Epithelial cell recognition AI training and visible component quantitative analysis method
By preprocessing urine samples and conducting AI training, combined with manual labeling, epithelial cells in urine can be identified and quantitatively analyzed, solving the problem of difficult sample collection and achieving efficient and accurate urine testing.
Patent Information
- Application Number
- CN202410771900.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2024-04-11
- Filing Date
- 2024-06-15
- Publication Date
- 2025-10-21
AI Technical Summary
In existing technologies, the detection of epithelial cells in urine is difficult to carry out effectively, especially the difficulty in collecting samples of renal tubular epithelial cells, transitional epithelial cells, and squamous epithelial cells, resulting in high AI training costs and low detection efficiency.
By pre-processing urine samples, including adding dyes and reference particles, obtaining microscopic samples after sedimentation and stratification, combining AI training and manual review, identifying and labeling epithelial cells, and using AI feature data sets for training and analysis, the impact of sample tiling height on quantitative analysis is eliminated.
It achieves accurate identification and quantitative analysis of epithelial cells in urine, reduces AI training costs, improves detection efficiency and accuracy, provides research sample support, and simplifies the urine testing process.
Smart Images
Figure CN120823596A_ABST
Abstract
Description
Technical Field
[0001] The present application belongs to the technical field of analysis of tangible elements based on microscopically magnified images, and particularly relates to an AI training method for identifying epithelial cells in urine, an epithelial cell analysis method, and a tangible element quantitative analysis method and computing, processing, and storage device. Background Art
[0002] The applicant filed a series of Chinese patents, such as
[0003] 1. CN2020112669290, “Cell analysis method and system and quantitative method and system”;
[0004] 2. CN2020112669182, “Cell Suspension Sample Imaging Method, System, and Kit”;
[0005] 3. CN2022104799126, “Fast focusing method for microscopic image acquisition device and microscopic image detection method”;
[0006] 4. CN2023100423151, “Blood imaging analysis system and method”;
[0007] Use a new technical solution to measure the content of formed elements in blood and urine samples.
[0008] Epithelial cells in urine are generally divided into three types: tubular epithelial cells, transitional epithelial cells, and stratified squamous epithelial cells.
[0009] 1. Renal tubular epithelial cells: They originate from the renal tubules and appear in the urine, which usually indicates renal tubular lesions. After kidney transplantation, it is necessary to monitor the epithelial cells in the urine to determine whether there is a rejection reaction; renal tubular epithelial cells refer to a layer of cells on the outside of the renal tubules. Renal tubular epithelial cells originate from the distal and proximal renal tubules. If the number of epithelial cells in the urine routine is high, it mainly indicates the possibility of acute glomerulonephritis, tubular necrosis, and interstitial nephritis. When acute tubular necrosis leads to acute renal failure, the most basic manifestation is the shedding of renal tubular epithelial cells, exposing the basement membrane. If the basement membrane is intact, the possibility of recovery from acute renal vascular necrosis is relatively high. If the basement membrane of the renal tubular epithelial cells is severely damaged, the possibility of recovery is small.
[0010] 2. Transitional epithelial cells: There are three layers in total. The surface transitional epithelial cells originate from the bladder, the middle transitional epithelial cells originate from the renal pelvis, and the bottom transitional epithelial cells originate from the ureter, bladder, and urethra. Normal people do not have transitional epithelial cells in their urine, or a small number of transitional epithelial cells can be seen occasionally. If transitional epithelial cells appear in urine, it usually indicates diseases such as pyelonephritis, cystitis, and urethritis. If they appear in large numbers, consider whether there is transitional epithelial carcinoma.
[0011] 3. Stratified squamous epithelial cells: These cells are primarily located in the anterior urethra and, in women, may be mixed with stratified squamous epithelial cells from the vagina. If stratified squamous epithelial cells are present in urine and are accompanied by white blood cells or pus cells, a urinary tract infection, particularly urethritis, is usually suspected.
[0012] 4. Basal transitional epithelial cells: Located in the basal or deeper layers of the epithelium, these cells are rounder in shape and smaller in size, slightly larger than tubular epithelial cells. Their nuclei and cytoplasm are smaller than those of tubular epithelial cells, and they have fewer cytoplasmic granules. They originate from the deeper layers of the ureters, bladder, and urethra. An increase in basal transitional epithelial cells, accompanied by leukocytosis, indicates urethral inflammation or severe inflammation, possibly even necrotizing lesions.
[0013] 5. Middle transitional epithelial cells: These cells vary in size, often showing fish, pear, spindle, or tadpole shapes. They are 20 to 40 μm long, with round or oval nuclei, often located on one side. Sources: renal pelvis, ureter, and bladder neck. An increase in middle transitional epithelial cells suggests pyelonephritis.
[0014] 6. Surface transitional epithelial cells: Located in the surface layer of the urothelium, these cells are large, 4 to 6 times larger than white blood cells, often irregularly round in shape, with small, often centrally located nuclei and abundant cytoplasm. Origin: Bladder. Increased numbers of surface transitional epithelial cells indicate acute or chronic cystitis.
[0015] 7. Squamous epithelial cells: Squamous epithelial cells often have an irregular, polygonal, and flat, thin cell body with many folded edges. They have small, round or oval nuclei and abundant cytoplasm. Origin: Urogenital tract. An increase in squamous epithelial cells suggests urinary tract inflammation.
