, which are called "neuronal process paths". These extracted paths are the reconstruction results of neuronal processes. The neuronal process reconstruction results are usually represented by vector files with three-dimensional spatial coordinates and cylinder radius data connected by multiple cylindrical models.

[0005] Currently, the main focus of these neuronal process reconstruction methods is the integrity of the neuronal process paths and the correctness of the connections between the paths. That is, these methods mainly depict the paths of the same neuronal process that people can see in the neuronal image in full, and ensure that different neuronal processes in the neuronal image will not be identified as the same neuronal process, and the connection relationship between the branches is consistent with the connection relationship seen by people through the image. However, few people pay attention to the shape reconstruction of neuronal processes and the accuracy of the reconstruction results.

[0006] Neuron shape reconstruction refers to reconstructing a vector file that can describe the shape of neurons based on the reconstruction of the neuron process path and combining morphological parameters such as the neuron radius. The accuracy of the reconstruction result includes two aspects. The first is the degree of fit between the reconstructed neuron process path and the center line of the neuron process in the image. The second is the accuracy of the neuron radius estimated by combining the nerve fiber image and the neuron process path. Compared with the reconstruction of the neuron process path, while ensuring the accuracy of the reconstruction results, the reconstruction of the shape of neurons can provide more indispensable key data for subsequent neuron morphological analysis and neuron simulation.

[0007] Regarding the problem of the degree of fit between the neuron process path and the center line of the neuron process in the image, which is involved in the accuracy of the neuron shape reconstruction result, the integrity of the neuron process signal and the correctness of the connection are very difficult for the machine automatic recognition algorithm without human intervention because the neuron process signal and the background signal in the image are complex and changeable. However, for the reconstruction process of the neuron process with human assistance, although the integrity and correctness of the neuron process are guaranteed, the accuracy of the reconstructed neuron process path needs to be improved. In order to improve the accuracy of the reconstructed neuron process path, the existing methods mostly use the grayscale characteristics of the neuron process in the image to correct the neuron process path. However, due to the influence of various factors such as sample labeling and imaging conditions, the neuron process signal and the background signal in the image are not constant. This complex signal will eventually lead to the phenomenon of jagged neuron process path. And this jagged neuron process path is quite different from the smooth shape of the neuron process in the image.

[0008] The accuracy of neuron shape reconstruction results involves the accuracy of neuron process radius estimation. The main idea of ​​the prior art is to extract the image grayscale features near the nodes of the neuron process path, and then segment the image and estimate its radius. Using this method, the accuracy of the radius of the extracted neuron process depends on the quality of the image. Due to the influence of various factors such as biomarkers and imaging, the signal intensity of the neuron process and the background signal intensity in the whole brain image data may not always remain consistent. For algorithms that rely heavily on image features, the changing signal will cause the accuracy of the estimated neuron process radius to be very unstable.

[0009] In summary, there are few existing neuron shape reconstruction technologies, and the currently commonly used automatic and semi-automatic neuron process reconstruction algorithms cannot guarantee the integrity and correctness of the reconstructed neuron process paths. The current neuron process reconstruction algorithms estimate the radius of neuron process paths roughly, and when applied to whole-brain image data, they are easily affected by image quality, resulting in low accuracy of the calculated radius data of neuron process path nodes. Summary of the invention

[0010] Therefore, in order to solve the above problems, the present invention provides a method and system for digitally reconstructing neuron shape.

[0011] The present invention is achieved through the following technical solutions:

[0012] A method for digitally reconstructing neuron shape, comprising the following steps:

[0013] S1: insert a series of equally spaced nodes P0 in the neuron process path to describe the center line of the neuron process path by linear interpolation;

[0014] S2: sequentially calculate the pixel node P1 with the strongest signal strength in the local range with P0 as the center point, and use P1 as the correction point of P0;

[0015] S3: cyclically checking and correcting the node P1 through angle constraints and signal strength constraints to obtain a finely corrected node P2, and connecting each node P2 to obtain a finely corrected neuron protrusion path;

[0016] S4: Extract the orthogonal planes of each node on the neuron process path;

[0017] S5: Segment each orthogonal plane and extract the foreground signal;

[0018] S6: Estimate the radius of each orthogonal plane;

[0019] S7: After cyclically judging the accuracy of the calculation results, the calculation results are output.

