Rat burn wound sepsis model method

Abstract

The invention discloses a rat burn wound sepsis model method, and relates to the technical field of image processing, and the method comprises the steps: collecting burn wound maps of a rat at different time and multiple angles through a camera, and collecting cell tissues of the rat according to the corresponding time; carrying out image processing on the burn wound map, and respectively processing the cell tissues according to biological experiment steps; corresponding pathological feature vectors to feature vectors in the wound texture type feature set to obtain a correlation vector set; and training the burn wound sepsis prediction model, and analyzing according to the burn wound sepsis prediction model to obtain pathological trend change according to wound change. According to the method, the field of image processing and the field of infection experiments are combined, a burn wound map and rat body functions are combined, a burn wound sepsis prediction model about a time sequence is constructed according to the relevance of the burn wound map and the rat body functions, and rat body cell changes can be predicted according to time or burn wounds at the same time, namely, the pathological trend of rats is predicted.

Description

Technical Field

[0001] The invention relates to the technical field of image processing, and in particular to a rat burn wound sepsis model method. Background Art

[0002] The burn wound sepsis model is an experimental animal model used to study the development of sepsis after burn-induced wound infection. Sepsis is one of the common fatal complications in burn patients. Especially after wound infection, bacteria enter the blood circulation and trigger a systemic inflammatory response, which may lead to organ failure or even death. Therefore, it is crucial to study the pathogenesis, clinical manifestations, early prediction and intervention measures of burn wound sepsis.

[0003] At present, the Chinese invention patent with application number CN201710168989.0 discloses a method for constructing a mouse burn sepsis model. The sepsis model prepared by using a small amount of standard Pseudomonas aeruginosa strains can be standardized and simulates the pathological and physiological process of burn sepsis more closely than other models. However, the existing technology cannot combine image processing and infection experiments, cannot combine burn wound images with rat body functions, cannot construct a burn wound sepsis prediction model for time series based on the correlation between the two, and cannot simultaneously predict rat body cell changes based on time or burn wounds, that is, predict rat pathological trends. Summary of the invention

[0004] The technical problem solved by the present invention is that the existing technology cannot combine image processing and infection experiments, cannot combine burn wound images with rat body functions, cannot construct a burn wound sepsis prediction model based on a time series according to the correlation between the two, and cannot predict rat body cell changes based on time or burn wounds at the same time, that is, predict the pathological trend of rats.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions: a rat burn wound sepsis model method, comprising the following steps:

[0006] Step S1: using a camera to collect burn wound images of rats at different times and angles, and collecting cell tissues of rats according to corresponding times;

[0007] Step S2: performing image processing on the burn wound image to obtain a feature set of wound texture types in a corresponding period, processing the cell tissues respectively according to biological experimental steps, and obtaining corresponding pathological trends according to time;

[0008] Step S3: extracting the pathological feature vector set of the pathological trend through a machine learning model according to time, respectively corresponding the pathological feature vectors in the pathological feature vector set to the feature vectors in the wound surface texture type feature set to form a two-dimensional matrix, and obtaining the association vector set between the burn wound surface image and the pathological trend;

[0009] Step S4: using the association vector set to train a burn wound sepsis prediction model, and analyzing the burn wound sepsis prediction model to obtain pathological trend changes according to wound changes.

[0010] Preferably, the cell tissues include wound skin cell tissues, liver cell tissues, lung cell tissues, lymphocyte tissues, kidney cell tissues, heart cell tissues and blood cells of rats;

[0011] The time is expressed in days.

[0012] Preferably, step S2 includes the following sub-steps:

[0013] Step S21: reconstructing a multi-view 3D point cloud through image segmentation and 3D reconstruction;

[0014] Step S22: Calculate the area of ​​the three-dimensional point cloud using triangulation;

[0015] Step S23: obtaining the surface areas of different types of wound textures through the three-dimensional point cloud areas;

[0016] Step S24: using the SIFT algorithm to obtain feature quantities corresponding to the wound surface texture types from the surface areas, and constructing a wound surface texture type feature set by combining the feature quantities and the corresponding wound surface texture types in a set form;

[0017] Step S25: extracting the pathological data of the cell tissue using the experimental steps corresponding to the cell tissue, and drawing the pathological change trend graphs corresponding to the cell tissue with time as the horizontal axis.