[0016] In the microscopic imaging + AI analysis technology system, AI training requires a large number of samples. However, the probability of renal tubular epithelial cells, transitional epithelial cells, squamous epithelial cells, or stratified squamous epithelial cells appearing in normal individual urine is low, making sample collection difficult and training costs very high. This makes epithelial cell detection ineffective. Summary of the Invention
[0017] In this application, the inventors proposed that by conducting AI training on microscopic sample images containing epithelial cells, epithelial cells in urine suspensions can be accurately classified and identified.
[0018] In the AI training method of this application, the manually reviewed and labeled images include epithelial cell labeling information and suspected epithelial cell labeling information, which not only improves training efficiency but also provides research samples for researchers. In particular, the network is used to aggregate the training sample collection work from various places to the server, allowing experts to be free from the front-line sample collection and directly perform labeling or research analysis remotely. Epithelial cells can effectively identify inflammation or lesions of the urogenital system. This application can effectively detect various epithelial cells and significantly reduce the amount of AI training engineering for epithelial cells, making the detection of urine epithelial cells feasible.
[0019] In this application, the inventors propose an epithelial cell analysis method that can accurately obtain the content of epithelial cells per unit volume in microscopic samples and perform accurate quantitative analysis.
[0020] In this application, the inventors propose a method for quantitative analysis of tangible elements. The method utilizes the condition that the reference particle concentration Q is known and combines the number of reference particles in the selected area Sn of the image to obtain the volume of the microscopic sample, thereby obtaining the volume of the microscopic sample for quantitative analysis of the number of target detection objects. The volume of the microscopic sample can be obtained indirectly, eliminating the influence of the microscopic sample tiling height H on the quantitative analysis.
[0021] The technical solution of the present application to solve the above-mentioned technical problems is a method for AI training to identify epithelial cells, wherein a urine sample is preprocessed to obtain a microscopic sample; the microscopic sample is flattened; the formed elements are allowed to settle to the bottom layer; the flattened microscopic sample is photographed to obtain a microscopic sample image; the epithelial cells in the microscopic sample image are identified and labeled to obtain a labeled picture; the above-mentioned labeled picture includes epithelial cell labeling information, and AI training is performed using the labeled picture to obtain an epithelial cell AI feature dataset A.
[0022] The pretreatment of urine samples includes any of the following technical features: TA1: adding a dye to the urine sample to dye the urine; TA2: adding a dye to the urine sample to dye the urine; the dye is a dry powder dye; TA3: adding reference particles to the urine sample, and the reference particles are located at the bottom of the microscopic sample; TA4: when photographing the flat microscopic sample, the focal plane of the camera is aligned with the center of the epithelial cells; TA5: adding a dye to the urine sample to dye the urine; the dye includes a new methylene blue component; TA6: mixing multiple urine samples, letting them settle and separate After the urine is layered, the bottom layer is taken to obtain a sample for microscopic examination; TA7: multiple urine samples are mixed, centrifuged and layered, the bottom layer is taken to obtain a sample for microscopic examination; TA8: urine samples of biological individuals with increased epithelial cells are taken; TA9: dyes are added to the urine sample to dye the urine; the dye is a dry powder particle dye; TA10: dyes are added to the urine sample to dye the urine; the dye is a dry powder particle dye; the dry powder particle dye is obtained by quantitatively dividing and drying a liquid dye reagent; the liquid dye reagent includes reference particles, dye and water, wherein the target concentration range of the reference particles is adjustable.
[0023] Epithelial cells are added to a urine sample and mixed to obtain a microscopic sample; obtaining epithelial cells includes any one of the following technical features: TD1: cutting open the animal kidney to expose the renal cortex, and obtaining cells from the renal cortex; TD2: adding water to the animal kidney and crushing it to obtain a suspension, precipitating and stratifying the suspension to obtain epithelial cells; TD3: cutting open the animal urinary system epithelial tissue sample and obtaining cells from the cortex.
[0024] The epithelial cells are any one or more of renal tubular epithelial cells, transitional epithelial cells, and squamous epithelial cells. The transitional epithelial cells are any one or more of basal transitional epithelial cells, middle transitional epithelial cells, and surface transitional epithelial cells.
[0025] The above-mentioned AI training method for identifying epithelial cells includes any one of the following technical features: TB1: the above-mentioned microscopic sample images are sent to the AI training server via the network; TB2: the above-mentioned microscopic sample images are sent to the AI training server via the network, and the segmented images are manually identified and labeled through the network.
[0026] Use epithelial cell AI feature dataset A as the feature dataset, use AI software to identify one or more pictures input into AI recognition, and obtain recognition output pictures. The above-mentioned recognition output pictures include recognized epithelial cell images. The recognition output pictures are manually reviewed to form manually reviewed marked pictures. The manually reviewed marked pictures are used for AI training to obtain epithelial cell AI feature dataset B; the above-mentioned manually reviewed marked pictures include epithelial cell labeling information and suspected epithelial cell labeling information.
[0027] In the above-mentioned AI training method for identifying epithelial cells, the epithelial cells are any one or more of renal tubular epithelial cells, transitional epithelial cells, and squamous epithelial cells.
[0028] In the above-mentioned AI training method for identifying epithelial cells, the transitional epithelial cells are any one or more of the bottom transitional epithelial cells, middle transitional epithelial cells, and surface transitional epithelial cells.
