A method and system for classifying and identifying myofibroblast images based on ATPase staining

Abstract

The present invention provides a method and system for classifying and identifying myofibroblast images based on ATPase staining. The method collects myofibroblast tissue that meets the requirements of morphological integrity and then processes it according to a set freezing operation, and then performs ATPase staining on myofibroblast tissue sections as myofibroblast images for identification. The type I cells in the myofibroblast images are extracted with the help of a pre-built basic classification model, and the cell images after excluding the interference of type I cells are automatically quality controlled. The connection relationship between cells is segmented and the interference of ice crystals is eliminated, and then the type II cells are refined and classified by combining color space description technology and sample clustering methods. The above scheme overcomes the problems of complicated operation and insufficient accuracy of existing myofibroblast identification technology. After a small amount of training samples are used to build a classification model, automatic basic classification can be achieved. In combination with cell image processing and clustering operation recognition, the accuracy and efficiency of myofibroblast identification are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of cell image optimization processing, and in particular to a myofibroblast image classification and recognition method and system based on ATPase staining. Background Art

[0002] With the continuous improvement of people's living standards and their higher yearning for a better life, the majority of people are no longer satisfied with having rice and vegetables to eat, but pursue healthier, more delicious and higher nutritional value of high-quality food, such as livestock meat. At present, pork accounts for the largest part of livestock meat consumption, up to more than 60%. To obtain a healthier and more delicious pork, it is necessary to conduct a comprehensive and accurate determination of its meat phenotype. The main traits for measuring meat phenotypes include pH value, dripping, meat color, marbling, and muscle fiber type and number in muscle slices.

[0003] However, in order to clearly identify the types and numbers of livestock meat muscle fibers, it is necessary to stain the muscle cell sections and observe and count them. The existing technology mostly collects sample sections and places them under a microscope for observation and counting. In order to improve the accuracy of cell counting under a microscope, each sample needs to be counted three times, and each section needs to be repeated five times in the microscopic field of view under the microscope. Manual counting is extremely time-consuming and labor-intensive, and prone to errors, which greatly limits the development of this field.

[0004] There are some existing technologies that use convolutional neural network technology to realize cell image recognition, but it requires a huge number of cell image samples and combined with theoretical mathematical calculations, it is impossible to achieve accurate cell characteristic characterization, and its reliability needs to be improved.

[0005] The information disclosed in the background technology section of the present invention is only intended to deepen the understanding of the general background technology of the present invention, and should not be regarded as acknowledging or suggesting in any form that the information constitutes the prior art known to those skilled in the art. Summary of the invention

[0006] To solve the above problems, the present invention provides a method for classifying and identifying myofibroblast images based on ATPase staining. In one embodiment, the method comprises:

[0007] Sample collection step: collecting muscle fiber tissue that meets the requirements of morphological integrity, cutting the muscle fiber tissue once according to a set standard, and freezing the cut muscle fiber tissue;

[0008] The sample staining step includes performing a secondary cutting of the frozen muscle fiber tissue in a constant temperature environment, staining the secondary cut muscle fiber tissue using ATPase, and saving the stained image as a muscle fiber sample image;

[0009] In the first-level classification step, all cells and type I cells in the sample image are identified and divided using a basic classification model pre-built based on the random forest algorithm, and a corresponding whole-cell database is constructed based on the identified and divided sample image. Type I cells are eliminated from the whole cells to extract type II cell images, and the result image for type I cell identification is placed in the set type I cell database;

[0010] The automatic quality control step is to first optimize the segmentation of the sample image of type II cells based on the morphological watershed method, and then perform ice crystal recognition and ice crystal screening on the segmented sample image of type II cells to obtain the sample image of type II cells to be calculated;

[0011] In the secondary classification step, for the type II cell sample image to be calculated, the color space description technology and the sample clustering method are combined to refine the type II cells into two categories and mark them.

[0012] Preferably, in one embodiment, the automatic quality control step includes the following operations:

[0013] Step A1, considering the topological theory, using the watershed method to segment the links between cells in the type II cell sample image based on the region;

[0014] Step A2: Use the four-point sampling method to identify ice crystals in the type II cell sample image by region. When the total number of ice crystals exceeds the set condition, use the set rules to divide the ice crystal size and remove the large ice crystals that meet the set size condition.

[0015] Furthermore, in one embodiment, the automatic quality control step further includes before step A2:

[0016] Step B: using a five-point sampling method to calculate the difference in the number of cells in the type II cell sample image by region, which is used to characterize the uniformity of cell distribution in the sample image, and to eliminate sample images that do not meet the uniformity requirements.

[0017] Optionally, in one embodiment, in the sample collection step, the muscle fiber tissue of the longissimus dorsi muscle of the last rib of the organism within 30 minutes of death is collected as a sample.

[0018] Furthermore, in one embodiment, in the sample collection step, the process of freezing the muscle fiber tissue after the first cutting includes:

[0019] The muscle fiber tissue after the first cutting is placed in a new plastic sampling tube with a plurality of uniform small holes on the tube wall for processing;

[0020] In liquid nitrogen, the muscle fiber tissue was frozen in a new porous plastic sampling tube to control the probability of ice crystal formation.

[0021] In a preferred embodiment, in the sample staining step, before staining, cross-section slices of muscle fiber tissue are cut using a low-temperature constant temperature microtome in a temperature environment of -20°C, and the cross-section slices must meet the set size requirements.

[0022] Specifically, in one embodiment, the basic classification model used in the primary classification step is pre-constructed according to the following operations:

[0023] Step C1, dividing the cell image samples of a set scale with basic classification labels into segmentation training samples and segmentation test samples, and establishing a segmentation training database and a segmentation test database respectively;

[0024] Step C2, select sample images and existing classification result data in the training database and input them into the ilastik system, and use the random forest algorithm to iteratively train the cell classification model;

[0025] Step C3, input the sample images in the segmentation test database into the latest model, evaluate the matching degree between the obtained identification results and the existing real classification label results, if the set conditions are not met, update in real time and return to step C2; if the set conditions are met, the final cell classification model is used as the target basic classification model.