[0018] Preferably, the step S21 includes:

[0019] The burn wound image is input into a Mean shift image segmentation algorithm based on six-dimensional input, wherein the Mean shift image segmentation algorithm based on six-dimensional input comprises:

[0020] The HSV color model is used to adjust the color saturation and correction value of the burn wound image, and each pixel of the burn wound image is represented as a six-dimensional vector. The mathematical expression of the six-dimensional vector is:

[0021] [H(x,y),S(x,y),V(x,y),x,y,δ]

[0022] Among them, H(x,y) represents hue, S(x,y) represents saturation, V(x,y) represents brightness, x and y represent pixel positions, and δ represents texture features;

[0023] Randomly select pixel center points in the six-dimensional space corresponding to the six-dimensional vector, and calculate the weighted offset mean value in the neighborhood of each pixel point σ respectively, where σ represents a rational constant, and the logic of calculating the weighted offset mean value includes: calculating a weighted average distance according to the position distances of adjacent pixel points in the six-dimensional space, the weight is obtained by calculating the Gaussian kernel of the position distance, iteratively calculating the weighted average distance until a predetermined number of iterations is reached, and when the weighted offset means of all pixel points converge, the pixels located at the same cluster center point are classified into the same category, the cluster is the image segmentation area of ​​the burn wound image, and the contour area of ​​the burn wound is obtained;

[0024] The burn wound image after image segmentation is converted into three-dimensional point cloud data using the multi-view geometric reconstruction technology SfM, and the three-dimensional point cloud data is stereo reconstructed and filtered to generate a three-dimensional point cloud model of the wound area.

[0025] Preferably, the step S22 includes:

[0026] Converting the three-dimensional point cloud model into a mesh composed of a plurality of triangular facets by using a Delaunay triangulation algorithm, wherein the triangular facets are obtained by linking adjacent points in the three-dimensional point cloud model with triangles;

[0027] According to the vertex coordinates of the triangular facets, the area of ​​each triangular facet in the grid is calculated using Heron's formula, and the areas of the triangular facets are added together to obtain the total surface area of ​​the burn wound region.

[0028] Preferably, the step S23 comprises:

[0029] According to the three-dimensional reconstruction results of the wound surface, the wound surface texture is classified. Different wound surface textures will be different in the three-dimensional point cloud. The burn wound surface texture is classified by CNN convolutional neural network. Based on the three-dimensional point cloud model and grid, the surface area of ​​each wound surface texture type is calculated respectively. The wound surface texture type characteristics correspond to granulation, epithelialization, bone, tendon, blood vessel, molting, eschar, n1, n2, ... and n m , the n1, n2, ... and n m They respectively represent a characteristic of wound texture type.

[0030] Preferably, the step S25 comprises:

[0031] Extracting the pathological data of the wound skin cell tissue includes: obtaining the skin cell proliferation area through tissue sectioning and immunohistochemical staining;

[0032] Extracting the pathological data of the liver cell tissue includes: obtaining liver function markers ALT and AST through liver tissue section detection;

[0033] Extracting the pathological data of the lung cell tissue includes: analyzing the immune response of lung endothelial cells by lung sectioning and immunohistochemistry to obtain indicators of lung endothelial cell proliferation and apoptosis;

[0034] Extracting the pathological data of the lymphocyte tissue includes: detecting the proportion of T lymphocytes and corresponding subpopulations in the total number of cells by flow cytometer;

[0035] Extracting the pathological data of the renal cell tissue includes: obtaining serum urea nitrogen and creatinine content through renal sections and experiments;

[0036] Extracting the pathological data of the cardiac cell tissue includes: obtaining the myocardial marker BNP through cardiac slices;

[0037] Extracting the pathological data of the blood cells includes: collecting a blood sample aseptically from the apex of the heart by a disposable syringe to obtain a complete blood cell count and changes in the complete blood cell count;

[0038] All pathological data were standardized to obtain pathological indicators in unified units, and the Matplotlib data visualization tool was used to draw a pathological change trend graph with days as the horizontal axis and various pathological indicators as the vertical axis.