[0020] Preferably, in step S2, the convolution of all pixel points in the local range with P0 as the center point and the Gaussian kernel function is determined in the three-axis directions respectively, so as to obtain the three-axis coordinates of the pixel node P1 with the strongest signal intensity in the local range with P0 as the center point, and the calculation formula of the P1 coordinate is:

[0021]

[0022] Among them, Gau represents the Gaussian convolution kernel, and G(P0) is the function of P0 on the three-dimensional coordinate axis.

[0023] Preferably, the angle constraint condition is: the angle formed by the neuron process path node P1 and the connecting line between its parent and child nodes is greater than or equal to 120°; the signal constraint condition is: judging whether the signal strength of the neuron process path node P1 is greater than the signal strength of its surrounding neighborhood and its parent and child nodes.

[0024] Preferably, in step S3, if the node P1 meets both the angle constraint condition and the signal constraint condition, the result is retained; if the node P1 does not meet one or both of the conditions, steps S2 and S3 are repeated.

[0025] Preferably, step S4 comprises:

[0026] S41: Based on the node P2 and its parent node P f , child node P c The position relationship is obtained, and the arc C1 is obtained;

[0027] S42: According to arc C1, node P2, P f and P c The topological relationship is used to calculate the tangent vector V1 of the arc C1 at the node P2;

[0028] S43: Obtain a plane A1 with V1 as the normal vector through the tangent vector V1, wherein A1 is an orthogonal plane of the neuron process path node P2.

[0029] Preferably, step S5 comprises:

[0030] S51: extract all pixel points on plane A1 to obtain an orthogonal plane image of the neuron process path node P2;

[0031] S52: enhancing the gap between the foreground signal and the background signal of the neuron process in the orthogonal plane image;

[0032] S53: Binarizing the pixels on the orthogonal plane image to extract foreground signals;

[0033] S54: Separate the interference signal by using the opening operation of image morphology;

[0034] S55: Use connected domain analysis to mark the orthogonal plane A2 after separating the interference signal.

[0035] Preferably, step S6 comprises:

[0036] S61: Calculate the number N2 of pixels of the foreground signal in the orthogonal plane A2;

[0037] S62: Calculate the area S2 occupied by all pixels in the orthogonal plane A2 using the image resolution and the number of pixels N2;

[0038] S63: Obtain the radius of the orthogonal plane A2 using a circular area calculation formula.

[0039] Preferably, in step S7, the accuracy of the calculation result is judged cyclically, and if the calculation result is accurate, the calculation result is output; if the calculation result is inaccurate, steps S2 to S6 are repeated until the calculation result is judged to be accurate and then output.

[0040] Preferably, in step S7, the accuracy of the calculation result is determined based on the following formula:

[0041]

[0042]

[0043] α=a*cosγ o +b (1, 4)

[0044] Where p is the point where the metric value F is maximized, γ is the angle between the two vectors formed by the input node P2 and its parent and child nodes on the search path, Gp represents the signal strength of point P on the path, β is the weight factor representing the signal strength, α is the weight factor representing the angle, Dis represents the distance between the search point and the calibration starting point, a, b, c are set values, and γ o Represents the angle between the two vectors formed by the input node P2 and its parent and child nodes.

