A method for obtaining the degree of oxidative damage of red blood cells based on deep learning

Abstract

The present invention discloses a method for obtaining the degree of oxidative damage of red blood cells based on deep learning, by obtaining a red blood cell image, marking a normal red blood cell image in the red blood cell image to form a label, and obtaining a label image of the normal red blood cell; using the red blood cell image and the label image of the normal red blood cell as a data set, dividing the data set into a training set and a test set, and the training set and the test set both contain the label image of the red blood cell image and the normal red blood cell; based on the U-Net model and the YOLOv3 model, constructing a prediction model for the degree of oxidative damage of red blood cells; using the training set to train the prediction model for the degree of oxidative damage of red blood cells, and obtaining a trained prediction model for the degree of oxidative damage of red blood cells; using the test set to test the trained prediction model for the degree of oxidative damage of red blood cells, and obtaining a test result of the degree of oxidative damage of red blood cells. The present invention can instantly obtain the degree of oxidative damage of red blood cells and intuitively observe the real-time changing state of cells.

Description

Technical Field

[0001] The present invention relates to the technical field of cell image recognition, and in particular to a method for acquiring the degree of oxidative damage of red blood cells based on deep learning. Background Art

[0002] With the continuous development of technologies for analyzing biologically active molecular components in single cells, the diagnosis of single-cell pathways has the potential to improve diagnostic accuracy based on knowledge of intracellular and intercellular networks. Cell shape is a basic biological feature that provides specific information about physiological or pathological cell conditions. Now, with the increasing development of computer hardware and artificial intelligence, the application of machine learning and deep learning in the medical field is becoming more and more extensive.

[0003] When using traditional methods to evaluate the extent of oxidative damage to red blood cells during oxidative stress, the magnitude of the oxidative damage cannot be directly obtained, which is not convenient enough and the real-time status of the cells cannot be observed intuitively. Summary of the invention

[0004] The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the present invention proposes a method for obtaining the degree of oxidative damage of red blood cells based on deep learning, which can directly obtain the magnitude of the degree of oxidative damage of red blood cells and can intuitively observe the real-time state of cells.

[0005] In a first aspect, an embodiment of the present invention provides a method for obtaining the degree of oxidative damage of red blood cells based on deep learning, comprising the steps of:

[0006] Acquire a red blood cell image, mark a normal red blood cell image in the red blood cell image to form a label, and obtain a label image of the normal red blood cell;

[0007] The red blood cell image and the label image of the normal red blood cell are used as a data set, and the data set is divided into a training set and a test set, wherein both the training set and the test set contain the red blood cell image and the label image of the normal red blood cell;

[0008] Based on the U-Net model and the YOLOv3 model, a prediction model for the degree of oxidative damage of red blood cells was constructed;

[0009] The training set is used to train the red blood cell oxidative damage degree prediction model to obtain a trained red blood cell oxidative damage degree prediction model, and the trained red blood cell oxidative damage degree prediction model is obtained by the following method:

[0010] Using the training set to train a preset U-Net model to obtain a mask image of the normal red blood cells;

[0011] Using the YOLOv3 model to perform rectangular anchor frame annotation and recognition on the normal red blood cells in the mask image of the normal red blood cells, and obtaining the number of anchor frames annotated with the normal red blood cells;

[0012] Using the YOLOv3 model to perform rectangular anchor box annotation and recognition on the red blood cells in all the red blood cell images, and obtaining the number of anchor boxes annotated with all the red blood cells;

[0013] Obtaining the degree of oxidative damage of red blood cells based on the number of anchor frames labeled with the normal red blood cells and the number of anchor frames labeled with all red blood cells;

[0014] The test set is used to test the trained red blood cell oxidative damage degree prediction model to obtain a red blood cell oxidative damage degree test result.

[0015] Compared with the prior art, the first aspect of the present invention has the following beneficial effects:

[0016] This method is based on the U-Net model and the YOLOv3 model to construct a prediction model for the degree of oxidative damage of red blood cells; the U-Net model is used to obtain the mask image of normal red blood cells to improve the segmentation accuracy; the YOLOv3 model is used to annotate and identify the normal red blood cells in the mask image of normal red blood cells with rectangular anchor frames to obtain the number of anchor frames annotated with normal red blood cells; the YOLOv3 model is used to annotate and identify the red blood cells in all red blood cell images with rectangular anchor frames to obtain the number of anchor frames annotated with all red blood cells; the YOLOv3 model is used for annotation and identification to improve the recognition accuracy and recognition speed. The YOLOv3 model can balance accuracy and speed well. This method directly obtains the degree of oxidative damage of red blood cells through the prediction model for the degree of oxidative damage of red blood cells. Therefore, this method can not only obtain the degree of oxidative damage of red blood cells in real time, but also can intuitively observe the real-time change state of cells through the degree of oxidative damage of red blood cells. By obtaining the degree of oxidative damage of red blood cells in real time, it is more convenient, saves analysis time, and improves efficiency.