[0039] Preferably, step S3 includes the following sub-steps:

[0040] Step S31: using the CNN convolutional neural network to extract pathological feature vectors from the time series data of each cell tissue respectively, to obtain pathological feature vectors, wherein the CNN convolutional neural network uses the same model parameters as those used in extracting the characteristics of the burn wound texture when extracting the pathological feature vectors, and the model parameters are the number of convolution kernels and the dimension of each convolution kernel;

[0041] Step S32: merging the pathological feature vectors at all time points to form a time series pathological feature vector set, wherein the pathological feature vector set is a collection of pathological feature vectors of all tissues at different time points;

[0042] Step S33: adjusting model parameters of several layers of the CNN convolutional neural network, and using the CNN convolutional neural network after adjusting the model parameters to secondary extract wound texture features and pathological features to obtain a deep wound texture feature vector and a deep pathological feature vector;

[0043] Step S34: using the Pearson correlation coefficient to calculate the correlation between the deep pathology feature vector and the deep wound texture feature vector, using the principal component analysis PCA matrix decomposition method to extract the potential relationship value between the deep pathology feature vector and the deep wound texture feature vector, using the potential relationship value to replace the corresponding correlation to obtain a new feature space, which is a correlation matrix;

[0044] Step S35: according to the correlation matrix, a correlation threshold is set, and the deep pathology feature vector and the deep wound texture feature vector whose correlation or potential relationship value is higher than the correlation threshold are input into the support vector regression SVR for mapping function matching, so as to align the dimension of the deep pathology feature vector and the deep wound texture feature vector;

[0045] Step S36: The pathological feature vectors in the pathological feature vector set are respectively matched with the feature vectors in the wound surface texture type feature set to form a two-dimensional matrix, and all two-dimensional matrices are integrated into a correlation vector set of the burn wound image and the pathological trend.

[0046] Preferably, the step S33 includes:

[0047] The last classification layer of the CNN convolutional neural network is removed, a new linear regression output layer is added, several low-level feature extraction layers of the CNN convolutional neural network are frozen, the convolutional layer of the CNN convolutional neural network is unfrozen, the model parameters of the convolutional layer are adjusted using an adaptive dynamic learning rate adjustment method, a batch normalization layer is added in front of the linear regression output layer, the batch normalization layer is used to stabilize the training process, the mean square error MSE is used as the loss function of the CNN convolutional neural network, a training cycle threshold is set, the wound texture feature vector and the pathological feature vector set are used as training sets, the CNN convolutional neural network is iteratively trained until the number of training times reaches the training cycle threshold, and the CNN convolutional neural network after adjusting the model parameters is obtained.

[0048] Preferably, the step S4 comprises:

[0049] The association vector set is input into a long short-term memory network as a training set to obtain a burn wound sepsis prediction model, and the pathological trend change is measured in time.

[0050] The beneficial effects of the present invention are as follows: by combining the fields of image processing and infection experiments, burn wound images and rat body functions are combined, and a burn wound sepsis prediction model based on a time series is constructed according to the correlation between the two. This model can simultaneously predict changes in rat body cells based on time or burn wounds, that is, predict the pathological trend of rats. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 A schematic diagram of the basic process of a rat burn wound sepsis model method provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0052] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, but not all of the embodiments.