[0045] The present invention also discloses a digital reconstruction system of neuron shape, comprising:

[0046] A neuron process path linear interpolation unit, used for inserting a series of equally spaced nodes P0 for describing the center line of the neuron process path into the neuron process path by linear interpolation;

[0047] A neuron protrusion path preliminary correction unit is used to sequentially calculate the pixel node P1 with the strongest signal intensity in the local range with P0 as the center point, and use P1 as the correction point of P0;

[0048] A neuron process path fine correction unit is used to perform cyclic inspection and correction on the node P1 through angle constraints and signal strength constraints to obtain a finely corrected node P2, connect each node P2, and obtain a finely corrected neuron process path;

[0049] A path node orthogonal plane extraction unit is used to extract the orthogonal planes of each node on the neuron process path;

[0050] A path node orthogonal plane segmentation unit is used to segment each orthogonal plane and extract the foreground signal;

[0051] A radius estimation unit, used for estimating the radius of each orthogonal plane;

[0052] The loop judgment unit is used to loop judge the accuracy of the calculation result and then output the calculation result.

[0053] The beneficial effects of the technical solution of the present invention are mainly reflected in:

[0054] 1. Through preliminary correction, the pixel point with the strongest signal in the neuron process path is found, and then the angle constraint and signal constraint are used to perform cyclic judgment and precise correction on the pixel point with the strongest signal in the neuron process path. The smooth characteristics of the neuron process are used as the constraint condition, so as to effectively avoid the jagged phenomenon of the neuron process path and improve the accuracy of the neuron process path.

[0055] 2. A series of nodes representing the center line of the neuron process path are used to realize the correction of the neuron process path and the reconstruction of the neuron shape. The spacing of the nodes can be adjusted according to the reconstruction accuracy and the setting is flexible. Compared with the existing method of using multiple cylindrical models to connect the reconstruction method, the reconstruction result is more accurate.

[0056] 3. The radius of the path nodes is estimated based on the topological structure of the neuronal process path nodes. It has low dependence on image quality. The accuracy of the radius of the estimated neuronal process path nodes is stable, which can effectively improve the accuracy of radius estimation. In the whole-brain three-dimensional neuronal process image data, the problem of quality fluctuation of the three-dimensional image caused by various factors such as labeling and imaging is avoided. It can be applied to whole-brain image data, solving the problem that traditional methods are heavily dependent on the quality and image characteristics of the whole-brain image.

[0057] 4. A method for reconstructing the neuron process path by using the local grayscale features of the nodes in the neuron process path to estimate the radius information, using the linear change of grayscale to appropriately enhance the gap between the neuron process signal and the background signal in the image, then using binarization to extract the foreground signal, and then using the opening operation of image morphology to separate the interference signal, thereby reducing the interference of the background signal.

[0058] 5. Before outputting the calculation results, the corrected neuron protrusion path and the orthogonal plane radius results are cyclically judged to further verify the calculation results, thereby improving the accuracy of the neuron shape reconstruction results.

[0059] 6. On the basis of neuronal process path reconstruction, the degree of fit between neuronal process path and the center line of neuronal process in the image is improved, and the radius information of neuronal path nodes is combined to achieve accurate reconstruction of neuronal shape.

[0060] 8. It runs fully automatically without human intervention and can process all neuronal morphological files by itself. With the support of appropriate computing resources and image data, it can greatly save manpower and time and improve the quality of neuronal morphological data. It is applicable to many types of computing platforms. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 : is a flow chart of a method for digitally reconstructing neuron shape according to the present invention;

[0062] Figure 2 : is a program flow chart of the digital reconstruction system of neuron shape of the present invention;

[0063] Figure 3 : is a schematic diagram of the neuron process path in the mouse experiment of the present invention;

[0064] Figure 4 :yes Figure 3 A magnified view of part A;

[0065] Figure 5 : Yes Figure 4 Schematic diagram after performing step S1;

[0066] Figure 6 : Yes Figure 5 Schematic diagram after preliminary calibration in step S2;

[0067] Figure 7 : Yes Figure 6 Schematic diagram after fine correction in step S3;