[0017] According to some embodiments of the present invention, acquiring the red blood cell image comprises:

[0018] Leica laser confocal microscope was used to take photos of red blood cell morphology, and the red blood cell images were exported in a fixed tif format.

[0019] According to some embodiments of the present invention, the U-Net model includes an encoder and a decoder, and the training set is used to train the preset U-Net model to obtain the mask image of the normal red blood cells, specifically including the steps of:

[0020] Inputting the red blood cell image and the label image of the normal red blood cells into the encoder of the U-Net model;

[0021] Using a convolution operation in the encoder, and using multiple continuous downsampling to gradually extract feature information of the red blood cell image and the label image of the normal red blood cells;

[0022] The feature information is subjected to multiple upsampling by the decoder of the U-Net model to fuse shallow features and deep features and restore image information;

[0023] According to the encoder and the decoder, a mask image of the normal red blood cells is obtained.

[0024] According to some embodiments of the present invention, the YOLOv3 model includes a feature extraction network Darknet-53 and a prediction network, and the YOLOv3 model is used to perform annotation recognition, specifically including the steps of:

[0025] The feature extraction network Darknet-53 uses five residual modules to extract features from the red blood cell image and the mask image of the normal red blood cells;

[0026] The prediction network uses a series of convolution, upsampling and splicing operations to obtain prediction boxes of three different sizes for labeling and identifying all cells in the red blood cell image and normal red blood cells in the mask image of normal red blood cells.

[0027] According to some embodiments of the present invention, the degree of oxidative damage to red blood cells is calculated by the following calculation formula:

[0028]

[0029] Wherein, d represents the degree of oxidative damage of the red blood cells, x represents the number of anchor frames annotated with the normal red blood cells, and y represents the number of anchor frames annotated with all the red blood cells.

[0030] In a second aspect, an embodiment of the present invention further provides a system for acquiring the degree of oxidative damage of red blood cells based on deep learning, comprising:

[0031] A label image acquisition unit is used to acquire a red blood cell image, mark a normal red blood cell image in the red blood cell image to form a label, and obtain a label image of the normal red blood cell;

[0032] a data set division unit, configured to use the red blood cell image and the label image of the normal red blood cell as a data set, and divide the data set into a training set and a test set, wherein both the training set and the test set contain the red blood cell image and the label image of the normal red blood cell;

[0033] A prediction model building unit, used to build a prediction model for the degree of oxidative damage of red blood cells based on the U-Net model and the YOLOv3 model;

[0034] The prediction model training unit is used to train the prediction model of the degree of oxidative damage of red blood cells using the training set to obtain a trained prediction model of the degree of oxidative damage of red blood cells. The trained prediction model of the degree of oxidative damage of red blood cells is obtained by:

[0035] Using the training set to train a preset U-Net model to obtain a mask image of the normal red blood cells;

[0036] Using the YOLOv3 model to perform rectangular anchor frame annotation and recognition on the normal red blood cells in the mask image of the normal red blood cells, and obtaining the number of anchor frames annotated with the normal red blood cells;

[0037] Using the YOLOv3 model to perform rectangular anchor box annotation and recognition on the red blood cells in all the red blood cell images, and obtaining the number of anchor boxes annotated with all the red blood cells;

[0038] Obtaining the degree of oxidative damage of red blood cells based on the number of anchor frames labeled with the normal red blood cells and the number of anchor frames labeled with all red blood cells;

[0039] The prediction model testing unit is used to test the trained red blood cell oxidative damage degree prediction model using the test set to obtain a red blood cell oxidative damage degree test result.

[0040] According to some embodiments of the present invention, the label image acquisition unit includes:

[0041] A Leica laser confocal microscope is used to take photos of the morphology of red blood cells, and the images are exported in a fixed tif format, and the images are used as the original images of the red blood cells;

[0042] Before shooting with the Leica laser confocal microscope, add erythrocyte sedimentation fluid and gently shake to mix, and then put it into a glass-bottomed confocal dish modified with poly-lysine for shooting.

[0043] According to some embodiments of the present invention, the parameter settings of the Leica laser confocal microscope include a Gain setting range of 300 to 350, an Offset setting of -0.15, a 552nm laser setting range of 1.4 to 2.60, a Format setting of 1024×1024, a Speed ​​setting of 400, and a Zoom factor setting range of 1 to 1.5.

[0044] In a third aspect, an embodiment of the present invention provides a device for acquiring the degree of oxidative damage of red blood cells based on deep learning, comprising at least one control processor and a memory for communicating with the at least one control processor; the memory stores instructions executable by the at least one control processor, and the instructions are executed by the at least one control processor so that the at least one control processor can execute a method for acquiring the degree of oxidative damage of red blood cells based on deep learning as described above.

[0045] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute the method for obtaining the degree of oxidative damage of red blood cells based on deep learning as described above.