[0053] Reference Figure 1 , as one embodiment of the present invention, provides a rat burn wound sepsis model method, comprising the following steps:

[0054] Step S1: using a camera to collect burn wound images of rats at different times and angles, and collecting cell tissues of rats according to corresponding times;

[0055] Step S2: performing image processing on the burn wound image to obtain a feature set of wound texture types in a corresponding period, processing the cell tissues respectively according to biological experimental steps, and obtaining corresponding pathological trends according to time;

[0056] Step S3: extracting the pathological feature vector set of the pathological trend through a machine learning model according to time, respectively corresponding the pathological feature vectors in the pathological feature vector set to the feature vectors in the wound surface texture type feature set to form a two-dimensional matrix, and obtaining the association vector set between the burn wound surface image and the pathological trend;

[0057] Step S4: using the association vector set to train a burn wound sepsis prediction model, and analyzing the burn wound sepsis prediction model to obtain pathological trend changes according to wound changes.

[0058] By combining the burn wound images and the body functions of rats, a burn wound sepsis prediction model based on the time series is constructed according to the correlation between the two, which can predict the changes in rat body cells based on time or burn wounds at the same time.

[0059] The cell tissues include wound skin cell tissues, liver cell tissues, lung cell tissues, lymphocyte tissues, kidney cell tissues, heart cell tissues and blood cells of rats;

[0060] The time is expressed in days.

[0061] The step S2 comprises the following sub-steps:

[0062] Step S21: reconstructing a multi-view 3D point cloud through image segmentation and 3D reconstruction;

[0063] Step S22: Calculate the area of ​​the three-dimensional point cloud using triangulation;

[0064] Step S23: obtaining the surface areas of different types of wound textures through the three-dimensional point cloud areas;

[0065] Step S24: obtaining feature quantities corresponding to wound texture types from the surface areas through the SIFT algorithm, and constructing a wound texture type feature set in the form of a set of the feature quantities and the corresponding wound texture types. The wound texture type feature set effectively represents the texture type of the wound, and the feature quantities extracted by the SIFT algorithm are local and have high stability;

[0066] Step S25: extracting the pathological data of the cell tissue using the experimental steps corresponding to the cell tissue, and drawing the pathological change trend graphs corresponding to the cell tissue with time as the horizontal axis.

[0067] The steps describe a complete process from image segmentation, 3D reconstruction, to pathological data extraction and trend analysis. By extracting wound texture feature vectors through 3D point cloud technology and neural network and combining with the pathological trend data of cell tissue, detailed quantitative analysis and prediction can be provided for the wound healing process.

[0068] The step S21 comprises:

[0069] The burn wound image is input into a Mean shift image segmentation algorithm based on six-dimensional input, wherein the Mean shift image segmentation algorithm based on six-dimensional input comprises:

[0070] The HSV color model is used to adjust the color saturation and correction value of the burn wound image, and each pixel of the burn wound image is represented as a six-dimensional vector. The mathematical expression of the six-dimensional vector is:

[0071] [H(x,y),S(x,y),V(x,y),x,y,δ]

[0072] Among them, H(x,y) represents hue, S(x,y) represents saturation, V(x,y) represents brightness, x and y represent pixel positions, and δ represents texture features;

[0073] Randomly select pixel center points in the six-dimensional space corresponding to the six-dimensional vector, and calculate the weighted offset mean in the neighborhood of each pixel point σ, where σ represents a rational constant. The logic of calculating the weighted offset mean includes: calculating a weighted average distance according to the position distances of adjacent pixel points in the six-dimensional space, the weight is obtained by calculating the Gaussian kernel of the position distance, iteratively calculating the weighted average distance until a predetermined number of iterations is reached, and when the weighted offset means of all pixel points converge, the pixels located at the center point of the same cluster are classified into the same category, the cluster is an image segmentation area of ​​the burn wound image, the contour area of ​​the burn wound is obtained, and the background is eliminated;

[0074] The burn wound image after image segmentation is converted into three-dimensional point cloud data using the multi-view geometric reconstruction technology SfM, and the three-dimensional point cloud data is stereo reconstructed and filtered to generate a three-dimensional point cloud model of the wound area.

[0075] The step S22 comprises:

[0076] Converting the three-dimensional point cloud model into a mesh composed of a plurality of triangular facets by using a Delaunay triangulation algorithm, wherein the triangular facets are obtained by linking adjacent points in the three-dimensional point cloud model with triangles;

[0077] According to the vertex coordinates of the triangular facets, the area of ​​each triangular facet in the grid is calculated using Heron's formula, and the areas of the triangular facets are added together to obtain the total surface area of ​​the burn wound region.