[0068] Figure 8 : Yes Figure 6 Schematic diagram of the tangent vector V1 at each node P2;

[0069] Fig. 9 : It is a schematic diagram of the orthogonal plane A1 of the node P2;

[0070] Fig.10 : It is a schematic diagram of the linear change of grayscale of orthogonal plane A1;

[0071] Fig.11 : Yes Fig.10 Schematic diagram after binarization operation;

[0072] Fig.12 : Yes Fig.11 Schematic diagram after image morphological opening operation;

[0073] Fig.13 : Yes Fig.12 Schematic diagram of the radius estimation operation of the orthogonal planes in . DETAILED DESCRIPTION

[0074] In order to make the purpose, advantages and features of the present invention more clearly and in detail, the following preferred embodiments are used for illustration and explanation. The embodiments are only typical examples of the application of the technical solution of the present invention. Any technical solution formed by equivalent replacement or equivalent transformation falls within the scope of protection claimed by the present invention.

[0075] At the same time, it is stated that in the description of the scheme, it should be noted that the directions or positional relationships indicated by the terms "center", "up", "down", "left", "right", "front", "back", "inside" and "outside" are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of description and simplified description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as a limitation on the present invention.

[0076] In addition, the terms "first" and "second" in this solution are used for descriptive purposes only and cannot be understood as indicating or implying the order of importance, or implicitly indicating the number of technical features shown. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features.

[0077] In the present invention, "plurality" means two or more than two, unless otherwise clearly and specifically defined.

[0078] The present invention discloses a method for digital reconstruction of neuron shape, which is applied to fMOST imaging and GFP fluorescent labeling, with an image resolution of 0.2×0.2×1μm. 3 The mouse whole brain 3D image dataset was used to obtain the attached Figure 3-Figure 13 .

[0079] like Figure 1 As shown, a method for digitally reconstructing neuron shape of the present invention comprises the following steps:

[0080] S1: insert a series of equally spaced nodes P0 in the neuron process path to describe the center line of the neuron process path by linear interpolation;

[0081] like Figure 3-Figure 5 As shown, Figure 3 Taking part A in the figure as an example, a series of equally spaced nodes P0 for describing the center line of the neuronal process path are inserted into the neuronal process path to provide multiple calculation samples for subsequent correction.

[0082] Specifically, in this embodiment, the linear interpolation method refers to: first presetting a minimum distance D in the neuron process path, when the distance between a pair of parent and child nodes is greater than D, inserting a number of nodes P0 at equal intervals between the two, and the type of the expanded node P0 is consistent with the parent node of the two skeleton points by default; in order to avoid adjacent skeleton points corresponding to the same voxel in the original data, the selection of the minimum distance D is mainly related to the voxel resolution of the original data. For example, if the resolution of the original data is 0.2μm×0.2μm×1μm, based on the above principle, the minimum distance D should be set to 2μm, which is twice the axial resolution.

[0083] S2: sequentially calculate the pixel node P1 with the strongest signal strength in the local range with P0 as the center point, and use P1 as the correction point of P0;

[0084] like Figure 6 As shown, in one embodiment, the pixel node P1 with the strongest signal intensity in the local range with P0 as the center point is calculated by the mean shift method, and the neuron protrusion path is preliminarily corrected. Figure 5 , Figure 6 It can be clearly seen that the change of the center line representing the neuron protrusion path is determined by convolution of all pixel points in the local range with P0 as the center point and the Gaussian kernel function in the three-axis direction, thereby obtaining the three-axis coordinates of the pixel node P1 with the strongest signal intensity in the local range with P0 as the center point. The calculation formula of P1 coordinates is:

[0085]

[0086] Among them, Gau represents the Gaussian convolution kernel, and G(P0) is the function of P0 on the three-dimensional coordinate axis.

