Electron-dense material identification method and device for electron microscope image of chronic kidney disease

Abstract

The application discloses a kind of chronic kidney disease electron microscope image electronic dense material identification method and device, by obtaining several chronic kidney disease electron microscope pathology images that have been labeled, and based on the model training of pre-set deep network architecture, obtain electronic dense material identification model, so that when obtaining the electron microscope image of chronic kidney disease to be measured, it can be input into electronic dense material identification model to identify the region of electronic dense material, realizes the automatic identification of the electronic dense material of chronic kidney disease electron microscope image.The pathological characteristics of the present application using deep learning model are learned to chronic kidney disease electron microscope picture, shorten the kidney biopsy electron microscope picture identification time, greatly improve the identification efficiency of the electronic dense material of chronic kidney disease electron microscope image, be conducive to auxiliary diagnosis immune complex type kidney disease, improve diagnosis efficiency, expand the use range of electron microscope.

Description

Technical Field

[0001] This invention relates to the field of image recognition, and more particularly to a method and apparatus for identifying electron compaction in electron micrographs of chronic kidney disease. Background Technology

[0002] Chronic kidney disease (CKD) is one of the most serious diseases affecting human health. Since there are currently no effective drugs or other means to cure it, it places a huge economic burden and psychological stress on families and society. Therefore, early detection, early diagnosis, and early treatment are crucial. Currently, kidney biopsy is the gold standard for diagnosing CKD. A final diagnosis is made through a comprehensive analysis of clinical symptoms, light microscopy, electron microscopy, and immunofluorescence results. Electron microscopy image analysis is often indispensable for CKD diagnosis. Transmission electron microscopy (TEM) can not only observe the ultrastructure of kidney tissue but also the presence and specific distribution of electron-dense materials. Especially for immune complex-type kidney diseases, the presence or absence of electron-dense materials in the electron microscopy report is critical.

[0003] However, electron-dense structures in TEM images exhibit complex textures, diverse shapes, and are difficult to distinguish from surrounding structures. Analyzing TEM images visually is not only time-consuming but also inaccurate. Furthermore, the large volume of specimens requiring electron microscopy, the limited number of experienced pathologists, and the inherent subjectivity in pathological diagnosis all restrict the effectiveness of electron microscopy in diagnosing chronic kidney disease. With the development of artificial intelligence (AI) technology, current deep learning models have achieved high accuracy rates in image recognition, some even surpassing the accuracy of the human eye. Therefore, automatically identifying pathological features in electron micrographs of chronic kidney disease using AI algorithms to assist physicians in diagnosing immune complex-type kidney disease is of significant value and importance in improving the efficiency of clinical diagnosis of chronic kidney disease, supporting scientific research, and promoting the widespread application of electron microscopy.

[0004] Therefore, there is an urgent need for a strategy to identify electron-dense structures in electron micrographs of chronic kidney disease to address the problem of low identification efficiency of electron-dense structures in these images. Summary of the Invention

[0005] This invention provides a method and apparatus for identifying electron-dense materials in electron micrographs of chronic kidney disease, thereby improving the efficiency of electron-dense material identification in these images.

[0006] To address the aforementioned problems, one embodiment of the present invention provides a method for identifying electron-dense structures in electron micrographs of chronic kidney disease, comprising:

[0007] Acquire several electron microscopic pathological images of chronic kidney disease; wherein, the electron microscopic pathological images of chronic kidney disease are images with the location features of electron-dense material already marked;

[0008] A model for recognizing electron-dense materials was obtained by training a pre-defined deep network architecture using several electron microscopic pathological images of the chronic kidney disease.

[0009] When acquiring electron microscopic images of chronic kidney disease, the electron-dense regions of the electron microscopic images are identified using the electron-dense recognition model.

[0010] As an improvement to the above scheme, the step of training a preset deep network architecture using several electron microscopic pathological images of chronic kidney disease to obtain an electron-dense material recognition model includes:

[0011] The preset deep network architecture includes: an encoder, a first decoder, a second decoder, and a deep network structure;

[0012] Several sample images are extracted from several electron microscopic pathological images of chronic kidney disease according to a preset sample number;

[0013] Each sample image is input into the encoder to extract several layers of initial features for each sample image, and the global features of each sample image are obtained through a preset stitching operation.

[0014] The global features of each sample image are input into the first decoder, and the first parsed features of each sample image are obtained through a preset gating mechanism.