[0029] A method for analyzing epithelial cells comprises preprocessing a urine sample to obtain a microscopic sample; flattening the microscopic sample; photographing the flattened microscopic sample to obtain a microscopic sample image; and using an AI recognition algorithm to identify the microscopic sample image and epithelial cells in the image; the AI recognition algorithm includes an epithelial cell AI feature data set, which includes feature data of any one or more of renal tubular epithelial cells, transitional epithelial cells, and squamous epithelial cells.
[0030] In the above-mentioned epithelial cell analysis method, the sample tiling height is H; a microscopic sample image is obtained, and an image area S1 is selected; epithelial cells in the urine in the selected image area S1 are identified, and the number of epithelial cells NUMS1 is obtained; in the microscopic sample, the content of epithelial cells per unit volume = NUMS1 / (S1×H).
[0031] In the above-mentioned epithelial cell analysis method, the volume of the microscopic sample is VT; the total test area of the microscopic sample is SA; the total test area is the total area of the epithelial cells of the microscopic sample collected and precipitated; an image of the microscopic sample is obtained, and the image area S1 is selected; epithelial cells in the urine in the selected image area S1 are identified, and the number of epithelial cells NUMS1 is obtained; in the microscopic sample, the content of epithelial cells per unit volume = SA × (NUMS1 / S1) / VT.
[0032] The above-mentioned epithelial cell analysis method, including the above-mentioned epithelial cell AI feature dataset, is obtained by training after manually annotating epithelial cell images in urine.
[0033] In the above-mentioned epithelial cell analysis method, the above-mentioned urine sample is pretreated and the concentration is concentrated N times. In the urine sample, the content of epithelial cells per unit volume = the content of epithelial cells per unit volume / N; the content of epithelial cells per unit volume = NUMS1 / (S1×H×N); the content of epithelial cells per unit volume = (SA×(NUMS1 / S1) / VT) / N.
[0034] In the above-mentioned epithelial cell analysis method, the urine sample is pre-treated and reference particles are added, and the reference particles are used to assist in focus adjustment.
[0035] In the above-mentioned epithelial cell analysis method, the urine sample is pre-treated and a dye is added thereto, and the dye dyes the urine sample.
[0036] In the above-mentioned epithelial cell analysis method, the urine sample is pre-treated and a dye is added thereto, and the dye dyes the urine sample; the dye is a dry powder dye.
[0037] In the above-mentioned epithelial cell analysis method, the urine sample is pre-treated and a dye is added, and the dye includes a new methylene blue component.
[0038] In the above-mentioned epithelial cell analysis method, the epithelial cells are any one or more of renal tubular epithelial cells, transitional epithelial cells, and squamous epithelial cells; in the above-mentioned epithelial cell analysis method, the epithelial cells are any one or more of basal transitional epithelial cells, middle transitional epithelial cells, and surface transitional epithelial cells.
[0039] A method for quantitative analysis of tangible elements, comprising: adding reference particles to a sample to obtain a microscopic sample, wherein the concentration Q of the reference particles in the microscopic sample is known; the microscopic sample is flattened, photographed, and two or more microscopic sample images are obtained, and an image area Sn is selected in image n; an AI recognition algorithm is used to recognize the microscopic sample image, and target detection objects in the selected image area Sn are identified to obtain the number NUM_Sn of target detection objects; the number NUM_S_All of target detection objects in all selected images is equal to the sum of the number NUM_Sn of each target detection object; the AI recognition algorithm is used to recognize the microscopic sample image, and reference particles in the selected image area Sn are identified to obtain the number NUM_Pn of reference particles; the number NUM_P_All of particles in all selected images is equal to the sum of the number NUM_Pn of each target detection object; the volume V2 of the microscopic sample corresponding to the selected image area is equal to NUM_P_All / reference particle concentration Q; the content of the target detection object per unit volume in the microscopic sample is equal to NUM_S_All / V2.
[0040] The above-mentioned method for quantitative analysis of tangible elements includes any one of the following technical features: TH1: When the reference particles in the selected area Sn of the image are identified, the number of reference particles at the boundary of the selected area Sn of the image is the number of cross-border reference particles NUM_P_B, and half of the number of cross-border reference particles NUM_P_B is taken and included in NUM_P_All; TH2: The above-mentioned sample includes any one of a blood sample, a fecal sample, a urine sample or other biological sample; TH3: The above-mentioned target detection object includes any one of red blood cells, white blood cells, platelets, urine casts, urine crystals, urinary tract epithelial cells, urine microorganisms, urinary tract parasite eggs, intestinal parasite eggs, intestinal protozoa, intestinal microorganisms, intestinal epithelial cells, urine lipid droplets, and intestinal undigested matter.
[0041] The above-mentioned quantitative analysis method of tangible elements includes any one of the following technical features: TF1: the range of the concentration Q of the above-mentioned reference particles is 1400 particles / ul to 2000 particles / ul; TF2: the above-mentioned reference particles are floating particles, and the specific gravity relative to water is in the range of 0.9 to 0.95; TF3: the above-mentioned reference particles are bottom particles, and the specific gravity relative to water is in the range of 1.05 to 1.1; TF4: the range of the diameter of the above-mentioned reference particles is 1um to 10um; TF5: the preferred diameter of the above-mentioned reference particles is 5um.