[0046] It can be understood that the beneficial effects of the second to fourth aspects compared with the related art are the same as the beneficial effects of the first aspect compared with the related art. Please refer to the relevant description in the first aspect, and no further details will be given here. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0048] Figure 1 It is a flow chart of a method for obtaining the degree of oxidative damage of red blood cells based on deep learning according to an embodiment of the present invention;

[0049] Figure 2 is a structural diagram of a U-Net model according to an embodiment of the present invention;

[0050] Figure 3 1 is a structural diagram of a YOLOv3 model according to an embodiment of the present invention;

[0051] Figure 4 It is a fitting diagram of the relationship between the normal red blood cell ratio measured by the flow cytometer and the red blood cell oxidative damage degree prediction model according to one embodiment of the present invention;

[0052] Figure 5 This is a structural diagram of a system for acquiring the degree of oxidative damage of red blood cells based on deep learning according to an embodiment of the present invention. DETAILED DESCRIPTION

[0053] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0054] In the description of the present invention, if there is a description of first, second, etc., it is only for the purpose of distinguishing technical features, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features or implicitly indicating the order of the indicated technical features.

[0055] In the description of the present invention, it should be understood that descriptions involving orientation, such as orientation or positional relationship indicated as up, down, etc., are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the present invention.

[0056] In the description of the present invention, it should be noted that, unless otherwise clearly defined, terms such as setting, installing, connecting, etc. should be understood in a broad sense, and technicians in the relevant technical field can reasonably determine the specific meanings of the above terms in the present invention in combination with the specific content of the technical solution.

[0057] With the continuous development of technologies for analyzing biologically active molecular components in single cells, the diagnosis of single-cell pathways has the potential to improve diagnostic accuracy based on knowledge of intracellular and intercellular networks. Cell shape is a basic biological feature that provides specific information about physiological or pathological cell conditions. Now, with the increasing development of computer hardware and artificial intelligence, the application of machine learning and deep learning in the medical field is becoming more and more extensive.

[0058] When using traditional methods to evaluate the extent of oxidative damage to red blood cells during oxidative stress, the magnitude of the oxidative damage cannot be directly obtained, which is not convenient enough and the real-time status of the cells cannot be observed intuitively.

[0059] To solve the above problems, the present invention is based on U-Net model and YOLOv3 model, constructs a prediction model for the degree of oxidative damage of red blood cells; obtains the mask image of normal red blood cells through the U-Net model, improves the segmentation accuracy, and uses the YOLOv3 model to mark and identify the normal red blood cells in the mask image of normal red blood cells with rectangular anchor frames, and obtains the number of anchor frames marked by normal red blood cells; uses the YOLOv3 model to mark and identify the red blood cells in all red blood cell images with rectangular anchor frames, obtains the number of anchor frames marked by all red blood cells, and uses the YOLOv3 model to mark and identify, improves the recognition accuracy and recognition speed, and the YOLOv3 model can balance the accuracy and speed well. The present invention directly obtains the degree of oxidative damage of red blood cells through the prediction model for the degree of oxidative damage of red blood cells, therefore, the present invention can immediately obtain the degree of oxidative damage of red blood cells, and can also intuitively observe the real-time change state of cells through the degree of oxidative damage of red blood cells.

[0060] Reference Figure 1 The embodiment of the present invention provides a method for obtaining the degree of oxidative damage of red blood cells based on deep learning, comprising the steps of:

[0061] Step S100, acquiring a red blood cell image, marking a normal red blood cell image in the red blood cell image to form a label, and obtaining a label image of the normal red blood cell.

[0062] Specifically, this embodiment uses a Leica laser confocal microscope to take photos of red blood cell morphology, and exports red blood cell images in a fixed tif format. The Leica laser confocal microscope (LEICA / TCS SP8) is used to take photos of red blood cell morphology under various treatment conditions in a bright field, and after taking photos, all red blood cell images are exported with a ruler in a tif format to obtain red blood cell images under different treatment conditions. The red blood cell shooting parameters under the same batch of treatment conditions should be controlled to be basically consistent. Therefore, the parameter settings of the Leica laser confocal microscope include Gain setting range of 300 to 350, Offset setting of -0.15, 552nm laser setting range of 1.4 to 2.60, Format setting of 1024×1024, Speed ​​setting of 400, and Zoom factor setting range of 1 to 1.5, wherein 552nm laser means that the wavelength of the laser is 552nm.

[0063] Since red blood cells are suspended cells, they are prone to collision and adhesion during the floating process, which makes ghosting easy to occur when photographing, and it is difficult to obtain single cells. Therefore, before photographing, the red blood cells are first added to the red blood cell sedimentation fluid and gently shaken to mix, and then placed in a glass-bottomed confocal dish modified with poly-lysine for photographing.

[0064] This embodiment obtains the red blood cell image through the above method, which can better obtain the state of a single cell and reduce the occurrence of ghosting between cells.

[0065] Step S200: taking the red blood cell images and the label images of normal red blood cells as a data set, dividing the data set into a training set and a test set, wherein both the training set and the test set contain the red blood cell images and the label images of normal red blood cells.