[0026] In one embodiment, in the secondary classification step, the image color data inside the cell and at the edge of the cell in the type II sample image are extracted respectively, and the RGB color space characteristics of the cell image are used to calculate the mean of the R value, G value, and B value inside the cell and at the edge of the cell, and then the difference is calculated to characterize the current cell, and then the type II cells are divided into type IIA and type IIB based on the obtained difference using a clustering algorithm.

[0027] Optionally, in one embodiment, in the secondary classification step, image color data of the cell interior and cell edge in the type II sample image are extracted respectively, and the HSV (Hue, Saturation, Value) color space characteristics of the cell image are used to calculate the ratio of the difference between the H value inside the cell and the H value at the cell edge to the S value of the entire cell to characterize the current cell, and then a clustering algorithm is used based on the obtained data to divide the type II cells into type IIA and type IIB.

[0028] Based on other aspects of the method described in any one or more of the above embodiments, the present invention also provides a myofibroblast image classification and recognition system based on ATPase staining, characterized in that the system executes the method described in any one or more of the above embodiments.

[0029] Compared with the closest prior art, the present invention also has the following beneficial effects:

[0030] The present invention provides a method and system for classifying and identifying myofibroblast images based on ATPase staining. The method collects myofibroblast tissue that meets the requirements of morphological integrity and then processes it according to a set freezing operation, and then performs ATPase staining on myofibroblast tissue slices as myofibroblast images for identification. Freezing treatment according to the set strategy can fundamentally control the proportion of ice crystal interference in the sample image. In addition, the myofibroblast slice image stained with ATPase only needs to be stained once, and the error probability is low, which will not lead to sample waste, saving the cost and time of staining.

[0031] Furthermore, the type I cells in the myofibroblast image are extracted with the help of the pre-built basic classification model. The random forest algorithm can achieve good recognition results with only a small sample size of data to establish a reliable basic classification model without the need to preset a large number of training samples. Under the premise of ensuring the reliability of the basic classification model, the time consumption of preparing training samples is saved.

[0032] In addition, the present invention further performs automated quality control processing on the cell images after eliminating the interference of type I cells, segments the connection relationship between cells and eliminates the interference of ice crystals, and then combines color space description technology and sample clustering methods to refine the classification of type II cells. It overcomes the problems of complicated operation and insufficient accuracy of existing myofibroblast identification technology, eliminates the interference of cell links and ice crystals, cooperates with cell image processing and clustering operation recognition, and fully combines unsupervised learning with the color space feature type of cells through clustering algorithms. A quantitative color feature value that can represent each cell type is found. Through such a value, the two types of cells are completely distinguished by clustering. It provides substantial improvements to the technology of myofibroblast image classification, improves the accuracy and efficiency of myofibroblast analysis and identification, and significantly improves the process of manual counting and identification required by scientific researchers.

[0033] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0035] Figure 1 It is a schematic flow chart of a method for classifying and identifying myofibroblast images based on ATPase staining provided in one embodiment of the present invention;

[0036] Figure 2 is a detailed diagram of the implementation principle of a method for classifying and identifying myofibroblasts in another embodiment of the present invention;

[0037] Figure 3 is an example diagram of cell classification results of the myofibroblast image classification and recognition method in an embodiment of the present invention;

[0038] Figure 4 It is a detailed schematic diagram of the execution process of the myofibroblast image classification and recognition method in one embodiment of the present invention;

[0039] Figure 5 It is a detailed schematic diagram of the execution process of the myofibroblast image classification and recognition method provided by another embodiment of the present invention;

[0040] Figure 6 It is a structural schematic diagram of a myofibroblast image classification and recognition system based on ATPase staining provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0041] The following will describe the implementation methods of the present invention in detail in conjunction with the accompanying drawings and embodiments, so that the implementers of the present invention can fully understand how the present invention applies technical means to solve technical problems and achieve the implementation process of technical effects and implement the present invention specifically according to the above implementation process. It should be noted that as long as there is no conflict, the various embodiments and various features of the embodiments in the present invention can be combined with each other, and the technical solutions formed are all within the protection scope of the present invention.

[0042] Although the flowcharts describe the operations as sequential processes, many of the operations may be performed in parallel, concurrently, or simultaneously. The order of the operations may be rearranged. A process may be terminated when its operations are completed, but may also have additional steps not included in the accompanying drawings. A process may correspond to a method, function, procedure, subroutine, subprogram, etc.

[0043] The terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments. Unless the context clearly indicates otherwise, the singular forms "one", "one" and "item" used herein are also intended to include plural numbers. It should also be understood that the terms "include" and / or "comprise" used herein specify the existence of stated features, integers, steps, operations, units and / or components, without excluding the existence or addition of one or more other features, integers, steps, operations, units, components and / or combinations thereof.

[0044] To determine the type and number of livestock meat muscle fibers, it is necessary to stain the muscle cell sections and observe and count them. The existing technology mostly collects sample sections and places them under a microscope for observation and counting. In order to improve the accuracy of cell counting under a microscope, each sample needs to be counted three times, and each section needs to be repeated five times in the microscopic field of view under the microscope. Manual counting is extremely time-consuming and labor-intensive, and prone to errors, which greatly limits the development of this field.

[0045] In addition, although there are some existing technologies that use convolutional neural network technology to achieve cell image recognition, it requires a huge number of cell image samples and combined with theoretical mathematical calculations, it is impossible to achieve accurate cell characteristic characterization, and its reliability needs to be improved.

[0046] In order to solve the above technical problems, the researchers of the present invention considered that the image classification system constructed based on the machine learning method can quickly segment the image area and count the area or other characteristic data of different regions, which can be suitable for the identification and counting of different muscle cells after muscle fiber sectioning. Based on this, the project team designed a series of cell image processing methods and classification and recognition operations, which constitute the muscle fiber cell image classification and recognition scheme based on ATPase staining in the present invention.