[0015] The first analytical feature of each sample image is input into the second decoder, and the second analytical feature of each sample image is obtained through a preset attention supervision mechanism;

[0016] The first and second analytical features of each sample image are input into a deep network structure for training to obtain the electronically dense object recognition model.

[0017] As an improvement to the above scheme, the step of inputting each sample image into the encoder, extracting several layers of initial features for each sample image, and obtaining the global features of each sample image through a preset stitching operation, specifically involves:

[0018] Each of the sample images is input into the encoder to extract several layers of initial features for each sample image; wherein the several layers of initial features are sorted according to depth;

[0019] Global pooling is performed on several initial features of each sample image, and the features are then concatenated to obtain multi-scale features of each sample image.

[0020] Based on the multi-scale features of each sample image, the fusion weights are calculated through a fully connected layer.

[0021] Based on the initial features of the aforementioned layers and the fusion weights, the updated features of each sample image are obtained by multiplying them.

[0022] Upsampling and stitching operations are performed on several layers of updated features of each sample image to obtain the global features of each sample image.

[0023] As an improvement to the above scheme, the preset gating mechanism includes: a gating progressive feature fusion mechanism.

[0024] As an improvement to the above scheme, the preset attention supervision mechanism includes: a reverse attention mechanism and a discontinuous boundary supervision mechanism.

[0025] As an improvement to the above scheme, after obtaining the electronically dense object identification model, the following further steps are included:

[0026] Obtain labeled images of the electron microscopy images of chronic kidney disease patients;

[0027] The accuracy of electron-dense object recognition is obtained by calculating the accuracy of the predicted and labeled images of electron-dense objects output by the electron-dense object recognition model using the Dice coefficient.

[0028] The electronically dense object identification model is evaluated using the accuracy of the electronically dense object identification.

[0029] Accordingly, one embodiment of the present invention also provides an electron compaction identification device for electron microscopic images of chronic kidney disease, comprising: a data acquisition module, a model training module, and a model application module;

[0030] The data acquisition module is used to acquire several electron microscopic pathological images of chronic kidney disease; wherein, the electron microscopic pathological images of chronic kidney disease are images with the location features of electron-dense material already marked.

[0031] The model training module is used to train a preset deep network architecture using several electron microscopic pathological images of chronic kidney disease to obtain an electron-dense material recognition model.

[0032] The model application module is used to identify the electron-dense region in the electron microscope image of chronic kidney disease by means of the electron-dense material recognition model.

[0033] As an improvement to the above scheme, the model training module includes: an architecture classification unit, a sample extraction unit, an encoding unit, a first decoding unit, a second decoding unit, and a training unit;

[0034] The sample extraction unit, used in the preset deep network architecture, includes: an encoder, a first decoder, a second decoder, and a deep network structure;

[0035] The encoding unit is used to extract several sample images from several electron microscopic pathological images of chronic kidney disease according to a preset number of samples;

[0036] The first decoding unit is used to input each of the sample images into the encoder, extract several layers of initial features of each sample image, and obtain the global features of each sample image through a preset stitching operation.

[0037] The second decoding unit is used to input the global features of each sample image into the first decoder, and obtain the first parsed features of each sample image through a preset gating mechanism;

[0038] The training unit is used to input the first analytical feature of each sample image into the second decoder, and obtain the second analytical feature of each sample image through a preset attention supervision mechanism;

[0039] The architecture classification unit is used to input the first and second analytical features of each sample image into the deep network structure for training to obtain the electronically dense object recognition model.

[0040] As an improvement to the above scheme, the step of inputting each sample image into the encoder, extracting several layers of initial features for each sample image, and obtaining the global features of each sample image through a preset stitching operation, specifically involves:

[0041] Each of the sample images is input into the encoder to extract several layers of initial features for each sample image; wherein the several layers of initial features are sorted according to depth;

[0042] Global pooling is performed on several initial features of each sample image, and the features are then concatenated to obtain multi-scale features of each sample image.

[0043] Based on the multi-scale features of each sample image, the fusion weights are calculated through a fully connected layer.

[0044] Based on the initial features of the aforementioned layers and the fusion weights, the updated features of each sample image are obtained by multiplying them.

[0045] Upsampling and stitching operations are performed on several layers of updated features of each sample image to obtain the global features of each sample image.

[0046] As an improvement to the above scheme, the preset gating mechanism includes: a gating progressive feature fusion mechanism.