[0087] Specifically, G(P) represents a function of a local image with point P as the center, P1 is a parameter in the function G(P0) that can obtain the maximum output value, and the above formula is used to calculate the maximum grayscale values ​​in the X-axis direction, the Y-axis direction and the Z-axis direction within the local range with P0 as the center point (when calculating the maximum grayscale value in one direction, the coordinates of the other two directions remain unchanged), and the corresponding coordinate values ​​of the maximum grayscale value in the three-axis directions are the three-axis coordinate values ​​of the pixel node P1 with the strongest signal intensity. After determining the three-axis coordinates of node P1, P1 is set as the correction point of P0.

[0088] Among them, considering the factors such as the diameter of the nerve fiber and the axial resolution, the initial value of the half-width of the axial coordinate movement interval in the present invention is selected as 2, that is, assuming that the initial coordinates of point P are (x, y, z), then the axial coordinate interval under the initial conditions is [z-2, z+2]

[0089] S3: cyclically checking and correcting the node P1 through angle constraints and signal strength constraints to obtain a finely corrected node P2, and connecting each node P2 to obtain a finely corrected neuron protrusion path;

[0090] Among them, when the angle formed by the connection between node P1 and its parent and child nodes is less than 120°, this skeleton point may deviate from the correct position, so the angle constraint condition is: the angle formed by the connection between the neuron process path node P1 and its parent and child nodes is greater than or equal to 120°; the signal constraint condition is: it is judged that the signal strength of the neuron process path node P1 is greater than the signal strength of its surrounding neighborhood and its parent and child nodes; it should be noted that the starting point and the end point lack parent nodes and child nodes respectively, so the starting point and the end point are not corrected.

[0091] like Figure 7 As shown, in this embodiment, if the node P1 meets both the angle constraint and the signal constraint, the result is retained; if the node P1 does not meet one or both of the conditions, steps S2 and S3 are repeated until a node P2 that meets both the angle constraint and the signal constraint is obtained, and the node P2 is set as the correction point of P1, and each node P2 is connected to obtain a neuron protrusion path that is highly aligned with the center line of the neuron protrusion in the image. Figure 6 , Figure 7 The center line representing the neuron protrusion path changes further as each node P2 makes a fine correction to the node P1, and each node P1 on the center line is corrected cyclically to improve the fit between the center line and the neuron protrusion path image.

[0092] S4: Extract the orthogonal planes of each node on the neuron process path;

[0093] In one embodiment, based on the topological relationship between the locations of the finely corrected neuronal process path node P2 and its parent and child nodes, the tangent vector along the center line of the neuronal fiber in the image at point P2 is calculated, and the plane with the tangent vector as the normal vector is the orthogonal plane A1 of the extracted neuronal process path node P2.

[0094] These include the following steps:

[0095] S41: Based on the node P2 and its parent node P f , child node P c The position relationship is obtained, and the arc C1 is obtained;

[0096] Among them, for the corrected neuron protrusion path node P2, its parent node P f and child node P c , we can determine an inscribed circle with a very large radius so that these three points exist on the arc C1 at the same time.

[0097] S42: According to arc C1, node P2, P f and P c The topological relationship is used to calculate the tangent vector V1 of the arc C1 at the node P2;

[0098] like Figure 8 As shown, when determining the parent node P of node P2 f and child node P c And arc C1, find the tangent vector V1 at node P2 on arc C1.

[0099] S43: Obtain a plane A1 with V1 as the normal vector through the tangent vector V1, wherein A1 is an orthogonal plane of the neuron process path node P2.

[0100] like Fig. 9 As shown, it is the orthogonal plane A1 at the node P2. At this time, the image of the orthogonal plane A1 is relatively blurred, and the foreground signal and the background signal cannot be clearly distinguished.