[0047] The myofibroblast image classification and recognition scheme of the present invention first prepares ATPase-stained myofibroblast images as training samples, then builds a training model through the random forest method of ilastik software, inputs the training samples into the model for training, stops label input when the training samples are basically accurate, and completes segmentation.

[0048] The cell edge links are further segmented using traditional image processing methods to obtain each individual cell and filter out ice crystals according to the set strategy.

[0049] Finally, the color space feature value of each cell is calculated, and it is divided into type I cells and whole cells through the random forest algorithm. The extracted type II cell images are classified into type IIA and type IIB cells through unsupervised learning methods. This method improves the accuracy and efficiency of cell number recognition calculation in myofibroblast images, optimizes the complicated and time-consuming diagnostic process, and greatly reduces the burden of medical cell recognition research and other work related to myofibroblast research.

[0050] Next, the detailed process of the method of the embodiment of the present invention is described in detail based on the accompanying drawings. The steps shown in the flowchart of the accompanying drawings can be executed in a computer system including a set of computer executable instructions. Although the logical order of each step is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0051] Embodiment 1

[0052] Figure 1 The schematic diagram of the process of the myofibroblast image classification and recognition method based on ATPase staining provided in the first embodiment of the present invention is shown, referring to Figure 1 It can be seen that the method includes the following steps.

[0053] Sample collection step: collecting muscle fiber tissue that meets the requirements of morphological integrity, cutting the muscle fiber tissue once according to a set standard, and freezing the cut muscle fiber tissue;

[0054] The sample staining step includes performing a secondary cutting of the frozen muscle fiber tissue in a specific environment, staining the secondary cut muscle fiber tissue using ATPase, and saving the stained image as a muscle fiber sample image;

[0055] In the first-level classification step, the basic classification model pre-built based on the random forest algorithm is used to identify and divide the type I cells and all cells in the sample image, and the type II cell image is extracted by eliminating the type I cells through the whole cell. The corresponding whole-cell database is constructed based on the extracted sample image, and the sample image and result image for type I cell identification are placed in the set type I cell database;

[0056] The automatic quality control step is to first optimize the segmentation of the sample image of type II cells based on the morphological watershed method, and then perform ice crystal recognition and ice crystal screening on the segmented sample image of type II cells to obtain the sample image of type II cells to be calculated;

[0057] In the secondary classification step, for the type II cell sample image to be calculated, the color space description technology and the sample clustering method are combined to refine the type II cells into two categories and mark them.

[0058] In practical applications, muscle fiber images include type I cells and type II cells. Type II cells can be further divided into type IIA cells and type IIB cells. The whole cell database represents the recognition database of the entire image, and the type I cell database represents the recognition database of type I cells in the current image. The type II cells represent the result of the whole cell recognition database minus the result of the type I cell database. First, identify the whole cell and separate all the cells in the whole image. Secondly, extract and identify type I cells, and then whole cell-type I cell=type II cell. Type II cells are classified into type IIA cells and type IIB cells through unsupervised learning.

[0059] Based on the logic in the above embodiment, after cutting according to the set requirements and freezing under the set environment, the cell sample is finely cut again and stained using ATPase, and the image of the stained cells is saved as a cell image sample for classification operation. In this way, only one staining is required, which overcomes the defect of multiple staining required in the prior art and effectively saves the cost and time of staining.

[0060] Further, Figure 2 FIG. 1 is a schematic diagram showing the implementation principle of the myofibroblast image classification and recognition method provided by an embodiment of the present invention, as shown in FIG. Figure 2 As shown, the basic classification model pre-built based on the random forest algorithm is used to identify and divide type I cells and all cells (whole cells) in the sample image. The value of small sample cell image samples can be maximized through the random forest machine learning algorithm. Without the need for a large number of training samples, a basic recognition database can be trained to create type I cell and whole cell recognition in unknown myofibroblast images.

[0061] In addition, considering that the presence of small ice crystals in a cell image will affect the accuracy of cell image recognition and classification, in the above embodiment, after the present invention optimizes the segmentation of the sample image of type II cells, it also performs ice crystal recognition and ice crystal removal processing on the segmented type II cell sample image, and removes ice crystals of different sizes in different areas, thereby increasing the accuracy of cell classification and recognition.

[0062] In the secondary classification step, the present invention combines the unsupervised learning technology with the color space characteristics of the cells through a classification algorithm, and finds a color feature value that can replace each cell. Based on this value, the two types of type II cells are completely distinguished by clustering. It can be seen that the above embodiment scheme improves the technology of ATPase-stained myofibroblast image classification, improves the accuracy and efficiency of myofibroblast counting, and improves the process of manual counting required by scientific researchers.

[0063] Furthermore, in practical application, in one embodiment, in the sample collection step, the muscle fiber tissue of the longissimus dorsi muscle of the last rib of the organism within 30 minutes of death is collected as a sample.

[0064] Specifically, in an optional embodiment, in the sample collection step, the process of freezing the muscle fiber tissue after the first cutting includes:

[0065] The muscle fiber tissue after the first cutting is placed in a new plastic sampling tube with a plurality of uniform small holes on the tube wall;

[0066] Myofiber tissue was frozen in liquid nitrogen to control the probability of ice crystal formation.

[0067] In actual application, the muscle fiber tissue to be collected comes from the longissimus dorsi muscle collected from the last rib within 30 minutes after the pig dies, and the sample is cut into a cube of about 0.5×0.5×1cm, completing one cut. It is placed on a plastic sample gasket, and then placed in a new plastic sampling tube with 16 round holes of the same size and a round hole of the same size at the bottom, placed under the liquid nitrogen surface at -196℃ for 15 to 20 seconds to fully contact with liquid nitrogen to prevent ice crystals from forming, and placed in a liquid nitrogen tank for long-term freezing and preservation.