[0066] Specifically, for all red blood cell images in the tif format taken by a Leica laser confocal microscope, first use marking software to mark normal red blood cells in a double-sided concave pancake shape, thereby forming a label image of a normal red blood cell, and then use the red blood cell image and the label image of the normal red blood cell as a data set, and divide the data set into a training set and a test set. For example, first put all the images in a folder, then write a code program to shuffle the order of the images, and then randomly put the images into two folders according to the corresponding ratio, one is the training set folder, and the other is the test set folder. The ratio of the training set and the test set of this embodiment can be set as needed, and this embodiment is not limited. The training set and the test set of this embodiment both contain red blood cell images and label images of normal red blood cells.

[0067] Step S300: construct a prediction model for the degree of oxidative damage of red blood cells based on the U-Net model and the YOLOv3 model.

[0068] Specifically, this embodiment uses the U-Net model and the YOLOv3 model to construct a prediction model for the degree of oxidative damage of red blood cells. It should be noted that the U-Net model and the YOLOv3 model used in this embodiment are models in the prior art and are not described in detail in this embodiment.

[0069] Step S400: training the red blood cell oxidative damage degree prediction model using the training set to obtain a trained red blood cell oxidative damage degree prediction model. The trained red blood cell oxidative damage degree prediction model is obtained by the following method:

[0070] The preset U-Net model is trained using the training set to obtain the mask image of normal red blood cells;

[0071] The YOLOv3 model is used to annotate and identify the normal red blood cells in the mask image of the normal red blood cells with rectangular anchor frames, and the number of anchor frames annotated with the normal red blood cells is obtained;

[0072] The YOLOv3 model is used to annotate and identify the red blood cells in all red blood cell images with rectangular anchor boxes, and the number of anchor boxes annotated with all red blood cells is obtained;

[0073] The degree of oxidative damage of red blood cells was obtained based on the number of anchor frames labeled with normal red blood cells and the number of anchor frames labeled with all red blood cells.

[0074] Specifically, refer to Figure 2, the U-Net model consists of an encoder and a decoder, the left half is the encoder, and the right half is the decoder. The encoder uses convolution operations and adopts multiple continuous downsampling to gradually extract image feature information. The decoder uses multiple continuous upsampling operations, and combines the downsampling layer information and the upsampling input information to fuse the shallow features and deep features to restore the image information. Therefore, in this embodiment, the red blood cell image and the label image of the normal red blood cell are input into the encoder of the U-Net model; the convolution operation is used in the encoder, and the feature information of the red blood cell image and the label image of the normal red blood cell is gradually extracted by multiple continuous downsampling; the feature information is multiple upsampled through the decoder of the U-Net model to fuse the shallow features and deep features and restore the image information; according to the encoder and the decoder, the mask image of the normal red blood cell is obtained.

[0075] Reference Figure 3 , the YOLOv3 model consists of two parts: a backbone feature extraction network Darknet-53 and a prediction network. The left half is the backbone feature extraction network Darknet-53, which uses five residual modules (Residual), and the right half is the prediction network, which uses a series of convolution, upsampling and splicing operations to obtain three prediction boxes of different sizes for detecting objects of different sizes. Therefore, in this embodiment, the feature extraction network Darknet-53 uses five residual modules to extract features from the red blood cell image and the mask image of the normal red blood cells; the prediction network uses a series of convolution, upsampling and splicing operations to obtain three prediction boxes of different sizes for marking and identifying all cells in the red blood cell image and the normal red blood cells in the mask image of the normal red blood cells.

[0076] In this embodiment, the YOLOv3 model is used to perform rectangular anchor frame annotation and recognition on the normal red blood cells in the mask image of the normal red blood cells, and the number of anchor frames annotated by the normal red blood cells is obtained; the YOLOv3 model is used to perform rectangular anchor frame annotation and recognition on the red blood cells in all red blood cell images, and the number of anchor frames annotated by all red blood cells is obtained; based on the number of anchor frames annotated by the normal red blood cells and the number of anchor frames annotated by all red blood cells, the degree of oxidative damage of the red blood cells is obtained, and the degree of oxidative damage of the red blood cells is calculated by the following calculation formula:

[0077]

[0078] Wherein, d represents the degree of oxidative damage of red blood cells, x represents the number of anchor frames annotated with normal red blood cells, and y represents the number of anchor frames annotated with all red blood cells.

[0079] In this embodiment, this embodiment is based on the U-Net model and the YOLOv3 model to construct a prediction model for the degree of oxidative damage of red blood cells; the mask image of normal red blood cells is obtained by the U-Net model to improve the segmentation accuracy, and the normal red blood cells in the mask image of normal red blood cells are annotated and identified by the YOLOv3 model with rectangular anchor frames to obtain the number of anchor frames annotated by normal red blood cells; the YOLOv3 model is used to annotate and identify the red blood cells in all red blood cell images with rectangular anchor frames to obtain the number of anchor frames annotated by all red blood cells, and the YOLOv3 model is used for annotation and identification to improve the recognition accuracy and recognition speed, and the YOLOv3 model can balance the accuracy and speed well. This embodiment directly obtains the degree of oxidative damage of red blood cells through the prediction model for the degree of oxidative damage of red blood cells. Therefore, this embodiment can obtain the degree of oxidative damage of red blood cells in real time, and can also observe the real-time change state of cells intuitively through the degree of oxidative damage of red blood cells. It is more convenient to obtain the degree of oxidative damage of red blood cells in real time, save analysis time, and improve efficiency.