[0101] S5: Segment each orthogonal plane and extract the foreground signal;

[0102] In one embodiment, the orthogonal plane A1 is first subjected to a grayscale linear transformation of the image, and the pixel signal of the neuron process path node P2 is enhanced. Then, the foreground signal is extracted through an adaptive image binarization method. Next, the image morphological opening operation is used to remove interference noise and signals. Finally, the connected domain analysis is used to segment the neuron process path node orthogonal plane A2 composed of the cylindrical structure pixel points of the neuron process.

[0103] These include the following steps:

[0104] S51: extract all pixel points on plane A1 to obtain an orthogonal plane image of the neuron process path node P2;

[0105] According to the calculated algebraic expression of A1, based on the whole brain three-dimensional image data, all pixel points on the orthogonal plane A1 of each neuronal process path node P2 can be extracted to obtain the orthogonal plane image at each neuronal process path node P2.

[0106] S52: enhancing the gap between the foreground signal and the background signal of the neuron process in the orthogonal plane image; Fig.10 As shown, the pixel signal of the neuron protrusion path node P2 is enhanced by using the grayscale linear transformation of the image. Fig. 9 , Fig.10 ,at this time Fig.10 The gap between the signal of the neuron and the background signal is Fig. 9 Significantly enhanced compared to the previous study.

[0107] S53: Binarizing the pixels on the orthogonal plane image to extract foreground signals;

[0108] like Fig.11 As shown, the image is binarized using an adaptive image binarization method, and the entire image presents an obvious visual effect of only black and white, further highlighting and extracting the foreground signal.

[0109] S54: Separate the interference signal by using the opening operation of image morphology;

[0110] The opening operation in image morphology is equivalent to performing an erosion operation on the image first and then a dilation operation, which can eliminate discrete points and "burrs" and separate two objects. In this embodiment, Fig.12 As shown, the foreground signal represented by the cylindrical structure of protruding fibers and the interference signal around it are separated by an opening operation.

[0111] S55: Use connected domain analysis to mark the orthogonal plane A2 after separating the interference signal.

[0112] S6: Estimate the radius of each orthogonal plane;

[0113] like Fig.13 As shown, in one embodiment, the number of pixels N2 of the foreground signal of the orthogonal plane A2 is calculated, and the area S2 corresponding to the N2 pixels is calculated according to the relevant parameters of the image resolution. The orthogonal plane shape of the protruding fiber is approximated by a circle, and according to the circular area calculation formula, it can be deduced that the estimated value of its radius is:

[0114] These include the following steps:

[0115] S61: Calculate the number N2 of pixels of the foreground signal in the orthogonal plane A2;

[0116] S62: Calculate the area S2 occupied by all pixels in the orthogonal plane A2 using the image resolution and the number of pixels N2;

[0117] S63: Obtain the radius of the orthogonal plane A2 using a circular area calculation formula.

[0118] S7: After cyclically judging the accuracy of the calculation results, the calculation results are output.

[0119] In step S7, the accuracy of the calculation result is judged cyclically. If the calculation result is accurate, the calculation result is output; if the calculation result is inaccurate, steps S2 to S6 are repeated until the calculation result is judged to be accurate and then output.

[0120] In step S7, the accuracy of the calculation result is determined based on the following formula:

[0121]

[0122]

[0123] α=a*cosγ o +b (1, 4)

[0124] Where p is the point where the metric value F is maximized, γ is the angle between the two vectors formed by the input node P2 and its parent and child nodes on the search path, Gp represents the signal strength of point P on the path, β is the weight factor representing the signal strength, α is the weight factor representing the angle, Dis represents the distance between the search point and the calibration starting point, a, b, c are set values, and γ o Represents the angle between the two vectors formed by the input node P2 and its parent and child nodes.

[0125] Specifically, the finely corrected neuron process path node P2 is input into formula (1, 2). If node P2 is the point on the search path that maximizes the metric value F, node P2 is judged to be the final corrected node, that is, the calculation result of node P2 is accurate.