[0047] As an improvement to the above scheme, the preset attention supervision mechanism includes: a reverse attention mechanism and a discontinuous boundary supervision mechanism.

[0048] As an improvement to the above scheme, after obtaining the electronically dense object identification model, the following further steps are included:

[0049] Obtain labeled images of the electron microscopy images of chronic kidney disease patients;

[0050] The accuracy of electron-dense object recognition is obtained by calculating the accuracy of the predicted and labeled images of electron-dense objects output by the electron-dense object recognition model using the Dice coefficient.

[0051] The electronically dense object identification model is evaluated using the accuracy of the electronically dense object identification.

[0052] Accordingly, one embodiment of the present invention also provides a computer terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements a method for identifying electron dense material in electron micrographs of nephrotic syndrome as described in the present invention.

[0053] Accordingly, one embodiment of the present invention also provides a computer-readable storage medium, the computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform an electron density identification method for electron microscopic images of nephrotic syndrome as described in the present invention.

[0054] As can be seen from the above, the present invention has the following beneficial effects:

[0055] This invention provides a method for identifying electron density in electron micrographs of chronic kidney disease. By acquiring several labeled pathological images of chronic kidney disease, and training a model based on a pre-defined deep network architecture, an electron density recognition model is obtained. This allows the acquisition of electron micrographs of chronic kidney disease to be input into the electron density recognition model for identification of electron density regions, thus achieving automatic identification of electron density in electron micrographs of chronic kidney disease. This invention utilizes a deep learning model to learn the pathological features of electron micrographs of chronic kidney disease, shortening the recognition time of kidney biopsy electron micrographs, significantly improving the efficiency of electron density recognition in electron micrographs of chronic kidney disease, which is beneficial for assisting in the diagnosis of immune complex-type kidney diseases, improving diagnostic efficiency, and expanding the application scope of electron microscopy. Attached Figure Description

[0056] Figure 1This is a flowchart illustrating a method for identifying electron dense material in electron micrographs of chronic kidney disease according to an embodiment of the present invention.

[0057] Figure 2 This is a schematic diagram of the structure of an electron compaction identification device for electron micrographs of chronic kidney disease provided in an embodiment of the present invention;

[0058] Figure 3 This is an original electron microscopic pathological image of chronic kidney disease provided in an embodiment of the present invention;

[0059] Figure 4 This is an image of artificially labeled electron-dense material location features provided in an embodiment of the present invention;

[0060] Figure 5 This invention provides a model for identifying electron-dense regions in an electron microscope image under test.

[0061] Figure 6 This is a schematic diagram of a terminal device structure provided in an embodiment of the present invention. Detailed Implementation

[0062] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0063] Example 1

[0064] See Figure 1 , Figure 1 This is a flowchart illustrating a method for identifying electron dense material in electron micrographs of chronic kidney disease according to an embodiment of the present invention. Figure 1 As shown, this embodiment includes steps 101 to 103, and the specific steps are as follows:

[0065] Step 101: Obtain several electron microscopic pathological images of chronic kidney disease; wherein, the electron microscopic pathological images of chronic kidney disease are images with the location characteristics of electron-dense material already marked.

[0066] In this embodiment, the electron microscopic pathological images of chronic kidney disease were preprocessed. The collected electron microscopic images had different magnifications. Because the observation range and the clarity of different structures varied at different magnifications, this could cause difficulties in subsequent image segmentation and recognition. Therefore, it was necessary to standardize the size and format of the electron microscopic images to reduce their impact on image segmentation. In all the kidney electron microscopic images, the magnification was mainly between 2000 and 5000 times. To ensure that the field of view was not too small and that the internal structures of the glomeruli could be clearly seen, we selected electron microscopic images with a magnification of 4000 times for analysis.

[0067] In one specific embodiment, the samples of electron microscopic pathological images of chronic kidney disease include a variety of common chronic kidney diseases, including but not limited to: IgA nephropathy, membranous nephropathy, lupus nephritis, etc.

[0068] In one specific embodiment, a professional nephrologist regionally labels the electron-dense material in the glomerular electron microscopy image sample to form tissue regional division labels, thereby serving as an electron microscopic pathological image of chronic kidney disease with labeled electron-dense material location characteristics.

[0069] Step 102: Use several electron microscopic pathological images of the chronic kidney disease to train the preset deep network architecture to obtain an electron-dense material recognition model.