[0080] Step S500: using a test set to test the trained red blood cell oxidative damage degree prediction model to obtain a red blood cell oxidative damage degree test result.

[0081] Specifically, the trained prediction model of oxidative damage degree of red blood cells is tested through the test set to obtain the test result of oxidative damage degree of red blood cells. The feasibility and reliability of the prediction model of oxidative damage degree of red blood cells can be verified through the test set.

[0082] For better explanation, this example conducts experimental analysis to analyze the feasibility and reliability of the prediction model of the degree of oxidative damage of red blood cells of the present invention. The specific process is as follows:

[0083] In the experiment of this embodiment, some biochemical indicators under the conditions of red blood cell oxidation and anti-oxidation were tested respectively: detection of lipid peroxide (MDA), detection of reactive oxygen species, and detection of methemoglobin (MetHb) content. The experimental scenario of the detection is as follows:

[0084] 1. Detection of lipid peroxide (MDA)

[0085] Control group: Take 12 ml of 1% red blood cell suspension and divide it into 6 groups, each group at 1500r / min. After centrifugation for 5 minutes, discard the supernatant and place the suspensions in 100μM H2O2, 200μM H2O2, 300μM H2O2, 40μM H2O2, 450μM H2O2, and 500μM H2O2, respectively, and incubate at 37℃ for 2 hours.

[0086] For experimental group 1, 12 ml of 1% red blood cell suspension was taken and divided into 6 groups, each group was centrifuged at 1500 r / min for 5 min, the supernatant was discarded, and the suspensions were placed in 1mM C3H3NaO3+100μM H2O2, 1mM C3H3NaO3+200μM H2O2, 1mM C3H3NaO3+300μM H2O2, 1mM C3H3NaO3+400μM H2O2, 1mM C3H3NaO3+450μM H2O2, and 1mM C3H3NaO3+500μM H2O2, respectively, and incubated at 37°C for 2 h.

[0087] For experimental group 2, 12 ml of 1% red blood cell suspension was taken and divided into 6 groups, each group was centrifuged at 1500 r / min for 5 min, the supernatant was discarded, and the suspensions were placed in 50 mM C3H3NaO3 + 100 μM H2O2, 50 mM C3H3NaO3 + 200 μM H2O2, 50 mM C3H3NaO3 + 300 μM H2O2, 50 mM C3H3NaO3 + 400 μM H2O2, 50 mM C3H3NaO3 + 450 μM H2O2, and 50 mM C3H3NaO3 + 500 μM H2O2, respectively, and incubated at 37 °C for 2 h.

[0088] For experimental group 3, 12 ml of 1% red blood cell suspension was taken and divided into 6 groups, each group was centrifuged at 1500 r / min for 5 min, the supernatant was discarded, and the suspensions were placed in 50 mM C3H3NaO3 + 50 μM H2O2, 50 mM C3H3NaO3 + 60 μM H2O2, 50 mM C3H3NaO3 + 70 μM H2O2, 50 mM C3H3NaO3 + 80 μM H2O2, 50 mM C3H3NaO3 + 90 μM H2O2, and 50 mM C3H3NaO3 + 100 μM H2O2, respectively, and incubated at 37 °C for 2 h.

[0089] All mixed solutions must be fully mixed and left to stand for 15 minutes before adding cells; the control group and the experimental group simultaneously used a lipid peroxidation (MDA) assay kit (Lipid Peroxidation MDA Assay Kit) to quantitatively detect the MDA produced by red blood cells under different treatment conditions.

[0090] 2. Detection of Reactive Oxygen Species

[0091] Blank group: 3 ml of 1% red blood cell suspension was taken without constant temperature incubation;

[0092] Control group: Take 3 ml of 1% red blood cell suspension and incubate at 37℃ for 24 hours.

[0093] For experimental group 1, 12 ml of 1% red blood cell suspension was taken and divided into 6 groups, each group was centrifuged at 1500 r / min for 5 minutes, the supernatant was discarded, and the suspensions were placed in 50 mM C3H3NaO3, 40 mM C3H3NaO3, 50 mM C3H3NaO3+50 μM H2O2, 50 mM C3H3NaO3+60 μM H2O2, 50 mM C3H3NaO3+70 μM H2O2, and 50 mM C3H3NaO3+80 μM H2O2, respectively, and incubated at 37°C for 2 hours. All mixed solutions had to be fully mixed and left to stand for 15 minutes before adding cells.

[0094] For experimental group 2, 10 ml of 1% red blood cell suspension was taken and divided into 5 groups, each group was centrifuged at 1500 r / min for 5 minutes, the supernatant was discarded, and the suspensions were placed in 50 mM C3H3NaO3+50 μM H2O2, 40 mM C3H3NaO3+50 μM H2O2, 30 mM C3H3NaO3+70 μM H2O2, 20 mM C3H3NaO3+80 μM H2O2, and 5 mM C3H3NaO3+80 μM H2O2, respectively, and incubated at 37°C for 2 hours.