[0126] Specifically, γ represents the angle between the point P on the search path and the two vectors formed by the parent and child nodes. The larger the γ, the smoother the searched path.

[0127] Specifically, during the search process, the larger the β is, the greater the impact of the signal strength on the metric F is. oWhen the signal strength is less than the reference signal strength, it means that the input node P2 is likely to deviate from the foreground signal. In the correction search process, it is necessary to consider the impact of signal strength changes on F more. On the contrary, if G o If the signal strength is greater than the reference signal strength, it means that the skeleton point is already in the foreground signal, and the signal strength change has little effect on F. o Represents the angle between the two vectors formed by the input node P2 and its parent and child nodes. When γ o The smaller it is, the larger α is. At this time, the greater the influence of the angle on the metric value F, which means that the skeleton point may have been affected by other fiber signals in the previous correction process and deviated from the correct position. Therefore, the influence of the angle needs to be considered more during the search process.

[0128] Specifically, a and b are two empirical factors, a is generally taken as 0.4, b is generally taken as 0.7, and c is a parameter less than 1, generally chosen to be 0.98.

[0129] like Figure 2 As shown, the present invention also discloses a digital reconstruction system of neuron shape, comprising:

[0130] A neuron process path linear interpolation unit, used for inserting a series of equally spaced nodes P0 for describing the center line of the neuron process path into the neuron process path by linear interpolation;

[0131] A neuron protrusion path preliminary correction unit is used to sequentially calculate the pixel node P1 with the strongest signal intensity in the local range with P0 as the center point, and use P1 as the correction point of P0;

[0132] A neuron process path fine correction unit is used to perform cyclic inspection and correction on the node P1 through angle constraints and signal strength constraints to obtain a finely corrected node P2, connect each node P2, and obtain a finely corrected neuron process path;

[0133] A path node orthogonal plane extraction unit is used to extract the orthogonal planes of each node on the neuron process path;

[0134] A path node orthogonal plane segmentation unit is used to segment each orthogonal plane and extract the foreground signal;

[0135] A radius estimation unit, used for estimating the radius of each orthogonal plane;

[0136] The loop judgment unit is used to loop judge the accuracy of the calculation result and then output the calculation result.

[0137] There are many implementation methods of the present invention, and all technical solutions formed by equivalent transformation or equivalent transformation fall within the protection scope of the present invention.

Claims

1. A method for digitally reconstructing neuron shape, characterized in that: The following steps are involved: S1: insert a series of equally spaced nodes P0 in the neuron process path to describe the center line of the neuron process path by linear interpolation; S2: sequentially calculate the pixel node P1 with the strongest signal strength in the local range with P0 as the center point, and use P1 as the correction point of P0; S3: cyclically checking and correcting the node P1 through angle constraints and signal strength constraints to obtain a finely corrected node P2, and connecting each node P2 to obtain a finely corrected neuron protrusion path; S4: Extract the orthogonal planes of each node on the neuron process path; S5: Segment each orthogonal plane and extract the foreground signal; S6: Estimate the radius of each orthogonal plane; S7: After cyclically judging the accuracy of the calculation results, the calculation results are output; Wherein, the step S4 specifically includes: S41: Based on the relationship between node P2 and its parent node P f , child node P c The position relationship is obtained, and the arc C1 is obtained; S42: According to arc C1, node P2, P f and P c The topological relationship is used to calculate the tangent vector V1 of the arc C1 at the node P2; S43: obtaining a plane A1 with V1 as a normal vector through the tangent vector V1, wherein A1 is an orthogonal plane of the neuron process path node P2; The step S5 specifically includes: S51: extract all pixel points on plane A1 to obtain an orthogonal plane image of the neuron process path node P2; S52: enhancing the gap between the foreground signal and the background signal of the neuron process in the orthogonal plane image; S53: Binarizing the pixels on the orthogonal plane image to extract foreground signals; S54: Separate the interference signal by using the opening operation of image morphology; S55: Use connected domain analysis to mark the orthogonal plane A2 after separating the interference signal.