[0070] In this embodiment, the step of training a preset deep network architecture using several electron microscopic pathological images of chronic kidney disease to obtain an electron-dense material recognition model includes:

[0071] The preset deep network architecture includes: an encoder, a first decoder, a second decoder, and a deep network structure;

[0072] Several sample images are extracted from several electron microscopic pathological images of chronic kidney disease according to a preset sample number;

[0073] Each sample image is input into the encoder to extract several layers of initial features for each sample image, and the global features of each sample image are obtained through a preset stitching operation.

[0074] The global features of each sample image are input into the first decoder, and the first parsed features of each sample image are obtained through a preset gating mechanism.

[0075] The first analytical feature of each sample image is input into the second decoder, and the second analytical feature of each sample image is obtained through a preset attention supervision mechanism;

[0076] The first and second analytical features of each sample image are input into a deep network structure for training to obtain the electronically dense object recognition model.

[0077] In one specific embodiment, the preset deep network architecture is constructed based on the Co-Net model.

[0078] As a further preferred option, a number of sample images are extracted from a number of electron microscopic pathological images of chronic kidney disease according to a preset number of samples, as follows:

[0079] Electron microscopic pathological images of chronic kidney disease in historical electron microscopy reports were screened, and glomerular TEM images with electron-dense structures and a magnification of 4000 were selected. The electron-dense structures of the glomerular TEM images were then labeled to form tissue region division labels, resulting in a dataset S, S = {X, Y}, where X and Y represent the image and label sets, respectively. X = {X1, X2, Xm, ..., Xi}, where Xi represents the i-th electron microscopy image. There are a total of i images in dataset S. Y = {y1, y2, ym, ..., yi}, where yi represents the electron-dense structure label corresponding to the i-th electron microscopy image.

[0080] In this embodiment, the step of inputting each sample image into the encoder, extracting several layers of initial features from each sample image, and obtaining the global features of each sample image through a preset stitching operation specifically involves:

[0081] Each of the sample images is input into the encoder to extract several layers of initial features for each sample image; wherein the several layers of initial features are sorted according to depth;

[0082] Global pooling is performed on several initial features of each sample image, and the features are then concatenated to obtain multi-scale features of each sample image.

[0083] Based on the multi-scale features of each sample image, the fusion weights are calculated through a fully connected layer.

[0084] Based on the initial features of the aforementioned layers and the fusion weights, the updated features of each sample image are obtained by multiplying them.

[0085] Upsampling and stitching operations are performed on several layers of updated features of each sample image to obtain the global features of each sample image.

[0086] To better illustrate, the global feature acquisition method is as follows: features are extracted from different layers of the encoder, where features are ordered according to their depth in the network. Global average pooling is then applied to the encoded features F. i,i∈{2,3,4,5}, and further concatenate them to construct a fused multi-scale representation Q. By operating two fully connected layers on Q, a layer-by-layer fusion weight Φ is obtained. The original features can be obtained through F i With Φ i The data is then multiplied and updated. Through this series of upsampling and concatenation operations, the global features are finally obtained.

[0087] In this embodiment, the preset gating mechanism includes a gating progressive feature fusion mechanism.

[0088] To better illustrate this, and to fully leverage the guiding and supplementary role of global feature F, making progressive fusion more refined, a gated progressive feature fusion was designed, where each layer of features can be connected to global features, and features from multiple layers can be stably fused.

[0089] The global features of each sample image are input into the first decoder, and the first analytical features of each sample image are obtained through a gated progressive feature fusion mechanism.

[0090] In this embodiment, the preset attention supervision mechanism includes: a reverse attention mechanism and a discontinuous boundary supervision mechanism.

[0091] To better illustrate, the pre-defined attention supervision mechanism uses a reverse attention mechanism to correct the segmentation results, and employs discontinuous boundary supervision to make the model pay more attention to the detailed segmentation at the boundary positions.

[0092] The first analytical feature The input is fed into the second decoder, where it is decoded using a reverse attention mechanism and a discontinuity boundary mechanism to obtain the second parsed features.

[0093] In one specific embodiment, the first decoder may be a decoder with a globally guided feature interaction module, and the second decoder may be a decoder with discontinuous boundary supervision.