[0068] Furthermore, in the sample staining step, before staining, a cross-section of the muscle fiber tissue is cut using a cryostat in a temperature environment of -20°C, and the cross-section must meet the set size requirements. Among them, a 12um cross-section is cut at -20°C in a cryostat, and a fine secondary cut is completed. Then, the muscle fiber cross-section obtained after the secondary cut is stained using the ATPase (myosin ATPase) acid staining method, and the image with better quality and less ice crystals is collected by electron microscopy. It is saved by taking a photo, and then a high-definition image of the stained muscle fiber section is obtained as the muscle fiber cell image sample data. During actual identification, the muscle fiber tissue to be detected is observed by electron microscopy, and 3 to 5 parts with fewer ice crystals and clear observation of each cell are extracted and used as the slice image.

[0069] In order to first identify and eliminate the type I cell part in the overall myofibroblast sample image, the present invention first prepares a set number of myofibroblast images stained with ATPase, one part of which is used as a training sample and the other part is used as a test sample. Based on the training samples, a basic classification model is constructed by training with a random forest algorithm, and the obtained basic classification model is verified and updated in real time using the test sample until the verification result meets the set accuracy requirement, and the obtained basic classification model is put into the practical application of realizing the recognition of unknown label cell sample images, so as to identify and mark the I cells and whole cells in all cell sample images, and divide them into two types of samples, type I cells and whole cells, and respectively construct a type I cell database and a whole cell database, and based on the above operation, the interference of type I cells can be eliminated to obtain type II cells.

[0070] Specifically, the basic classification model used in the primary classification step is pre-trained and constructed using training samples of known Class I cell and whole cell classification label results. In one embodiment, it is constructed according to the following operations:

[0071] Step C1, dividing the cell image samples of a set scale with basic classification labels into segmentation training samples and segmentation test samples, and establishing a segmentation training database and a segmentation test database respectively;

[0072] Step C2, select sample images and existing classification result data in the training database and input them into the ilastik system, and use the random forest algorithm to iteratively train the cell classification model;

[0073] Step C3, input the sample images in the segmentation test database into the latest model, evaluate the matching degree between the obtained identification results and the existing real classification label results, if the set conditions are not met, update in real time and return to step C2; if the set conditions are met, the final cell classification model is used as the target basic classification model. Among them, the segmentation training samples and test samples can be manually labeled by professionals according to needs.

[0074] After obtaining the target basic classification model based on the above strategy, the cell sample image to be identified is input into the basic classification model, and a cell sample image with a type I cell identifier and a whole cell identifier can be obtained. Furthermore, the cell sample image is divided according to the cell identifier to obtain a sample image of only type I cells and a sample image of only whole cells, respectively. The type I cell image is placed in a preset type I cell database, and the whole cell is placed in a whole cell database. At the same time, as the cell sample image to be identified for further classification and identification operations, the interference of type I cells can be effectively eliminated based on the above operation to obtain type II cells.

[0075] In actual applications, type I cells and all cells are marked separately in the form of manual labels, and the labels are set in the same position as type I cells, cell edges, and all cells. The training begins until the program can accurately identify the fragments in the scanned image and identify type I fiber cells and whole cells. The "live update" function is used to check whether the training is sufficient, and the type I cell database and the whole cell database can be constructed to predict the myofibroblast images to be classified and identified.

[0076] Specifically, the Pixel classification and Object classification in the ilastik software are used to jointly complete the construction of the basic classification module. First, the training set cell images are added to the ilastik software. The cells and edges in the images are marked separately to build a full-cell recognition database. The same image data is marked again with type I cells as the marker, and other cells and edges as another marker. The "live update" function is used to check whether the training is sufficient, and the standard is set as complete recognition of the marked areas.

[0077] Specifically, before further performing the classification and recognition process, the obtained whole-cell sample image is subjected to unqualified factor screening based on the designed automatic quality control step. In a preferred embodiment, the automatic quality control step specifically includes the following operations:

[0078] Step A1, considering the topological theory, using the watershed method to segment the links between cells in the type II cell sample image based on the region;

[0079] Step A2: Use the four-point sampling method to identify ice crystals in the type II cell sample image by region. When the total number of ice crystals exceeds the set condition, use the set rules to divide the ice crystal size and remove the large ice crystals that meet the set size condition.

[0080] Among them, step A1 can effectively separate the links between the connected whole cells (type I and II cells) in the sample image to obtain independent cells. Further, the operation of step A2 is used to divide the entire whole cell sample image into four blocks. The closed elliptical patterns in each area are considered to be ice crystals, and their number is calculated as the number of ice crystals. The number of ice crystals in these four images is calculated by region. If the total number of ice crystals is less than 1200, jump to output; when the number of all ice crystals is greater than 1200, it is necessary to eliminate the influence of ice crystals, calculate the area of ​​each ice crystal (closed area), calculate the mean area of ​​ice crystals in each of the four areas, and the standard deviation value, and black out each ice crystal whose area is less than the area mean plus the area standard deviation value, which is equivalent to deleting the ice crystal. The formula is: mean area +std area The four areas need to remove ice crystals according to this requirement and go to output.

[0081] The above logic is used to first eliminate the interference of type I cells in the whole cell recognition process to obtain a finely processed type II cell image, and then the type II cells are classified into type IIA and type IIB cells by clustering. The value representing the type of each cell is calculated by the value of the color space, and then the value is divided into two categories by clustering, thus representing type IIA cells and IIB cells, and the classification result can be obtained.

[0082] Furthermore, the present invention classifies type II cells into type IIA and type IIB cells by clustering. The numerical value representing each cell type is calculated by the features of the color space, and then the numerical value is divided into two categories by clustering, thereby representing type IIA cells and IIB cells, and the classification result can be obtained.

[0083] On the basis of obtaining type I cells and whole cells, we can obtain type II cells, and type II cells can be classified into type IIA cells and type IIB cells. For example, when the image size is 1920*2560 pixels, the biggest difference between the two types of cells is the different colors at the 10-pixel edge of the cell. Therefore, in one embodiment, in the secondary classification step, the image color data of the cell interior and the cell edge in the type II sample image are extracted respectively, and the RGB color space characteristics of the cell image are used to extract the R value (red), G value (green), and B value (blue) in the image through array calculation, and the mean of the R value, G value, and B value inside the cell and at the cell edge is calculated, and then the difference is calculated to characterize the current cell, and then the type II cells are divided into type IIA and type IIB based on the obtained difference using a clustering algorithm.