[0095] All mixed solutions must be fully mixed and left to stand for 15 minutes before adding cells. The control group and the experimental group were stained with a Reactive Oxygen Species Assay Kit and detected by flow cytometry.

[0096] 3. Detection of hemoglobin content

[0097] Take 10ml of 1% red blood cell suspension and divide it into 5 groups, each group is centrifuged at 1500r / min for 5min, and the supernatant is discarded. The suspension is placed in 50mM C3H3NaO3, 40mM C3H3NaO3, 50mM C3H3NaO3+50μM H2O2, 50mM C3H3NaO3+100μM H2O2, and PBS and incubated at 37℃ for 2h. Then take the incubated red blood cells, break them with water, centrifuge at 12000r / min for 3 to 5min to precipitate the red blood cell membrane, take the supernatant, scan it on an ultraviolet spectrophotometer (UV-1900) from a wavelength of 500nm to 700nm, and calculate the content of methemoglobin by the following formula:

[0098] MetHb=279*A680-3.0*A577

[0099] Wherein, MetHb represents the content of methemoglobin, A680 represents the absorbance value at a wavelength of 680 nm, and A577 represents the absorbance value at a wavelength of 577 nm.

[0100] Based on the above experimental scenario, this embodiment measures the reactive oxygen content, cell clustering and methemoglobin content of cells under flow cytometry, and can obtain the proportion of normal red blood cell clustering measured under flow cytometry; lipid peroxides (MDA) are also measured through the red blood cell oxidative damage degree prediction model to obtain the ratio of lipid peroxides (MDA) to normal red blood cells.

[0101] Reference Figure 4 , R 2 represents the degree of fit, Y=1.292X-0.282 represents the linear fitting function, the vertical axis in the figure represents the normal red blood cell ratio measured by flow cytometry, and the horizontal axis in the figure represents the normal red blood cell ratio measured by the red blood cell oxidative damage prediction model. It can be seen from the figure that the R of the fitting line of the relationship between the normal red blood cell ratio measured by flow cytometry and the red blood cell oxidative damage prediction model is 2The value is 0.998, and there is a good linear correlation between the two. The normal red blood cell ratio measured by flow cytometry is divided by the normal red blood cell ratio after identification by the prediction model of red blood cell oxidative damage degree to obtain the consistency between the two. And through a series of experiments, it was proved that the ratio of normal red blood cells calculated after the prediction model of the degree of oxidative damage of red blood cells was identified and the results of the normal red blood cell population ratio measured by flow cytometry were consistent with 94.2%. Specifically, the ratio of normal red blood cells in different antioxidant groups under flow cytometry was: the ratio of red blood cells with normal morphology under 50mM sodium pyruvate was: 94.47%; the ratio of red blood cells with normal morphology under 50μM hydrogen peroxide plus 50mM sodium pyruvate was: 91.96%; the ratio of red blood cells with normal morphology under 50μM hydrogen peroxide plus 60mM sodium pyruvate was: 89.17%; the ratio of red blood cells with normal morphology under 50μM hydrogen peroxide plus 70mM sodium pyruvate was: 69.49%; the ratio of red blood cells with normal morphology under 50μM hydrogen peroxide plus 80mM sodium pyruvate was: 47.95%. The ratio of normal red blood cells calculated by different oxidation groups after the prediction model of the degree of oxidative damage of red blood cells was identified was: The normal morphology of red blood cells under 0mM sodium pyruvate is 95%; the normal morphology of red blood cells under 50μM hydrogen peroxide plus 50mM sodium pyruvate is 93%; the normal morphology of red blood cells under 50μM hydrogen peroxide plus 60mM sodium pyruvate is 90%; the normal morphology of red blood cells under 50μM hydrogen peroxide plus 70mM sodium pyruvate is 75%; the normal morphology of red blood cells under 50μM hydrogen peroxide plus 80mM sodium pyruvate is: 60%, corresponding to the values ​​of the normal red blood cell ratio measured by the flow cytometer under different antioxidant groups divided by the normal red blood cell ratio after intelligent image recognition (consistency) were: 0.994, 0.988, 0.990, 0.927, 0.799, with an average value of 0.942, that is, the consistency between the normal red blood cell ratio calculated after the red blood cell oxidative damage prediction model was identified and the normal red blood cell population proportion measured by the flow cytometer was 94.2%.

[0102] In this embodiment, the above results prove that it is feasible and reliable to characterize the degree of oxidative damage to red blood cells by the image method (i.e., the prediction model of the degree of oxidative damage to red blood cells). This method can instantly obtain the degree of oxidative damage to red blood cells, and can intuitively observe the real-time changing state of cells through the degree of oxidative damage to red blood cells. In addition, it is relatively quick to instantly obtain the degree of oxidative damage to red blood cells.