[0094] In one specific embodiment, the first parsed feature obtained by the first decoder is... The second parsed features obtained by the second decoder Dynamic fusion is achieved by training the model using a simplified SE module of the SE Net network. Based on a deep learning model with two decoders, the electron microscopy feature maps extracted from the shallow network and the features extracted from the deep network are processed by the decoder of the globally guided feature interaction module, enabling the stable fusion of features from multiple layers. At the same time, the model is guided to utilize missing details by a decoder with discontinuous boundary supervision. Finally, a simplified SE module from the SE Net network is used to obtain a more refined prediction map.

[0095] In this embodiment, after obtaining the electronically dense material identification model, the method further includes:

[0096] Obtain labeled images of the electron microscopy images of chronic kidney disease patients;

[0097] The accuracy of electron-dense object recognition is obtained by calculating the accuracy of the predicted and labeled images of electron-dense objects output by the electron-dense object recognition model using the Dice coefficient.

[0098] The electronically dense object identification model is evaluated using the accuracy of the electronically dense object identification.

[0099] In one specific embodiment, in addition to visual effects, this example also objectively evaluates the effectiveness of the invention through quantitative indicators. This example uses the gold standard image manually segmented by pathology experts and employs the Dice coefficient of the image to evaluate the segmentation effect. The Dice coefficient is a geometric similarity function, ranging from [0,1]. A larger Dice coefficient indicates a more accurate segmentation result, and its definition is:

[0100]

[0101] In the formula, A represents the labeled image, and B represents the predicted image of electron compaction output by the model.

[0102] Step 103: When acquiring the electron microscopic image of the patient with chronic kidney disease, the electron dense region of the electron microscopic image is identified by the electron dense material identification model.

[0103] In one specific embodiment, the method of the present invention was evaluated based on quantitative indicators. The average Dice coefficient was 0.797, indicating that the segmentation results of the invention are close to the gold standard.

[0104] For a better illustration of this embodiment, please refer to [link / reference]. Figure 3 , Figure 4 and Figure 5 ,in, Figure 3 These are original electron microscopic pathological images of chronic kidney disease. Figure 4 Images with human-annotated location features of electron-dense structures. Figure 5 To identify electron-dense regions in the electron microscope image under test for the model.

[0105] This embodiment acquires several labeled electron microscopic pathological images of chronic kidney disease and trains a model based on a preset deep network architecture to obtain an electron-dense feature recognition model. This allows the model to automatically identify electron-dense feature regions in electron microscopic images of chronic kidney disease when acquiring such images. This approach utilizes artificial intelligence algorithms to address the problem of automatically and intelligently recognizing pathological features in electron microscopic images of chronic kidney disease, shortening the image recognition time for kidney biopsy electron microscopy images. This opens up possibilities for the large-scale use of electron microscopy technology in kidney biopsies and helps shorten the treatment cycle for kidney disease.

[0106] Example 2

[0107] See Figure 2 , Figure 2 This is a schematic diagram of the structure of an electron compaction identification device for electron micrographs of chronic kidney disease provided in an embodiment of the present invention, including: a data acquisition module 201, a model training module 202 and a model application module 203;

[0108] The data acquisition module 201 is used to acquire several electron microscopic pathological images of chronic kidney disease; wherein, the electron microscopic pathological images of chronic kidney disease are images with the location features of electron-dense material already marked.

[0109] The model training module 202 is used to train a preset deep network architecture using several electron microscopic pathological images of chronic kidney disease to obtain an electron-dense material recognition model.

[0110] The model application module 203 is used to identify the electron-dense region of the electron-dense image of the target electron microscope image when acquiring the target electron microscope image of chronic kidney disease through the electron-dense material recognition model.

[0111] As an improvement to the above scheme, the model training module 202 includes: an architecture classification unit, a sample extraction unit, an encoding unit, a first decoding unit, a second decoding unit, and a training unit;

[0112] The sample extraction unit, used in the preset deep network architecture, includes: an encoder, a first decoder, a second decoder, and a deep network structure;

[0113] The encoding unit is used to extract several sample images from several electron microscopic pathological images of chronic kidney disease according to a preset number of samples;

[0114] The first decoding unit is used to input each of the sample images into the encoder, extract several layers of initial features of each sample image, and obtain the global features of each sample image through a preset stitching operation.

[0115] The second decoding unit is used to input the global features of each sample image into the first decoder, and obtain the first parsed features of each sample image through a preset gating mechanism;

[0116] The training unit is used to input the first analytical feature of each sample image into the second decoder, and obtain the second analytical feature of each sample image through a preset attention supervision mechanism;

[0117] The architecture classification unit is used to input the first and second analytical features of each sample image into the deep network structure for training to obtain the electronically dense object recognition model.