[0084] In practical applications, the RGB feature value representing the current cell can be calculated according to the function described in the following formula:

[0085] Δmean R，G，B =|mean R，G，B(inner) -mean R，G，B(edge) | (1)

[0086] In the formula, mean R，G，B(inner) Represents the mean of the R value, G value, and B value inside the cell in the sample image, mean R，G，B(edge) Represents the mean of the R, G, and B values ​​at the edge of the cell in the sample image.

[0087] Furthermore, based on the calculated RGB feature values, three clustering methods, namely, Bayesian Gaussian mixture model, Gaussian mixture model or K-means, are used to classify type II cells into type IIA cells and type IIB cells, and the cells are output.

[0088] Based on the scheme of the above embodiment, first, the image to be predicted is firstly extracted to obtain type I cells and whole cells through the training database. Second, the cell image is checked and adjusted through the automated quality control scheme to see if it meets the requirements, and small ice crystals can be removed. Third, each cell is extracted through the classification algorithm to give each cell a specific color space-related quantitative feature value to represent the cell type. Based on the quantitative data, all type II cells are divided into two categories through the clustering method, so that type IIA cells and type IIB cells can be obtained, thereby completing the typing operation, as shown in the attached figure. Figure 3 shown.

[0089] This provides a way to directly operate on cells, using pixels as feature values ​​to process each cell in color and give the cell different attributes. Clustering can reduce errors, thereby improving the accuracy of cell classification. All automated operations increase efficiency, making muscle adaptation assessment more comprehensive, sensitive and quantitative.

[0090] By using the methods and steps in any one or more of the above embodiments, it is possible to timely and effectively realize the fine identification and classification of myofibroblasts, which has at least the following advantages compared with the existing cell identification technology in the industry:

[0091] 1) The present invention uses the random forest algorithm to achieve good recognition results and establish a reliable basic classification model with only a small sample size of data, without the need to preset a large number of training samples. Under the premise of ensuring the reliability of the basic classification model, the preparation time of the training samples is saved;

[0092] 2) For ATPase-stained myofibroblast sections, only one staining is required, saving staining cost and time;

[0093] 3) The present invention provides an automated basic classification method, which greatly improves the efficiency of cell number determination and type discrimination. Compared with manual counting, human errors are greatly reduced, thereby improving the accuracy of measurement;

[0094] 4) The ice crystal elements that affect the recognition accuracy in the sample image are removed through an automated quality control solution, which improves the processing efficiency compared to manual removal and further ensures the reliability of the final cell recognition;

[0095] The cell type recognition algorithm designed in any one or more embodiments of the present invention can not only be applied to the quality identification of myofibroblasts, but also lay the foundation for algorithms for other cell identifications in the biological or medical fields. It can automatically distinguish cell types through unsupervised learning, effectively reduce the error rate, and provide support for the analysis and optimization of cell characteristics.

[0096] Embodiment 2

[0097] Figure 4 FIG. 4 shows a detailed flow chart of the execution of the method for classifying and identifying myofibroblasts based on ATPase staining provided in another embodiment 2 of the present invention, as shown in FIG. Figure 4 As shown, the myofibroblast image classification method of the present invention can be summarized into three core stages, namely, cell image segmentation processing, automated quality control and cell classification. Before performing cell image segmentation, it is necessary to use ATPase to stain the target myofibroblast sections and obtain corresponding images. Therefore, the myofibroblast image classification and recognition method based on ATPase staining in the present invention generally includes the following steps:

[0098] Sample collection step: collecting muscle fiber tissue that meets the requirements of morphological integrity, cutting the muscle fiber tissue once according to a set standard, and freezing the cut muscle fiber tissue;

[0099] The sample staining step includes: performing a second cutting on the frozen muscle fiber tissue in a specific environment, staining the second-cut muscle fiber tissue using an ATPase acid staining method, and saving the stained image as a muscle fiber sample image;

[0100] In the first-level classification step, the basic classification model pre-built based on the random forest algorithm is used to identify and divide the type I cells and all cells in the sample image, and the type II cell image is extracted by eliminating the type I cells through the whole cell. The corresponding whole-cell database is constructed based on the extracted sample image, and the sample image and result image for type I cell identification are placed in the set type I cell database;

[0101] The automatic quality control step is to first optimize the segmentation of the sample image of type II cells based on the morphological watershed method, and then perform ice crystal recognition and ice crystal screening on the segmented sample image of type II cells to obtain the sample image of type II cells to be calculated;

[0102] In the secondary classification step, for the type II cell sample image to be calculated, the color space description technology and the sample clustering method are combined to refine the type II cells into two categories and mark them.

[0103] The same operations as those in the above embodiment will not be elaborated here, and only the different technical means will be described in detail with reference to the drawings.

[0104] Specifically, before further performing the secondary classification and recognition process, the obtained whole-cell sample image is subjected to unqualified factor screening based on the designed automatic quality control step. In a preferred embodiment, the automatic quality control step specifically includes the following operations:

[0105] Step A1, considering the topological theory, using the watershed method to segment the links between cells in the type II cell sample image based on the region;

[0106] Step A2: Use the four-point sampling method to identify ice crystals in the type II cell sample image by region. When the total number of ice crystals exceeds the set condition, use the set rules to divide the ice crystal size and remove the large ice crystals that meet the set size condition.

[0107] Among them, step A1 can effectively separate the links between the connected whole cells (type I and II cells) in the sample image to obtain independent cells. Further, the operation of step A2 is used to divide the entire whole cell sample image into four blocks. The closed elliptical patterns in each area are considered to be ice crystals, and their number is calculated as the number of ice crystals. The number of ice crystals in these four images is calculated by region. If the total number of ice crystals is less than 1200, jump to output; when the number of all ice crystals is greater than 1200, it is necessary to eliminate the influence of ice crystals, calculate the area of ​​each ice crystal (closed area), calculate the mean area of ​​ice crystals in each of the four areas, and the standard deviation value, and black out each ice crystal whose area is less than the area mean plus the area standard deviation value, which is equivalent to deleting the ice crystal. The formula is: mean area +std area The four areas need to remove ice crystals according to this requirement and go to output.