[0042] A computing and processing device, comprising any one of the following technical features: TEA1: used to run all or part of the above-mentioned AI training method for identifying epithelial cells according to any one of claims 1 to 5; TEA2: the memory of the above-mentioned computing and processing device includes the above-mentioned epithelial cell AI feature dataset A according to claims 1 to 5; TEA3: the memory of the above-mentioned computing and processing device includes the above-mentioned epithelial cell AI feature dataset B according to claim 5; TEA4: used to run all or part of the above-mentioned epithelial cell analysis method according to any one of claims 6 to 7; TEA5: used to run all or part of the above-mentioned method for quantitative analysis of formed elements according to any one of claims 8 to 10.
[0043] A data storage device, comprising any one of the following technical features: TEB1: storing all or part of the program code for executing any one of the above-mentioned AI training methods for identifying epithelial cells according to claims 1 to 7; TEB2: storing the above-mentioned epithelial cell AI feature data set A according to claims 1 to 6; TEB3: storing the above-mentioned epithelial cell AI feature data set B according to claim 7; TEB4: storing all or part of the program code for executing any one of the above-mentioned epithelial cell analysis methods according to claims 6 to 7; TEB5: storing all or part of the program code for running any one of the above-mentioned tangible element quantitative analysis methods according to claims 8 to 10.
[0044] A detection device comprising any one of the following technical features: TEC1: used to run all or part of the above-mentioned AI training method for identifying epithelial cells according to any one of claims 1 to 5; TEC4: used to run all or part of the above-mentioned epithelial cell analysis method according to any one of claims 6 to 7; TEC5: used to run all or part of the above-mentioned quantitative analysis method of formed elements according to any one of claims 8 to 10.
[0045] The technical effects of the above technical solution include: using advanced AI computing power to complete the identification of epithelial cells, making the analysis of the formed components to be identified at the bottom more efficient and more accurate.
[0046] The technical effects of the above technical solution include: encapsulating the trained epithelial cell AI feature dataset A into independent data, which can be sold separately, reducing the engineering difficulty of the entire industry and allowing new technologies to be promoted and implemented as soon as possible.
[0047] The technical effects of the above technical solution include: the epithelial cell AI feature dataset B evolved from the epithelial cell AI feature dataset A is encapsulated into independent data, so that the AI recognition capability can be iterated and improved.
[0048] The technical effects of the above technical solution include: the purpose of urine staining is to detect cells, casts and other tangible objects in urine. When training the AI model, using the same staining can improve the recognition accuracy.
[0049] The technical effects of the above technical solution include: directly adding the dry powder dye, thereby avoiding the problem of increasing the urine volume due to adding the liquid dye.
[0050] The technical effects of the above technical solution include: the addition of reference particles makes the microscopic imaging of urine more efficient, solving the problem that it is difficult to quickly complete focusing due to the relatively low content of formed components in urine.
[0051] The technical effects of the above technical solution include: the dry powder particle dye can be added with the reference particles and the dye at the same time, and the efficiency of sample pretreatment is higher.
[0052] The technical effects of the above technical solution include: the dry powder particle dye is obtained by quantitatively dividing and drying the liquid dye reagent; the liquid dye reagent can be weighed based on a large scale, achieving more accurate and more uniform quantification of reference particles and dyes. A single portion of dry powder particle dye is usually at the microliter level for extremely small samples, and the corresponding single portion of dry powder particle dye is also very small, which is difficult to weigh accurately, making it difficult to ensure the accuracy of the reference particles and dye concentrations. Under the condition that a single portion of dry powder particle dye is difficult to weigh accurately, the above method can obtain a precise single portion of dry powder particle dye, achieve more accurate and more uniform quantification of reference particles and dyes, and ensure the accuracy of the subsequent quantitative analysis benchmark.
[0053] The technical effects of the above technical solution include: the focal plane of the camera is aligned with the center position of the epithelial cells, which can obtain the clearest image of the epithelial cells, which is beneficial for subsequent recognition training.
[0054] The technical benefits of the above-mentioned technical solution include: by mixing multiple urine samples to obtain microscopic samples, the efficiency of obtaining epithelial cell samples is improved. Since not all samples contain epithelial cells, the above-mentioned method can improve the efficiency of obtaining characteristic data.
[0055] The technical effects of the above technical solution include: after multiple urine samples are mixed and subjected to static sedimentation stratification or routine stratification, the bottom layer is taken to obtain a microscopic examination sample for use as a microscopic examination sample, thereby further improving the efficiency of obtaining epithelial cell characteristic data.
[0056] The technical effects of the above technical solution include: taking urine samples from biological individuals with increased epithelial cells, further improving the efficiency of obtaining characteristic data.
[0057] The technical effects of the above technical solution include: adding kidney cells to urine samples to obtain microscopic samples; and further improving the efficiency of obtaining characteristic data.
[0058] The technical effects of the above technical solution include: microscopic sample images are sent to the AI training server through the network, the segmented images are manually identified and labeled through the network, and the identified output images are manually reviewed and marked through the network. This can aggregate and utilize more resources, improve the efficiency of acquiring feature data, and improve the accuracy of AI training and analysis.