[0078] The step S23 comprises:

[0079] According to the three-dimensional reconstruction results of the wound surface, the wound surface texture is classified. Different wound surface textures will be different in the three-dimensional point cloud. The burn wound surface texture is classified by CNN convolutional neural network. Through multi-layer convolution and pooling operations, CNN convolutional neural network can effectively capture the characteristics of the burn wound surface. Based on the three-dimensional point cloud model and grid, the surface area of ​​each wound surface texture type is calculated respectively. The wound surface texture type characteristics correspond to granulation, epithelialization, bone, tendon, blood vessel, molting, eschar, n1, n2, ... and n m , the n1, n2, ... and n m They respectively represent a characteristic of wound texture type.

[0080] CNN convolutional neural network uses convolution kernels to extract local information and realizes learning and classification of high-level features through fully connected layers, thereby achieving accurate detection of wound textures of different sizes and shapes.

[0081] The step S25 comprises:

[0082] Extracting the pathological data of the wound skin cell tissue includes: obtaining the skin cell proliferation area through tissue sectioning and immunohistochemical staining;

[0083] Extracting the pathological data of the liver cell tissue includes: obtaining liver function markers ALT and AST through liver tissue section detection;

[0084] Extracting the pathological data of the lung cell tissue includes: analyzing the immune response of spleen and lung endothelial cells by lung sectioning and immunohistochemistry to obtain the proliferation and apoptosis indexes of lung endothelial cells;

[0085] Extracting the pathological data of the lymphocyte tissue includes: detecting the proportion of T lymphocytes and corresponding subpopulations in the total number of cells by flow cytometer;

[0086] Extracting the pathological data of the renal cell tissue includes: obtaining serum urea nitrogen and creatinine content through renal sections and experiments;

[0087] Extracting the pathological data of the cardiac cell tissue includes: obtaining the myocardial marker BNP as pathological data through cardiac slices to detect the degree of damage to the cardiac tissue;

[0088] Extracting the pathological data of the blood cells includes: collecting blood samples from the apex of the heart aseptically by using a disposable syringe to obtain a complete blood cell count and changes in the complete blood cell count, reflecting the immune status and post-traumatic response of the whole body;

[0089] Standardize all pathological data to obtain pathological indicators of unified units to ensure that different types of data can be compared;

[0090] The Matplotlib data visualization tool was used to draw a pathological change trend graph with days as the horizontal axis and various pathological indicators as the vertical axis.

[0091] The pathological change trend diagram is used to understand the repair process of different tissues after trauma, as well as the relationship between immune response and wound healing, providing strong support for research on burn wound healing, post-traumatic immune response, cell repair mechanism, etc.

[0092] The step S3 comprises the following sub-steps:

[0093] Step S31: using the CNN convolutional neural network to extract pathological feature vectors from the time series data of each cell tissue respectively, to obtain pathological feature vectors, wherein the CNN convolutional neural network uses the same model parameters as those used in extracting the characteristics of the burn wound texture when extracting the pathological feature vectors, and the model parameters are the number of convolution kernels and the dimension of each convolution kernel;

[0094] Step S32: merging the pathological feature vectors at all time points to form a time series pathological feature vector set, wherein the pathological feature vector set is a collection of pathological feature vectors of all tissues at different time points;

[0095] Step S33: adjusting model parameters of several layers of the CNN convolutional neural network, and using the CNN convolutional neural network after adjusting the model parameters to secondary extract wound texture features and pathological features to obtain a deep wound texture feature vector and a deep pathological feature vector;

[0096] Step S34: using the Pearson correlation coefficient to calculate the correlation between the deep pathology feature vector and the deep wound texture feature vector, using the principal component analysis PCA matrix decomposition method to extract the potential relationship value between the deep pathology feature vector and the deep wound texture feature vector, using the potential relationship value to replace the corresponding correlation to obtain a new feature space, which is a correlation matrix;

[0097] Step S35: According to the association matrix, a correlation threshold is manually set, and the deep pathology feature vector and the deep wound texture feature vector whose correlation or potential relationship value is higher than the correlation threshold are input into the support vector regression SVR for mapping function matching, and the deep pathology feature vector and the deep wound texture feature vector are dimensionally aligned; since there may be data scarcity or feature space differences in these two fields, transfer learning can make up for the lack of data in the target field to a certain extent by migrating existing features.