[0108] In the traditional image segmentation processing, the morphological watershed method is a more efficient method. It is a region-based image segmentation algorithm, which is proposed based on the theory of topology. Furthermore, the researchers of the present invention consider that although the images can be distinguished through the current database training, there are still connection problems at the edges of some cells, and there may still be a situation where the segmentation rate is not 100%, or because of other factors of the muscle fiber tissue sample, the cells in the sample image are unevenly distributed. Such muscle fiber cell sample images are not suitable for further cell type identification. In order to control the factors that affect the fine classification of cells to a minimum and maximize the accuracy of cell classification and identification, the present invention provides a judgment scheme for the uniformity of cell distribution in the sample image to screen out extreme cell sample images whose uniformity does not meet the degree requirements.

[0109] Based on this, the automatic quality control step designed in the present invention further includes before step A2:

[0110] Step B: using a five-point sampling method to calculate the difference in the number of cells in the type II cell sample image by region, which is used to characterize the uniformity of cell distribution in the sample image, and to eliminate sample images that do not meet the uniformity requirements.

[0111] Specifically, in practical application, the whole cell sample image can be divided into four blocks, and a fifth block of the same size as the four blocks is extracted from a point located in the center, and the number of cells contained in each block is calculated respectively. The difference between the number of cells in the five blocks is calculated as the uniformity coefficient of the current sample image. When the difference in the number is less than 70, it is a uniform image. The uniformity coefficient is used to characterize the uniformity of each image, and images whose uniformity does not meet the set requirements are eliminated. Specifically, a preset uniformity threshold can be compared with the uniformity coefficient to determine whether the current sample image meets the set requirements, wherein the uniformity threshold can be flexibly set by the staff according to the needs of cell identification and analysis applications; images with poor predictions are eliminated through an automated quality control solution, overcoming the problems of low efficiency and high error probability of manual sample elimination.

[0112] Based on the operation of the above embodiment, the five-point sampling method can be used to quickly identify cell sample images with poor cell uniformity in the sample, thereby eliminating the interference of uneven cell distribution caused by insufficient cell segmentation and other factors on the cell image recognition results, improving the quality of the input sample during the calculation, and further improving the accuracy of the cell image classification of the present invention.

[0113] Embodiment 3

[0114] Figure 5 FIG. 4 shows a schematic diagram of the detailed execution process of the myofibroblast image classification and recognition method based on ATPase staining provided in Example 3 of the present invention, as shown in FIG. Figure 5 As shown, the myofibroblast image classification method of the present invention can be summarized into three core stages, namely, cell image segmentation processing, automated quality control and cell classification. Before performing cell image segmentation, it is necessary to use ATPase to stain the target myofibroblast sections and obtain corresponding images. Therefore, the myofibroblast image classification and recognition method based on ATPase staining in the present invention generally includes the following steps:

[0115] Sample collection step: collecting muscle fiber tissue that meets the requirements of morphological integrity, cutting the muscle fiber tissue once according to a set standard, and freezing the cut muscle fiber tissue;

[0116] The sample staining step includes performing a secondary cutting of the frozen muscle fiber tissue in a specific environment, staining the secondary cut muscle fiber tissue using ATPase, and saving the stained image as a muscle fiber sample image;

[0117] In the first-level classification step, the basic classification model pre-built based on the random forest algorithm is used to identify and divide the type I cells and all cells in the sample image, and the type II cell image is extracted by eliminating the type I cells through the whole cell. The corresponding whole-cell database is constructed based on the extracted sample image, and the sample image and result image for type I cell identification are placed in the set type I cell database;

[0118] The automatic quality control step is to first optimize the segmentation of the sample image of type II cells based on the morphological watershed method, and then perform ice crystal recognition and ice crystal screening on the segmented sample image of type II cells to obtain the sample image of type II cells to be calculated;

[0119] In the secondary classification step, for the type II cell sample image to be calculated, the color space description technology and the sample clustering method are combined to refine the type II cells into two categories and mark them.

[0120] The same operations as those in the above embodiment will not be elaborated here, and only the different technical means will be described in detail with reference to the drawings.

[0121] The present invention builds an automated quality control solution to automatically identify and screen extreme cell sample slice images and remove them through quality control methods. When applied, the quality control method can be set to include an image watershed method, a five-point sampling method, and a four-point sampling method to remove ice crystals. The watershed method can separate connected cells, and the five-point sampling method can be used to determine the uniformity of cells in the image, thereby determining whether the cells in the image are sufficiently segmented. The four-point sampling method is used to remove ice crystals affecting elements inside the cells in different regions.

[0122] In actual application, based on the type I cells and whole cells, we can obtain type II cells. Type II cells can be classified into type IIA cells and type IIB cells. The biggest difference between these two types of cells is that the colors at the 10-pixel edge of the cells are different. Based on this characteristic, and in order to better eliminate the influence of cell color difference, the researchers of the present invention found that the HSV color space algorithm can also be used to realize the characterization of cell color space characteristics.

[0123] Therefore, in an optional embodiment, in the secondary classification step, by respectively extracting the image color data of the cell interior and the cell edge in the type II sample image, the HSV color space characteristics of the cell image are used to calculate the ratio of the difference between the H value inside the cell and the H value at the cell edge to the S value of the entire cell to characterize the current cell, and then based on the obtained data, a clustering algorithm is used to divide the type II cells into two categories: type IIA and type IIB.

[0124] Specifically, the HSV characteristic value characterizing the current cell type is calculated according to the function described in the following formula:

[0125]

[0126] In the formula, H inner Indicates the H value in the cell, H edge represents the H value at the cell edge, S cell Represents the S value of the entire cell image.

[0127] Then, based on the calculated values, clustering methods are used to divide all type II cells into two categories. For example, three clustering methods, namely Bayesian Gaussian mixture model, Gaussian mixture model, and K-means, can be used to classify type II cells into type IIA cells and type IIB cells and output them.