[0118] As an improvement to the above scheme, the step of inputting each sample image into the encoder, extracting several layers of initial features for each sample image, and obtaining the global features of each sample image through a preset stitching operation, specifically involves:

[0119] Each of the sample images is input into the encoder to extract several layers of initial features for each sample image; wherein the several layers of initial features are sorted according to depth;

[0120] Global pooling is performed on several initial features of each sample image, and the features are then concatenated to obtain multi-scale features of each sample image.

[0121] Based on the multi-scale features of each sample image, the fusion weights are calculated through a fully connected layer.

[0122] Based on the initial features of the aforementioned layers and the fusion weights, the updated features of each sample image are obtained by multiplying them.

[0123] Upsampling and stitching operations are performed on several layers of updated features of each sample image to obtain the global features of each sample image.

[0124] As an improvement to the above scheme, the preset gating mechanism includes: a gating progressive feature fusion mechanism.

[0125] As an improvement to the above scheme, the preset attention supervision mechanism includes: a reverse attention mechanism and a discontinuous boundary supervision mechanism.

[0126] As an improvement to the above scheme, after obtaining the electronically dense object identification model, the following further steps are included:

[0127] Obtain labeled images of the electron microscopy images of chronic kidney disease patients;

[0128] The accuracy of electron-dense object recognition is obtained by calculating the accuracy of the predicted and labeled images of electron-dense objects output by the electron-dense object recognition model using the Dice coefficient.

[0129] The electronically dense object identification model is evaluated using the accuracy of the electronically dense object identification.

[0130] This embodiment acquires electron microscopic pathological images of chronic kidney disease through a data acquisition module, and inputs these images into a model training module to train an electron density recognition model. Then, when the model application module obtains electron microscopic images of chronic kidney disease, the electron density recognition model identifies the electron density regions in the images, achieving automatic identification of electron density in chronic kidney disease electron microscopic images. This invention utilizes a deep learning model to learn the pathological features of electron microscopic images of chronic kidney disease, shortening the image recognition time for kidney biopsy electron microscopy images, significantly improving the efficiency of electron density recognition in chronic kidney disease electron microscopic images, which is beneficial for assisting in the diagnosis of immune complex-type kidney diseases, improving diagnostic efficiency, and expanding the application scope of electron microscopy.

[0131] Example 3

[0132] See Figure 6 , Figure 6 This is a schematic diagram of the terminal device structure provided in an embodiment of the present invention.

[0133] One terminal device in this embodiment includes: a processor 601, a memory 602, and a computer program stored in the memory 602 and executable on the processor 601. When the processor 601 executes the computer program, it implements the steps of the aforementioned methods for identifying electron dense material in electron micrographs of chronic kidney disease, as described in the embodiments, for example... Figure 1 All steps of the method for identifying electron dense material in electron micrographs of chronic kidney disease, as shown. Alternatively, when the processor executes the computer program, it implements the functions of each module in the above-described device embodiments, for example: Figure 2 All modules of the electron compaction identification device for electron microscopic images of chronic kidney disease are shown.

[0134] In addition, embodiments of the present invention also provide a computer-readable storage medium, the computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform the electron density identification method for electron micrographs of chronic kidney disease as described in any of the above embodiments.

[0135] Those skilled in the art will understand that the schematic diagram is merely an example of a terminal device and does not constitute a limitation on the terminal device. It may include more or fewer components than shown in the diagram, or combine certain components, or different components. For example, the terminal device may also include input / output devices, network access devices, buses, etc.

[0136] The processor 601 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor. The processor 601 is the control center of the terminal device, connecting various parts of the terminal device through various interfaces and lines.

[0137] The memory 602 can be used to store the computer programs and / or modules. The processor 601 implements various functions of the terminal device by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory 602. The memory 602 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0138] Wherein, if the modules / units integrated in the terminal device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when the computer program is executed by a processor, it can implement the steps of the various method embodiments described above. Wherein, the computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc.