[0059] The technical effects of the above technical solution include: epithelial cell AI feature dataset B, which is derived from epithelial cell AI feature dataset A, is encapsulated into independent data, enabling iterative and improved AI recognition capabilities. Suspected epithelial cell information provides clinical research material.
[0060] The technical effects of the above technical solution include: the epithelial cell AI feature data set is obtained by training after manually annotating epithelial cell images in urine, and the accuracy of AI recognition is more guaranteed.
[0061] The technical effects of the above technical solution include: defining a new parameter, the number of epithelial cells in a unit volume of urine = the number of epithelial cells in urine in the urine sample volume / the urine sample volume.
[0062] The technical effects of the above technical solution include: identifying and calculating the number of epithelial cells NUMS1 in the microscopic sample image, providing basic parameters for quantitative analysis of the formed elements to be identified at the bottom, and based on the AI recognition algorithm, it has high efficiency and accuracy.
[0063] The technical effects of the above technical solution include: based on classification recognition, combined with image-based sample quantification, accurate epithelial cell counting is achieved, which takes the accurate quantitative analysis of formed elements in urine a step further and can perform precise quantitative analysis of epithelial cells.
[0064] The technical effects of the above technical solution include: the volume quantitative calculation process of the tangible component quantitative analysis method is determined based on the reference particle concentration Q and the number of reference particles in the selected image area Sn, rather than based on the sample tiling height H, eliminating the error caused by the deviation of the sample tiling height H in the calculation, and the quantitative analysis is more accurate. BRIEF DESCRIPTION OF THE DRAWINGS
[0065] Figure 1 This is one of the steps in the AI training method for identifying epithelial cells;
[0066] Figure 2 This is the second step of the AI training method for identifying epithelial cells;
[0067] Figure 3 This is the third step of the AI training method for identifying epithelial cells;
[0068] Figure 4 This is one of the schematic diagrams of the urine sample pretreatment method;
[0069] Figure 5 This is the second diagram of the urine sample pretreatment method;
[0070] Figure 6 This is the third diagram of the urine sample pretreatment method;
[0071] Figure 7 This is one of the schematic diagrams of the method for obtaining dry powder particle dye;
[0072] Figure 8 This is the second schematic diagram of the method for obtaining dry powder particle dye;
[0073] Figure 9 This is the fourth diagram of the urine sample pretreatment method;
[0074] Figure 10 This is the fifth diagram of the urine sample pretreatment method;
[0075] Figure 11 This is the sixth diagram of the urine sample pretreatment method;
[0076] Figure 12 is a schematic diagram of an epithelial cell;
[0077] Figure 13 This is a schematic diagram of the distribution of various formed elements when a urine sample is imaged in a detection chip. The cells in the figure include epithelial cells.
[0078] Figure 14 This is one of the schematic diagrams of the epithelial cell analysis method;
[0079] Figure 15 This is the second schematic diagram of the epithelial cell analysis method;
[0080] Figure 16 It is a pictorial diagram of the method for quantitative analysis of formed elements;
[0081] Figure 17 It is a labeled schematic diagram of the underlying transitional epithelium;
[0082] Figure 18 It is a labeled schematic diagram of the middle transitional epithelium;
[0083] Figure 19 It is a labeled schematic diagram of the superficial transitional epithelium;
[0084] Figure 20 It is a labeled schematic diagram of squamous epithelial cells. DETAILED DESCRIPTION
[0085] The contents of this application are further described in detail below in conjunction with the accompanying drawings. It should be noted that the following is a description of the preferred embodiments of the present invention and does not constitute any limitation to the present invention. The description of the preferred embodiments of the present invention is only an illustration of the general principles of the present invention. The numbers such as "first", "second" and "A" and "B" involved in the present invention are only for the convenience of explanation and do not represent the order relationship in time or space. The combination of letters and numbers "TA", "TB" and "H" involved in the present invention are only for the convenience of explanation, and the specific meaning is determined by the specific words referred to.
[0086] like Figure 1 A method for AI training to identify epithelial cells is disclosed. The urine sample is preprocessed to obtain a microscopic sample; the microscopic sample is flattened; the formed elements are allowed to settle to the bottom layer; the flattened microscopic sample is photographed to obtain a microscopic sample image; the epithelial cells in the microscopic sample image are identified and labeled to obtain a labeled image; the labeled image includes epithelial cell labeling information, and the labeled image is used for AI training to obtain an epithelial cell AI feature dataset A.
[0087] like Figure 2 and Figure 13 When photographing a flat microscopic sample, the camera's focal plane is aligned with the center of the epithelial cell. Figure 13 Schematic diagram of the distribution of various visible elements when a urine sample is imaged in a detection chip. The visible elements in the figure include reference particles, cells, crystals, and casts, among which the cells include epithelial cells. Figure 13 As shown, you can focus from bottom to top to avoid the influence of other floating objects on the focus.
[0088] like Figure 4 Urine sample pretreatment involves adding a dye to the urine sample to stain the urine; the dye is a dry powder dye. The dye includes a new methylene blue component. Urine sample pretreatment can include adding reference particles to the urine sample, with the reference particles located at the bottom layer of the microscopic sample. Urine sample pretreatment can include obtaining urine samples from individuals with increased epithelial cells.
[0089] like Figure 5 and Figure 9, urine sample pretreatment, multiple urine samples are mixed, allowed to settle and stratify, and the bottom layer is taken to obtain a microscopic sample. Figure 6 and Figure 9 , pretreatment of urine samples, multiple urine samples are mixed, centrifuged and layered, the bottom layer is taken, and microscopic examination samples are obtained.