[0098] Step S36: The pathological feature vectors in the pathological feature vector set are respectively matched with the feature vectors in the wound surface texture type feature set to form a two-dimensional matrix, and all two-dimensional matrices are integrated into a correlation vector set of the burn wound image and the pathological trend.

[0099] The step S33 comprises:

[0100] The last classification layer of the CNN convolutional neural network is removed, a new linear regression output layer is added, several low-level feature extraction layers of the CNN convolutional neural network are frozen, the convolutional layer of the CNN convolutional neural network is unfrozen, the model parameters of the convolutional layer are adjusted using an adaptive dynamic learning rate adjustment method, the weight should be kept unchanged when adjusting the convolutional layer, a batch normalization layer is added in front of the linear regression output layer, the batch normalization layer is used to stabilize the training process, it helps to reduce the gradient vanishing problem and accelerate the training process by normalizing the input of each layer, the mean square error MSE is used as the loss function of the CNN convolutional neural network, a training cycle threshold is set, the wound texture feature vector and the pathological feature vector set are used as training sets, the CNN convolutional neural network is iteratively trained until the number of training times reaches the training cycle threshold, the loss and accuracy changes of the CNN convolutional neural network are observed during the training process, the number of training times is manually adjusted, and the CNN convolutional neural network after the model parameters are adjusted is obtained, and the CNN convolutional neural network after the parameters are adjusted can simultaneously adapt to the extraction of the wound texture features and pathological features.

[0101] Through the above steps, combined with transfer learning and adjustment of CNN model parameters, good results can be effectively achieved in the automatic adaptation of wound texture features. Through multi-level adjustments, transfer learning can help the model bridge the differences between pathological feature vectors and wound texture feature vectors, allowing them to be better aligned in high-dimensional feature space.

[0102] The step S4 comprises:

[0103] The association vector set is input into a long short-term memory network as a training set to obtain a burn wound sepsis prediction model, and the pathological trend change is measured in time.

[0104] As time goes by, the burn wound sepsis prediction model will predict the changes in the burn wound and the risk of sepsis. It can analyze the pathological changes in different time windows, and judge whether the condition has worsened through long-term trends (such as the changing trends in the past week or the past three days), thereby warning of the occurrence of sepsis.

[0105] The present invention combines the fields of image processing and infection experiments, combines burn wound images with rat body functions, and constructs a burn wound sepsis prediction model based on a time series according to the correlation between the two. It can simultaneously predict changes in rat body cells based on time or burn wounds, that is, predict the pathological trend of rats.

[0106] It should be understood by those skilled in the art that the embodiments of the present invention can be provided as methods, systems or computer program products. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program codes. Among them, the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, referred to as EPROM), programmable read-only memory (Programmable Red-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic memory, flash memory, disk or optical disk. These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0107] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A rat burn wound sepsis model method, characterized in that: The following steps are involved: Step S1: using a camera to collect burn wound images of rats at different times and angles, and collecting cell tissues of rats according to corresponding times; Step S2: performing image processing on the burn wound image to obtain a feature set of wound texture types in a corresponding period, processing the cell tissues respectively according to biological experimental steps, and obtaining corresponding pathological trends according to time; Step S3: extracting the pathological feature vector set of the pathological trend through a machine learning model according to time, respectively corresponding the pathological feature vectors in the pathological feature vector set to the feature vectors in the wound surface texture type feature set to form a two-dimensional matrix, and obtaining the association vector set between the burn wound surface image and the pathological trend; Step S4: using the association vector set to train a burn wound sepsis prediction model, and analyzing the burn wound sepsis prediction model to obtain pathological trend changes according to wound changes.