[0103] Reference Figure 5 The embodiment of the present invention further provides a system for acquiring the degree of oxidative damage of red blood cells based on deep learning, comprising:

[0104] The label image acquisition unit 100 is used to acquire a red blood cell image, mark a normal red blood cell image in the red blood cell image to form a label, and obtain a label image of the normal red blood cell;

[0105] The data set division unit 200 is used to divide the data set into a training set and a test set by taking the red blood cell image and the label image of the normal red blood cell as the data set, wherein the training set and the test set both contain the red blood cell image and the label image of the normal red blood cell;

[0106] A prediction model building unit 300 is used to build a prediction model for the degree of oxidative damage of red blood cells based on a U-Net model and a YOLOv3 model;

[0107] The prediction model training unit 400 is used to train the prediction model of the degree of oxidative damage of red blood cells using the training set to obtain the trained prediction model of the degree of oxidative damage of red blood cells. The trained prediction model of the degree of oxidative damage of red blood cells is obtained by the following method:

[0108] The preset U-Net model is trained using the training set to obtain the mask image of normal red blood cells;

[0109] The YOLOv3 model is used to annotate and identify the normal red blood cells in the mask image of the normal red blood cells with rectangular anchor frames, and the number of anchor frames annotated with the normal red blood cells is obtained;

[0110] The YOLOv3 model is used to annotate and identify the red blood cells in all red blood cell images with rectangular anchor boxes, and the number of anchor boxes annotated with all red blood cells is obtained;

[0111] Based on the number of anchor frames labeled with normal red blood cells and the number of anchor frames labeled with all red blood cells, the degree of red blood cell oxidative damage is obtained;

[0112] The prediction model testing unit 500 is used to test the trained red blood cell oxidative damage degree prediction model using a test set to obtain a red blood cell oxidative damage degree test result.

[0113] In some embodiments, the label image acquisition unit further includes:

[0114] Leica laser confocal microscope was used to take photos of red blood cell morphology, and the images were exported in a fixed pic.tif format as the original images of red blood cells.

[0115] In some embodiments, the parameter settings for shooting with a Leica laser confocal microscope include Gain settings ranging from 300 to 350, Offset settings ranging from -0.15, 552nm laser settings ranging from 1.4 to 2.60, Format settings ranging from 1024×1024, Speed ​​settings ranging from 400, and Zoom factor settings ranging from 1 to 1.5.

[0116] It should be noted that since the system for acquiring the degree of oxidative damage of red blood cells based on deep learning in this embodiment and the method for acquiring the degree of oxidative damage of red blood cells based on deep learning described above are based on the same inventive concept, the corresponding contents in the method embodiment are also applicable to the system embodiment and will not be described in detail here.

[0117] An embodiment of the present invention also provides a device for acquiring the degree of oxidative damage of red blood cells based on deep learning, comprising: at least one control processor and a memory for communicating with the at least one control processor.

[0118] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0119] The non-transient software program and instructions required to implement the method for obtaining the degree of oxidative damage of red blood cells based on deep learning in the above embodiment are stored in the memory. When executed by the processor, the method for obtaining the degree of oxidative damage of red blood cells based on deep learning in the above embodiment is executed, for example, the above described method is executed. Figure 1 The method comprises steps S100 to S500.

[0120] The system embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, i.e., may be located in one place, or may be distributed over multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0121] The embodiment of the present invention further provides a computer-readable storage medium, which stores computer-executable instructions. The computer-executable instructions are executed by one or more control processors, so that the one or more control processors can execute a method for obtaining the degree of oxidative damage of red blood cells based on deep learning in the above method embodiment, for example, executing the above described Figure 1 The functions of method steps S100 to S500 in the method.

[0122] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a general hardware platform. Those skilled in the art can understand that all or part of the processes in the above embodiment method can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and the program can include the process of the embodiment of the above method when executed. Among them, the storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

[0123] The embodiments of the present invention are described in detail above in conjunction with the accompanying drawings, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge scope of ordinary technicians in the relevant technical field without departing from the purpose of the present invention.

Claims

1. A method for obtaining the degree of oxidative damage of red blood cells based on deep learning, characterized in that: Includes steps: Acquire a red blood cell image, mark a normal red blood cell image in the red blood cell image to form a label, and obtain a label image of the normal red blood cell; The red blood cell image and the label image of the normal red blood cell are used as a data set, and the data set is divided into a training set and a test set, wherein both the training set and the test set contain the red blood cell image and the label image of the normal red blood cell; Based on the U-Net model and the YOLOv3 model, a prediction model for the degree of oxidative damage of red blood cells was constructed; The training set is used to train the red blood cell oxidative damage degree prediction model to obtain a trained red blood cell oxidative damage degree prediction model, and the trained red blood cell oxidative damage degree prediction model is obtained by the following method: Using the training set to train a preset U-Net model to obtain a mask image of the normal red blood cells; Using the YOLOv3 model to perform rectangular anchor frame annotation and recognition on the normal red blood cells in the mask image of the normal red blood cells, and obtaining the number of anchor frames annotated with the normal red blood cells; Using the YOLOv3 model to perform rectangular anchor box annotation and recognition on the red blood cells in all the red blood cell images, and obtaining the number of anchor boxes annotated with all the red blood cells; Obtaining the degree of oxidative damage of red blood cells based on the number of anchor frames labeled with the normal red blood cells and the number of anchor frames labeled with all red blood cells; The test set is used to test the trained red blood cell oxidative damage degree prediction model to obtain a red blood cell oxidative damage degree test result.