[0090] like Figure 5 and Figure 6 , add dye to the urine sample to dye the urine; the dye is a dry powder particle dye. Figure 5 , dry powder particle stain is a mixture of reference particles and infection stain. Figure 6 , after the reference particles, dye and water are made into liquid dye, dry powder particle dye is obtained after drying. Figure 7 After the reference particles, dye and water are made into liquid dye, the liquid dye is quantitatively divided into portions and dried to obtain a single portion of dry powder particle dye. Figure 8 ,exist Figure 7 Before the quantitative portioning, there is a quality inspection step for the liquid dye. When the particle concentration in the liquid dye meets the standard, the quantitative portioning is carried out. If the particle concentration in the liquid dye does not meet the standard, the content of the reference particles, dye and water can be further adjusted until the particle concentration meets the standard.
[0091] like Figures 6 to 9 , adding a dye to the urine sample to dye the urine; the dye is a dry powder particle dye; the dry powder particle dye is obtained by quantitatively dividing and drying a liquid dye reagent; the liquid dye reagent includes reference particles, a dye and water, wherein the target concentration range of the reference particles is adjustable.
[0092] like Figure 10 , pretreatment of urine samples, adding kidney cells to the urine sample and mixing to obtain a microscopic sample; to obtain kidney cells, the animal kidney can be cut open to expose the renal cortex, and cells from the above renal cortex can be taken; or the animal kidney can be crushed with water to obtain a suspension, and the above suspension can be precipitated and layered to obtain epithelial cells.
[0093] like Figure 12 , is a schematic diagram of epithelial cells. The image on the left shows native epithelial cells from a microscopic sample, while the image on the right shows a microscopic sample after kidney cells were added to a urine sample. The images show granular cytoplasm, a single, round nucleus that is large and prominent, and easily visible. The nuclear membrane is clear and easy to identify. As can be seen from the comparison in the images, the epithelial cells obtained by adding kidney cells have the same properties as native epithelial cells, making them suitable for AI training and significantly improving the efficiency of obtaining AI training datasets.
[0094] Figure 17 It is a labeled schematic diagram of the underlying transitional epithelium; Figure 18It is a labeled schematic diagram of the middle transitional epithelium;
[0095] Figure 19 It is a labeled schematic diagram of the superficial transitional epithelium; Figure 20 It is a labeled schematic diagram of squamous epithelial cells.
[0096] like Figures 17 to 20 In urine, there are various tangible objects unique to urine. By adding kidney cells to the urine sample and mixing them, the microscopic sample image includes not only various epithelial cells, but also the unique tangible objects in urine. The image obtained after this mixture can effectively improve the accuracy of AI recognition.
[0097] like Figure 3 , using epithelial cell AI feature dataset A as the feature dataset, using AI software to identify one or more images input into AI recognition to obtain a recognition output image, the above-mentioned recognition output image includes the recognized epithelial cell image, manually reviewing the recognition output image to form a manually reviewed and marked image, using the manually reviewed and marked image for AI training to obtain an epithelial cell AI feature dataset B; the above-mentioned manually reviewed and marked image includes epithelial cell annotation information and suspected epithelial cell annotation information. The microscopic sample image is sent to the AI training server via the network; the segmented image is manually identified and labeled via the network. The above-mentioned epithelial cell AI training method manually reviews and marks the recognition output image through the network by experts, and the experts identify the above-mentioned suspected epithelial cell information.
[0098] like Figure 14 A method for analyzing epithelial cells comprises: pre-processing a urine sample to obtain a microscopic sample; the microscopic sample is tiled at a height of H; photographing the tiled microscopic sample to obtain a microscopic sample image, and selecting an image area S1; using an AI recognition algorithm to identify the microscopic sample image, identifying epithelial cells in the urine within the selected image area S1, and obtaining the number of epithelial cells NUMS1; the AI recognition algorithm includes an epithelial cell AI feature dataset; in the microscopic sample, the content of epithelial cells per unit volume = NUMS1 / (S1×H). The epithelial cell AI feature dataset is obtained by training after manually annotating epithelial cell images in urine; the urine sample is pre-processed to be concentrated N times, and the content of epithelial cells per unit volume in the urine sample = NUMS1 / (S1×H×N).
[0099] like Figure 15A method for quantitative analysis of tangible elements, comprising: adding reference particles to a sample to obtain a microscopic sample, wherein the concentration Q of the reference particles in the microscopic sample is known; flattening the microscopic sample, photographing the microscopic sample, obtaining two or more microscopic sample images, and selecting an image area Sn in the image n; using an AI recognition algorithm to recognize the microscopic sample image, identifying the target detection object in the selected image area Sn, and obtaining the number of target detection objects NUM_Sn; the number of target detection objects NUM_S_All in all selected images = the sum of the number of each target detection object NUM_Sn; using an AI recognition algorithm to recognize the microscopic sample image, identifying the reference particles in the selected image area Sn, and obtaining the number of reference particles NUM_Pn; the number of particles NUM_P_All in all selected images = the sum of the number of each reference particle NUM_Pn; the volume V2 of the microscopic sample corresponding to the selected image area = NUM_P_All / reference particle concentration Q; the content of the target detection object per unit volume in the microscopic sample = NUM_S_All / V2.