2. The rat burn wound sepsis model method according to claim 1, characterized in that: The cell tissues include wound skin cell tissues, liver cell tissues, lung cell tissues, lymphocyte tissues, kidney cell tissues, heart cell tissues and blood cells of rats; The time is expressed in days.

3. The rat burn wound sepsis model method according to claim 2, characterized in that: The step S2 comprises the following sub-steps: Step S21: reconstructing a multi-view 3D point cloud through image segmentation and 3D reconstruction; Step S22: Calculate the area of ​​the three-dimensional point cloud using triangulation; Step S23: obtaining the surface areas of different types of wound textures through the three-dimensional point cloud areas; Step S24: using the SIFT algorithm to obtain feature quantities corresponding to the wound surface texture types from the surface areas, and constructing a wound surface texture type feature set by combining the feature quantities and the corresponding wound surface texture types in a set form; Step S25: extracting the pathological data of the cell tissue using the experimental steps corresponding to the cell tissue, and drawing the pathological change trend graphs corresponding to the cell tissue with time as the horizontal axis.

4. The rat burn wound sepsis model method according to claim 3, characterized in that: The step S21 comprises: The burn wound image is input into a Mean shift image segmentation algorithm based on six-dimensional input, wherein the Mean shift image segmentation algorithm based on six-dimensional input comprises: The HSV color model is used to adjust the color saturation and correction value of the burn wound image, and each pixel of the burn wound image is represented as a six-dimensional vector. The mathematical expression of the six-dimensional vector is: [H(x,y),S(x,y),V(x,y),x,y,δ] Among them, H(x,y) represents hue, S(x,y) represents saturation, V(x,y) represents brightness, x and y represent pixel positions, and δ represents texture features; Randomly select pixel center points in the six-dimensional space corresponding to the six-dimensional vector, and calculate the weighted offset mean value in the neighborhood of each pixel point σ respectively, where σ represents a rational constant, and the logic of calculating the weighted offset mean value includes: calculating a weighted average distance according to the position distances of adjacent pixel points in the six-dimensional space, the weight is obtained by calculating the Gaussian kernel of the position distance, iteratively calculating the weighted average distance until a predetermined number of iterations is reached, and when the weighted offset means of all pixel points converge, the pixels located at the same cluster center point are classified into the same category, the cluster is the image segmentation area of ​​the burn wound image, and the contour area of ​​the burn wound is obtained; The burn wound image after image segmentation is converted into three-dimensional point cloud data using the multi-view geometric reconstruction technology SfM, and the three-dimensional point cloud data is stereo reconstructed and filtered to generate a three-dimensional point cloud model of the wound area.

5. The rat burn wound sepsis model method according to claim 4, characterized in that: The step S22 comprises: Converting the three-dimensional point cloud model into a mesh composed of a plurality of triangular facets by using a Delaunay triangulation algorithm, wherein the triangular facets are obtained by linking adjacent points in the three-dimensional point cloud model with triangles; According to the vertex coordinates of the triangular facets, the area of ​​each triangular facet in the grid is calculated using Heron's formula, and the areas of the triangular facets are added together to obtain the total surface area of ​​the burn wound region.

6. The rat burn wound sepsis model method according to claim 5, characterized in that: The step S23 comprises: According to the three-dimensional reconstruction results of the wound surface, the wound surface texture is classified. Different wound surface textures will be different in the three-dimensional point cloud. The burn wound surface texture is classified by CNN convolutional neural network. Based on the three-dimensional point cloud model and grid, the surface area of ​​each wound surface texture type is calculated respectively. The wound surface texture type characteristics correspond to granulation, epithelialization, bone, tendon, blood vessel, molting, eschar, n1, n2, ... and n m , the n1, n2, ... and n m They respectively represent a characteristic of wound texture type.