2. The method for obtaining the degree of oxidative damage of red blood cells based on deep learning according to claim 1, characterized in that: The acquiring of the red blood cell image comprises: Leica laser confocal microscope was used to take photos of red blood cell morphology, and the red blood cell images were exported in a fixed tif format.

3. The method for obtaining the degree of oxidative damage of red blood cells based on deep learning according to claim 1, characterized in that: The U-Net model includes an encoder and a decoder. The training set is used to train the preset U-Net model to obtain the mask image of the normal red blood cells, which specifically includes the steps of: Inputting the red blood cell image and the label image of the normal red blood cells into the encoder of the U-Net model; Using a convolution operation in the encoder, and using multiple continuous downsampling to gradually extract feature information of the red blood cell image and the label image of the normal red blood cells; The feature information is subjected to multiple upsampling by the decoder of the U-Net model to fuse shallow features and deep features and restore image information; According to the encoder and the decoder, a mask image of the normal red blood cells is obtained.

4. The method for obtaining the degree of oxidative damage of red blood cells based on deep learning according to claim 3, characterized in that: The YOLOv3 model includes a feature extraction network Darknet-53 and a prediction network. The YOLOv3 model is used for labeling and recognition, specifically including the following steps: The feature extraction network Darknet-53 uses five residual modules to extract features from the red blood cell image and the mask image of the normal red blood cells; The prediction network uses a series of convolution, upsampling and splicing operations to obtain prediction boxes of three different sizes for labeling and identifying all cells in the red blood cell image and normal red blood cells in the mask image of the normal red blood cells.

5. The method for obtaining the degree of oxidative damage of red blood cells based on deep learning according to claim 1, characterized in that: The degree of oxidative damage of red blood cells was calculated by the following calculation formula: Wherein, d represents the degree of oxidative damage of the red blood cells, x represents the number of anchor frames annotated with the normal red blood cells, and y represents the number of anchor frames annotated with all the red blood cells.

6. A system for acquiring the degree of oxidative damage of red blood cells based on deep learning, characterized in that: include: A label image acquisition unit is used to acquire a red blood cell image, mark a normal red blood cell image in the red blood cell image to form a label, and obtain a label image of the normal red blood cell; a data set division unit, configured to use the red blood cell image and the label image of the normal red blood cell as a data set, and divide the data set into a training set and a test set, wherein both the training set and the test set contain the red blood cell image and the label image of the normal red blood cell; A prediction model building unit, used to build a prediction model for the degree of oxidative damage of red blood cells based on the U-Net model and the YOLOv3 model; The prediction model training unit is used to train the prediction model of the degree of oxidative damage of red blood cells using the training set to obtain a trained prediction model of the degree of oxidative damage of red blood cells. The trained prediction model of the degree of oxidative damage of red blood cells is obtained by: Using the training set to train a preset U-Net model to obtain a mask image of the normal red blood cells; Using the YOLOv3 model to perform rectangular anchor frame annotation and recognition on the normal red blood cells in the mask image of the normal red blood cells, and obtaining the number of anchor frames annotated with the normal red blood cells; Using the YOLOv3 model to perform rectangular anchor box annotation and recognition on the red blood cells in all the red blood cell images, and obtaining the number of anchor boxes annotated with all the red blood cells; Obtaining the degree of oxidative damage of red blood cells based on the number of anchor frames labeled with the normal red blood cells and the number of anchor frames labeled with all red blood cells; The prediction model testing unit is used to test the trained red blood cell oxidative damage degree prediction model using the test set to obtain a red blood cell oxidative damage degree test result.

7. The system for acquiring the degree of oxidative damage of red blood cells based on deep learning according to claim 6, characterized in that: The label image acquisition unit further includes: A Leica laser confocal microscope was used to take photos of the morphology of red blood cells, and the images were exported in a fixed tif format, which were used as the original images of the red blood cells.

8. The system for acquiring the degree of oxidative damage of red blood cells based on deep learning according to claim 7, characterized in that: The parameter settings of the Leica laser confocal microscope include Gain setting range of 300 to 350, Offset setting range of -0.15, 552nm laser setting range of 1.4 to 2.60, Format setting range of 1024×1024, Speed ​​setting range of 400, and Zoom factor setting range of 1 to 1.

5.

9. A device for acquiring the degree of oxidative damage of red blood cells based on deep learning, characterized in that: comprising at least one control processor and a memory for communicatively coupling with the at least one control processor; The memory stores instructions that can be executed by the at least one control processor, and the instructions are executed by the at least one control processor so that the at least one control processor can execute a method for obtaining the degree of oxidative damage of red blood cells based on deep learning as described in any one of claims 1 to 5.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute a method for obtaining the degree of oxidative damage of red blood cells based on deep learning as described in any one of claims 1 to 5.

Images