[0100] When testing urine, the volume of raw urine required varies depending on the requirements of each inspection specification. The larger the volume, the more images are required. Assuming one image corresponds to a sample volume of 0.2 × 0.3 × 0.1 = 0.006 ml on the chip; if the minimum sample volume required for analysis is 1 ml, 167 images are required. Errors in chip height will also introduce volume errors, affecting the accuracy of the final quantitative analysis of formed components. The greater the height error, the greater the cumulative error in the quantitative analysis of formed components. By introducing reference particles and converting their concentration into volume, the volume of the sample of equal volume is obtained. This conversion, as long as the reference particle concentration is stable and balanced, can eliminate volume deviations and errors in the quantitative analysis of formed components caused by chip height.
[0101] When the chip height accuracy and volume accuracy are stable, the concentration of the reference particles is converted into volume, which can be verified with the volume calculation based on the image area to ensure the calculation accuracy.
[0102] like Figure 16 When identifying reference particles within the selected image area Sn, the number of reference particles at the boundary of the selected image area Sn is the number of cross-boundary reference particles NUM_P_B. Half of this number is taken and included in NUM_P_All. The aforementioned samples include any one of blood, feces, urine, or other biological samples; and the aforementioned target detection objects include any one of red blood cells, white blood cells, platelets, urine casts, urine crystals, urinary tract epithelial cells, urine microorganisms, urinary tract parasite eggs, intestinal parasite eggs, intestinal protozoa, intestinal microorganisms, intestinal epithelial cells, urine lipid droplets, and intestinal undigested matter.
[0103] like Figure 16 The sizes of the fields of view S21 and S22 of the left and right variables can be different, but this does not affect the accuracy of the calculation of the content of the target detection object per unit volume. As long as the range of the reference particle concentration Q is appropriate and the size of the reference particles is comparable to the size of the target detection object, the accuracy of the calculation of the content of the target detection object per unit volume can be guaranteed, eliminating the process of highly participating in the calculation of the microscopic sample and reducing the influence of the chip processing accuracy on the analysis and measurement accuracy.
[0104] In the quantitative analysis method of tangible components, the concentration Q of the above-mentioned reference particles ranges from 1400 particles / ul to 2000 particles / ul; TF2: the above-mentioned reference particles are floating particles, and the specific gravity relative to water ranges from 0.9 to 0.95; TF3: the above-mentioned reference particles are sinking particles, and the specific gravity relative to water ranges from 1.05 to 1.1; TF4: the diameter of the above-mentioned reference particles ranges from 1um to 10um; TF5: the preferred diameter of the above-mentioned reference particles is 5um.
[0105] A computing and processing device for running all or part of the above-mentioned AI training method for identifying epithelial cells; the memory of the above-mentioned computing and processing device includes the above-mentioned epithelial cell AI feature dataset A; the memory of the above-mentioned computing and processing device includes the above-mentioned epithelial cell AI feature dataset B; used to run all or part of the above-mentioned epithelial cell analysis method; used to run all or part of the above-mentioned quantitative analysis method of formed elements.
[0106] A data storage device storing all or part of the program code for executing the above-mentioned AI training method for identifying epithelial cells; storing the above-mentioned AI feature data set A for epithelial cells; storing the above-mentioned AI feature data set B for epithelial cells; storing all or part of the program code for executing the above-mentioned epithelial cell analysis method; storing all or part of the program code for running the above-mentioned quantitative analysis method for formed elements.
[0107] A detection device for running all or part of the above-mentioned AI training method for identifying epithelial cells; for running all or part of the above-mentioned epithelial cell analysis method; for running all or part of the above-mentioned quantitative analysis method of formed elements.
[0108] Although the present invention is illustrated and described based on the preferred embodiment and several alternatives, the invention is not limited by the specific description in this specification. Other additional replacement or equivalent components can also be used to practice the present invention.
Claims
1. A method for AI training to identify epithelial cells, characterized in that: Urine samples were pre-processed to obtain samples for microscopic examination; Tiling of samples for microscopic examination; photographing the flattened microscopic sample to obtain an image of the microscopic sample; Epithelial cells in the microscopic sample images are identified and labeled to obtain labeled images; the above labeled images include epithelial cell labeling information, and AI training is performed using the labeled images to obtain epithelial cell AI feature dataset A.
2. The AI training method for identifying epithelial cells according to claim 1, characterized in that: Pretreatment of urine samples, including any of the following technical features: TA1: Add dye to the urine sample to stain the urine; TA2: Add dye to the urine sample to dye the urine; the dye is a dry powder dye; TA3: Add reference particles to the urine sample, and the reference particles are located at the bottom layer of the microscopic sample; TA4: When photographing a flat-mounted microscopic specimen, the camera's focal plane is aligned with the center of the epithelial cell; TA5: Add dye to the urine sample to stain the urine; the dye includes a new methylene blue component; TA6: Mix multiple urine samples, let them settle and separate into layers, and then take the bottom layer to obtain a sample for microscopic examination; TA7: Mix multiple urine samples, centrifuge and separate the layers, take the bottom layer, and obtain a sample for microscopic examination; TA8: Obtain a urine sample from an individual with epithelial cell hyperplasia; TA9: Add dye to the urine sample to dye the urine; the dye is a dry powder granular dye; TA10: Add a dye to the urine sample to stain the urine; the dye is a dry powder granular dye; the dry powder granular dye is obtained by quantitatively dividing and drying a liquid dye reagent; the liquid dye reagent includes reference particles, dye, and water, wherein the target concentration range of the reference particles is adjustable.