[0139] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0140] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for identifying electron-dense material in electron micrographs of chronic kidney disease, characterized in that, include: Acquire several electron microscopic pathological images of chronic kidney disease; wherein, the electron microscopic pathological images of chronic kidney disease are images with the location features of electron-dense material already marked; A model for recognizing electron-dense materials is obtained by training a pre-defined deep network architecture using several electron microscopic pathological images of chronic kidney disease. The process includes: the pre-defined deep network architecture comprising an encoder, a first decoder, a second decoder, and a deep network structure; extracting several sample images from the pre-defined electron microscopic pathological images of chronic kidney disease according to a pre-defined number of samples; inputting each sample image into the encoder to extract several layers of initial features for each sample image, and obtaining global features for each sample image through a pre-defined stitching operation; inputting the global features of each sample image into the first decoder, and obtaining first analytical features for each sample image through a pre-defined gating mechanism; inputting the first analytical features of each sample image into the second decoder, and obtaining second analytical features for each sample image through a pre-defined attention supervision mechanism; and inputting the first and second analytical features of each sample image into the deep network structure for training to obtain the electron-dense material recognition model. When acquiring electron microscopic images of chronic kidney disease, the electron-dense regions of the electron microscopic images are identified using the electron-dense recognition model.

2. The method for identifying electron-dense matter in electron micrographs of chronic kidney disease according to claim 1, characterized in that, The process involves inputting each sample image into an encoder to extract several layers of initial features from each sample image, and then obtaining the global features of each sample image through a preset stitching operation. Specifically: Each of the sample images is input into the encoder to extract several layers of initial features for each sample image; wherein the several layers of initial features are sorted according to depth; Global pooling is performed on several initial features of each sample image, and the features are then concatenated to obtain multi-scale features of each sample image. Based on the multi-scale features of each sample image, the fusion weights are calculated through a fully connected layer. Based on the initial features of the aforementioned layers and the fusion weights, the updated features of each sample image are obtained by multiplying them. Upsampling and stitching operations are performed on several layers of updated features of each sample image to obtain the global features of each sample image.

3. The method for identifying electron-dense matter in electron micrographs of chronic kidney disease according to claim 1, characterized in that, The preset gating mechanism includes: a gating progressive feature fusion mechanism.

4. The method for identifying electron-dense matter in electron micrographs of chronic kidney disease according to claim 1, characterized in that, The preset attention supervision mechanism includes: a reverse attention mechanism and a discontinuous boundary supervision mechanism.

5. The method for identifying electron-dense matter in electron micrographs of chronic kidney disease according to claim 1, characterized in that, After obtaining the electronically dense material identification model, the method further includes: Obtain labeled images of the electron microscopy images of chronic kidney disease patients; The accuracy of electron-dense object recognition is obtained by calculating the accuracy of the predicted and labeled images of electron-dense objects output by the electron-dense object recognition model using the Dice coefficient. The electronically dense object identification model is evaluated using the accuracy of the electronically dense object identification.

6. A device for identifying electron dense matter in electron microscopic images of chronic kidney disease, characterized in that, include: Data acquisition module, model training module, and model application module; The data acquisition module is used to acquire several electron microscopic pathological images of chronic kidney disease; wherein, the electron microscopic pathological images of chronic kidney disease are images with the location features of electron-dense material already marked. The model training module is used to train a preset deep network architecture using several electron microscopic pathological images of chronic kidney disease to obtain an electron-dense object recognition model. The process of training the preset deep network architecture using several electron microscopic pathological images of chronic kidney disease to obtain the electron-dense object recognition model includes: the preset deep network architecture comprising an encoder, a first decoder, a second decoder, and a deep network structure; extracting several sample images from the several electron microscopic pathological images of chronic kidney disease according to a preset number of samples; inputting each sample image into the encoder to extract several layers of initial features for each sample image, and obtaining global features for each sample image through a preset stitching operation; inputting the global features of each sample image into the first decoder, and obtaining first analytical features for each sample image through a preset gating mechanism; inputting the first analytical features of each sample image into the second decoder, and obtaining second analytical features for each sample image through a preset attention supervision mechanism; and inputting the first and second analytical features of each sample image into the deep network structure for training to obtain the electron-dense object recognition model. The model application module is used to identify the electron-dense region in the electron microscope image of chronic kidney disease by means of the electron-dense material recognition model.

7. A computer terminal device, characterized in that, The method includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements a method for identifying electron dense material in electron micrographs of chronic kidney disease as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device on which the computer-readable storage medium is located to perform a method for identifying electron dense matter in electron micrographs of chronic kidney disease as described in any one of claims 1 to 5.

Images