[0128] By adopting the means in the above-mentioned embodiment, each cell is identified in color with the color attribute of the cell pixel as the characteristic value to give the cell different quantitative attributes, and the secondary fine classification of the cell image is realized by using the clustering method based on the quantitative attribute. The clustering method can reduce the error, thereby improving the accuracy of cell classification, and it can be realized by automated operation, which can effectively improve the efficiency of cell recognition and make the assessment of muscle fiber tissue adaptability more comprehensive, sensitive and quantitative (intuitive).

[0129] For the above-mentioned method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should know that the present invention is not limited by the order of the actions described, because according to the present invention, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0130] It should be pointed out that in other embodiments of the present invention, the method can also obtain new cell image classification and recognition by combining one or several of the above embodiments to optimize the cell tissue evaluation.

[0131] It should be noted that, based on the method in any one or more of the above embodiments of the present invention, the present invention also provides a storage medium, on which is stored a program code that can implement the method as described in any one or more of the above embodiments, and when the code is executed by the operating system, it can implement the above-mentioned myofibroblast image classification and recognition method based on ATPase staining.

[0132] Embodiment 4

[0133] The method is described in detail in the embodiments disclosed in the present invention. The method of the present invention can be implemented by various forms of devices or systems. Therefore, based on other aspects of the method described in any one or more embodiments, the present invention also provides a myofibroblast image classification and recognition system based on ATPase staining, which is used to execute the myofibroblast image classification and recognition method based on ATPase staining described in any one or more embodiments. Specific embodiments are given below for detailed description.

[0134] Specifically, Figure 6 FIG. 4 shows a schematic diagram of the structure of the myofibroblast image classification and recognition system based on ATPase staining provided in an embodiment of the present invention. Figure 6 As shown, the system includes:

[0135] A sample collection module, configured to collect muscle fiber tissue that meets the requirements of morphological integrity, cut the muscle fiber tissue once according to a set standard, and freeze the cut muscle fiber tissue;

[0136] A sample staining module is configured to perform secondary cutting of the muscle fiber tissue obtained after freezing in a specific environment, stain the secondary cut muscle fiber tissue using ATPase, and save the stained image as a muscle fiber sample image;

[0137] A primary classification module is configured to identify and divide type I cells and all cells in the sample image using a basic classification model pre-built based on a random forest algorithm, extract type II cell images by eliminating type I cells from the whole cell, build a corresponding whole cell database based on the extracted sample image, and place the sample image and result image for type I cell identification into the set type I cell database;

[0138] An automatic quality control module is configured to first optimize the segmentation of the sample image of type II cells based on the morphological watershed method, and then perform ice crystal recognition and ice crystal screening on the segmented sample image of type II cells to obtain the sample image of type II cells to be calculated;

[0139] The secondary classification module is configured to refine and classify the type II cells into two categories and identify them by combining the color space description technology and the sample clustering method for the type II cell sample image to be calculated.

[0140] Furthermore, in one embodiment, the sample collection module is configured to collect the muscle fiber tissue of the longissimus dorsi muscle of the last rib of the organism within 30 minutes of death as a sample.

[0141] In a preferred embodiment, the sample collection module performs a freezing operation on the muscle fiber tissue after one cut according to the following rules:

[0142] The muscle fiber tissue after the first cutting is placed in a new plastic sampling tube with a plurality of uniform small holes on the tube wall;

[0143] Myofiber tissue was frozen in liquid nitrogen to control the probability of ice crystal formation.

[0144] In addition, in actual application, the sample staining module is configured as follows: before staining, cross-section slices of muscle fiber tissue are cut using a low-temperature constant temperature microtome in a temperature environment of -20°C, and the cross-section slices must meet the set size requirements.

[0145] In a specific embodiment, the primary classification module constructs the basic classification model according to the following operations:

[0146] Step C1, dividing the cell image samples of a set scale with basic classification labels into segmentation training samples and segmentation test samples, and establishing a segmentation training database and a segmentation test database respectively;

[0147] Step C2, select sample images and existing classification result data in the training database and input them into the ilastik system, and use the random forest algorithm to iteratively train the cell classification model;

[0148] Step C3, input the sample images in the segmentation test database into the latest model, evaluate the matching degree between the obtained identification results and the existing real classification label results, if the set conditions are not met, update in real time and return to step C2; if the set conditions are met, the final cell classification model is used as the target basic classification model.

[0149] In a preferred embodiment, the automatic quality control module processes the type II cell sample image by performing the following operations:

[0150] Step A1, considering the topological theory, using the watershed method to segment the links between cells in the type II cell sample image based on the region;

[0151] Step A2: Use the four-point sampling method to identify ice crystals in the type II cell sample image by region. When the total number of ice crystals exceeds the set condition, use the set rules to divide the ice crystal size and remove the large ice crystals that meet the set size condition.

[0152] Furthermore, in one embodiment, the automatic quality control module further performs the following operations before executing step A2:

[0153] Step B: using a five-point sampling method to calculate the difference in the number of cells in the type II cell sample image by region, which is used to characterize the uniformity of cell distribution in the sample image, and to eliminate sample images that do not meet the uniformity requirements.

[0154] In one embodiment, the secondary classification module performs the following operations to achieve fine classification of type II cells: image color data of the cell interior and cell edge in the type II sample image are extracted respectively, and the mean of the R value, G value, and B value inside the cell and at the cell edge are calculated using the RGB color space characteristics of the cell image, and then the difference is calculated to characterize the current cell, and then the type II cells are divided into type IIA and type IIB based on the obtained difference using a clustering algorithm.

[0155] In another embodiment, the secondary classification module can also perform the following operations to achieve fine classification of type II cells: in the secondary classification step, the image color data of the cell interior and the cell edge in the type II sample image are extracted respectively, and the HSV (Hue, Saturation, Value) color space characteristics of the cell image are used to calculate the ratio of the difference between the H value inside the cell and the H value at the cell edge to the S value of the entire cell to characterize the current cell, and then based on the obtained data, a clustering algorithm is used to divide the type II cells into two categories: type IIA and type IIB.