2. The method for digitally reconstructing neuron shape according to claim 1, characterized in that: In step S2, the convolution of all pixel points in the local range with P0 as the center point and the Gaussian kernel function is determined in the three-axis directions respectively, so as to obtain the three-axis coordinates of the pixel node P1 with the strongest signal intensity in the local range with P0 as the center point. The calculation formula of P1 coordinates is: （1，1） Among them, Gau represents the Gaussian convolution kernel, and G(P0) is the function of P0 on the three-dimensional coordinate axis.

3. The method for digitally reconstructing neuron shape according to claim 1, characterized in that: The angle constraint condition is: the angle formed by the neuron process path node P1 and the connecting line between its parent and child nodes is greater than or equal to 120°; the signal strength constraint condition is: judging that the signal strength of the neuron process path node P1 is greater than the signal strength of its surrounding neighborhood and its parent and child nodes.

4. The method for digitally reconstructing neuron shape according to claim 3, characterized in that: In step S3, if the node P1 meets both the angle constraint and the signal strength constraint, the result is retained; if the node P1 does not meet one or both of the conditions, steps S2 and S3 are repeated.

5. The method for digitally reconstructing neuron shape according to claim 1, characterized in that: The step S6 comprises: S61: Calculate the number N2 of pixels of the foreground signal in the orthogonal plane A2; S62: Calculate the area S2 occupied by all pixels in the orthogonal plane A2 using the image resolution and the number of pixels N2; S63: Obtain the radius r of the orthogonal plane A2 using a circular area calculation formula.

6. The method for digitally reconstructing neuron shape according to claim 1, characterized in that: In step S7, the accuracy of the calculation result is judged cyclically. If the calculation result is accurate, the calculation result is output; if the calculation result is inaccurate, steps S2 to S6 are repeated until the calculation result is judged to be accurate and then output.

7. A digital reconstruction system of neuron shape, characterized by: include: A neuron process path linear interpolation unit, used for inserting a series of equally spaced nodes P0 for describing the center line of the neuron process path into the neuron process path by linear interpolation; A neuron protrusion path preliminary correction unit is used to sequentially calculate the pixel node P1 with the strongest signal intensity in the local range with P0 as the center point, and use P1 as the correction point of P0; A neuron process path fine correction unit is used to perform cyclic inspection and correction on the node P1 through angle constraints and signal strength constraints to obtain a finely corrected node P2, connect each node P2, and obtain a finely corrected neuron process path; A path node orthogonal plane extraction unit is used to extract the orthogonal planes of each node on the neuron process path; A path node orthogonal plane segmentation unit is used to segment each orthogonal plane and extract the foreground signal; A radius estimation unit, used for estimating the radius of each orthogonal plane; A loop judgment unit, used for loop judgment of the accuracy of the calculation result and then outputting the calculation result; The orthogonal plane extraction unit for extracting the orthogonal planes of each node on the neuron process path specifically includes: According to the node P2 and its parent node P f , child node P c The position relationship is obtained, and the arc C1 is obtained; According to arc C1, node P2, P f and P c The topological relationship is used to calculate the tangent vector V1 of the arc C1 at the node P2; A plane A1 with V1 as the normal vector is obtained through the tangent vector V1, wherein A1 is an orthogonal plane of the neuron process path node P2; The path node orthogonal plane segmentation unit segments each orthogonal plane to extract the foreground signal specifically includes: Extract all the pixels on plane A1 to obtain the orthogonal plane image of the neuron process path node P2; enhancing the gap between the foreground signal and the background signal of the neuron process in the orthogonal plane image; Binarizing the pixels on the orthogonal plane image to extract foreground signals; Use the opening operation of image morphology to separate the interference signal; Using connected component analysis, mark the orthogonal plane A2 after separating the interference signal.

Images