7. The rat burn wound sepsis model method according to claim 6, characterized in that: The step S25 comprises: Extracting the pathological data of the wound skin cell tissue includes: obtaining the skin cell proliferation area through tissue sectioning and immunohistochemical staining; Extracting the pathological data of the liver cell tissue includes: obtaining liver function markers ALT and AST through liver tissue section detection; Extracting the pathological data of the lung cell tissue includes: analyzing the immune response of spleen cells through lung sections and immunohistochemistry to obtain lung cell proliferation and apoptosis indicators; Extracting the pathological data of the lymphocyte tissue includes: detecting the proportion of T lymphocytes and corresponding subpopulations in the total number of cells by flow cytometer; Extracting the pathological data of the renal cell tissue includes: obtaining serum urea nitrogen and creatinine content through renal sections and experiments; Extracting the pathological data of the cardiac cell tissue includes: obtaining the myocardial marker BNP through cardiac slices; Extracting the pathological data of the blood cells includes: collecting a blood sample aseptically from the apex of the heart by a disposable syringe to obtain a complete blood cell count and changes in the complete blood cell count; All pathological data are standardized to obtain pathological indicators in a unified unit; The Matplotlib data visualization tool was used to draw a pathological change trend graph with days as the horizontal axis and various pathological indicators as the vertical axis.

8. The rat burn wound sepsis model method according to claim 7, characterized in that: The step S3 comprises the following sub-steps: Step S31: using the CNN convolutional neural network to extract pathological feature vectors from the time series data of each cell tissue respectively, to obtain pathological feature vectors, wherein the CNN convolutional neural network uses the same model parameters as those used in extracting the characteristics of the burn wound texture when extracting the pathological feature vectors, and the model parameters are the number of convolution kernels and the dimension of each convolution kernel; Step S32: merging the pathological feature vectors at all time points to form a time series pathological feature vector set, wherein the pathological feature vector set is a collection of pathological feature vectors of all tissues at different time points; Step S33: adjusting model parameters of several layers of the CNN convolutional neural network, and using the CNN convolutional neural network after adjusting the model parameters to secondary extract wound texture features and pathological features to obtain a deep wound texture feature vector and a deep pathological feature vector; Step S34: using the Pearson correlation coefficient to calculate the correlation between the deep pathology feature vector and the deep wound texture feature vector, using the principal component analysis PCA matrix decomposition method to extract the potential relationship value between the deep pathology feature vector and the deep wound texture feature vector, using the potential relationship value to replace the corresponding correlation to obtain a new feature space, which is a correlation matrix; Step S35: according to the correlation matrix, a correlation threshold is set, and the deep pathology feature vector and the deep wound texture feature vector whose correlation or potential relationship value is higher than the correlation threshold are input into the support vector regression SVR for mapping function matching, so as to align the dimension of the deep pathology feature vector and the deep wound texture feature vector; Step S36: The pathological feature vectors in the pathological feature vector set are respectively matched with the feature vectors in the wound surface texture type feature set to form a two-dimensional matrix, and all two-dimensional matrices are integrated into a correlation vector set of the burn wound image and the pathological trend.

9. The rat burn wound sepsis model method according to claim 8, characterized in that: The step S33 comprises: The last classification layer of the CNN convolutional neural network is removed, a new linear regression output layer is added, several low-level feature extraction layers of the CNN convolutional neural network are frozen, the convolutional layer of the CNN convolutional neural network is unfrozen, the model parameters of the convolutional layer are adjusted using an adaptive dynamic learning rate adjustment method, a batch normalization layer is added in front of the linear regression output layer, the batch normalization layer is used to stabilize the training process, the mean square error MSE is used as the loss function of the CNN convolutional neural network, a training cycle threshold is set, the wound texture feature vector and the pathological feature vector set are used as training sets, the CNN convolutional neural network is iteratively trained until the number of training times reaches the training cycle threshold, and the CNN convolutional neural network after adjusting the model parameters is obtained.

10. The rat burn wound sepsis model method according to claim 9, characterized in that: The step S4 comprises: The association vector set is input into a long short-term memory network as a training set to obtain a burn wound sepsis prediction model, and the pathological trend change is measured in time.

Images