[0156] In the myofibroblast image classification and recognition system based on ATPase staining provided by the embodiment of the present invention, each module or unit structure can be operated independently or in combination according to actual recognition and calculation requirements to achieve corresponding technical effects.

[0157] In addition, it should be noted that, according to actual application requirements, in order to facilitate the use of users, we can build a website or platform including all the functional modules in the above embodiments, which can greatly reduce the user's use cost and calculation cost. Moreover, the method of the present invention has a fast classification speed, is easy to use in engineering practice, and has broad application value and market prospects.

[0158] It should be understood that the embodiments disclosed in the present invention are not limited to the specific structures, processing steps or materials disclosed herein, but should be extended to equivalent substitutions of these features understood by ordinary technicians in the relevant field. It should also be understood that the terms used herein are only used for the purpose of describing specific embodiments and are not meant to be limiting.

[0159] The "one embodiment" mentioned in the specification means that a particular feature, structure or characteristic described in conjunction with the embodiment is included in at least one embodiment of the present invention. Therefore, the phrase "one embodiment" appearing in various places throughout the specification does not necessarily refer to the same embodiment.

[0160] Although the embodiments disclosed in the present invention are as above, the above contents are only embodiments adopted for facilitating the understanding of the present invention and are not intended to limit the present invention. Any technician in the technical field to which the present invention belongs can make any modifications and changes in the form and details of the implementation without departing from the spirit and scope disclosed in the present invention, but the patent protection scope of the present invention shall still be subject to the scope defined in the attached claims.

Claims

1. A method for classifying and identifying myofibroblast images based on ATPase staining, characterized in that: The method comprises: Sample collection step: collecting muscle fiber tissue that meets the requirements of morphological integrity, cutting the muscle fiber tissue once according to a set standard, and freezing the cut muscle fiber tissue; The sample staining step includes performing a secondary cutting of the frozen muscle fiber tissue in a constant temperature environment, staining the secondary cut muscle fiber tissue using ATPase, and saving the stained image as a muscle fiber sample image; In the first-level classification step, all cells and type I cells in the sample image are identified and divided using a basic classification model pre-built based on the random forest algorithm, and a corresponding whole-cell database is constructed based on the identified and divided sample image. Type I cells are eliminated from the whole cells to extract type II cell images, and the result image for type I cell identification is placed in the set type I cell database; The automatic quality control step first optimizes the segmentation of the sample image of type II cells based on the morphological watershed method, and then performs ice crystal recognition and ice crystal screening on the segmented sample image of type II cells to obtain the sample image of type II cells to be calculated; the secondary classification step combines the color space description technology and the sample clustering method to refine the type II cells into two categories and mark them; In the automatic quality control step, the process of performing ice crystal recognition and ice crystal screening on the segmented type II cell sample image includes: The four-point sampling method is used to identify ice crystals in the type II cell sample image after optimization and segmentation. When the total number of ice crystals exceeds the set condition, the set rules are used to divide the ice crystal size, and the large ice crystals that meet the set size conditions are removed. In the secondary classification step, type II cells are refined and divided into two categories and marked using one of the following methods: The image color data of the cell interior and cell edge in the type II sample image are extracted respectively, and the mean of the R value, G value, and B value in the cell and at the cell edge are calculated by using the RGB color space characteristics of the cell image, and then the difference is calculated to characterize the current cell. Based on the obtained difference, the type II cells are divided into type IIA and type IIB using a clustering algorithm; The image color data of the cell interior and cell edge in the type II sample image are extracted respectively. The HSV (Hue, Saturation, Value) color space characteristics of the cell image are used to calculate the ratio of the difference between the H value inside the cell and the H value at the cell edge to the S value of the entire cell to characterize the current cell. Based on the obtained data, a clustering algorithm is used to divide type II cells into type IIA and type IIB.

2. The method according to claim 1, characterized in that In the automatic quality control step, the process of optimizing the segmentation of the sample image of type II cells using the morphological watershed method includes: Considering the topological theory, the watershed method is used to segment the links between cells in the type II cell sample image based on the region.

3. The method according to claim 2, characterized in that The automatic quality control step further includes before step A2: Step B: using a five-point sampling method to calculate the difference in the number of cells in the type II cell sample image by region, which is used to characterize the uniformity of cell distribution in the sample image, and to eliminate sample images that do not meet the uniformity requirements.

4. The method according to claim 1, characterized in that In the sample collection step, the muscle fiber tissue of the longissimus dorsi muscle of the last rib of the organism within 30 minutes of death is collected as a sample.

5. The method according to claim 1, characterized in that In the sample collection step, the process of freezing the muscle fiber tissue after the first cutting includes: The muscle fiber tissue after the first cutting is placed in a new plastic sampling tube with a plurality of uniform small holes on the tube wall for processing; In liquid nitrogen, the muscle fiber tissue was frozen in a new porous plastic sampling tube to control the probability of ice crystal formation.

6. The method according to claim 1, characterized in that In the sample staining step, before staining, a cross-section of the muscle fiber tissue is cut using a low-temperature constant temperature microtome in a temperature environment of -20°C, and the cross-section must meet the set size requirements.

7. The method according to claim 1, characterized in that The basic classification model used in the first-level classification step is pre-built according to the following operations: Step C1, dividing the cell image samples of a set scale with basic classification labels into segmentation training samples and segmentation test samples, and establishing a segmentation training database and a segmentation test database respectively; Step C2, select sample images and existing classification result data in the training database and input them into the ilastik system, and use the random forest algorithm to iteratively train the cell classification model; Step C3, input the sample images in the segmentation test database into the latest model, evaluate the matching degree between the obtained identification results and the existing real classification label results, if the set conditions are not met, update in real time and return to step C2; if the set conditions are met, the final cell classification model is used as the target basic classification model.

8. A myofibroblast image classification and recognition system based on ATPase staining, characterized in that: The system executes the method according to any one of claims 1 to 7.

Images