3. The AI training method for identifying epithelial cells according to claim 1, characterized in that: Epithelial cells are added to a urine sample and mixed to obtain a sample for microscopic examination; Obtaining epithelial cells, including any of the following technical features: TD1: The animal kidney is cut open to expose the renal cortex, and cells from the renal cortex are obtained; TD2: adding water to the animal kidney and crushing it to obtain a suspension, and precipitating and separating the suspension to obtain epithelial cells; TD3: Cut open the animal's urinary epithelial tissue sample and take cells from the cortical part.
4. The AI training method for identifying epithelial cells according to claim 1, characterized in that: Including any one of the following technical features, TB1: The microscopic sample image is sent to the AI training server via the network; TB2: The microscopic sample images are sent to the AI training server via the network, and the segmented images are manually identified and labeled via the network.
5. The AI training method for identifying epithelial cells according to any one of claims 1 to 4, characterized in that: Using epithelial cell AI feature dataset A as the feature dataset, AI software is used to identify one or more pictures input into AI to obtain a recognition output picture, wherein the recognition output picture includes the recognized epithelial cell image, and the recognition output picture is manually reviewed to form a manually reviewed marked picture. The manually reviewed marked picture is used for AI training to obtain an epithelial cell AI feature dataset B; the manually reviewed marked picture includes epithelial cell labeling information and suspected epithelial cell labeling information.
6. The AI training method for identifying epithelial cells according to any one of claims 1 to 5, characterized in that: The epithelial cells are any one or more of renal tubular epithelial cells, transitional epithelial cells, and squamous epithelial cells.
7. The AI training method for identifying epithelial cells according to claim 6, characterized in that: The transitional epithelial cells are any one or more of bottom layer transitional epithelial cells, middle layer transitional epithelial cells, and surface layer transitional epithelial cells.
8. A method for analyzing epithelial cells, characterized in that: Urine samples were pre-processed to obtain samples for microscopic examination; Tiling of samples for microscopic examination; photographing the flattened microscopic sample to obtain an image of the microscopic sample; Identifying the microscopic sample image using an AI recognition algorithm to identify epithelial cells in the image; The AI recognition algorithm includes an epithelial cell AI feature data set, and the epithelial cell AI feature data set includes any one or more feature data of renal tubular epithelial cells, transitional epithelial cells, and squamous epithelial cells.
9. The epithelial cell analysis method according to claim 8, characterized in that: The sample tile height is H; Obtain a microscopic sample image and select an image area S1; Identify the epithelial cells in the urine in the selected area S1 of the image and obtain the number of epithelial cells NUMS1; In the microscopic specimens, the content of epithelial cells per unit volume = NUMS1 / (S1×H).
10. The epithelial cell analysis method according to claim 8, characterized in that: The volume of the microscopic sample is VT; the total test area of the microscopic sample is SA; the total test area is the total area of the epithelial cells in the microscopic sample. Obtain a microscopic sample image and select an image area S1; Identify the epithelial cells in the urine in the selected area S1 of the image and obtain the number of epithelial cells NUMS1; In the microscopic specimen, the content of epithelial cells per unit volume = SA × (NUMS1 / S1) / VT.
11. The epithelial cell analysis method according to claim 9 or 10, characterized in that: Include any one of the following technical features: TG1: The epithelial cell AI feature dataset is obtained by manually annotating epithelial cell images in urine and then training; TG2: The urine sample is pretreated and concentrated N times. The content of epithelial cells per unit volume in the urine sample = the content of epithelial cells per unit volume / N; TG3: urine samples were pre-treated and reference particles were added to assist in focus adjustment; TG4: The urine sample is pre-treated and dye is added to the sample. The dye dyes the urine sample. TG5: The urine sample is pre-treated and a dye is added to the sample, and the dye is used to dye the urine sample; the dye is a dry powder dye: TG6: urine samples were pretreated with dyes including a new methylene blue component; TG7: The epithelial cells are any one or more of renal tubular epithelial cells, transitional epithelial cells, and squamous epithelial cells; TG8: The epithelial cells are any one or more of the following: bottom layer transitional epithelial cells, middle layer transitional epithelial cells, and surface layer transitional epithelial cells.
12. A method for quantitative analysis of tangible elements, characterized in that: Reference particles are added to the sample to obtain a microscopic sample, and the concentration Q of the reference particles in the microscopic sample is known; The microscopic sample is tiled, and the microscopic sample is photographed to obtain two or more microscopic sample images, and an image area Sn is selected from the image n; Identify the microscopic sample image using an AI recognition algorithm, identify the target detection objects in the selected area Sn of the image, and obtain the number of target detection objects NUM_Sn; the number of target detection objects NUM_S_All in all selected images = the sum of the number of each target detection object NUM_Sn; The microscopic sample image is identified using an AI recognition algorithm to identify the reference particles in the selected area Sn of the image to obtain the number of reference particles NUM_Pn; the number of reference particles NUM_P_All in all selected images = the sum of the number of reference particles NUM_Pn of each; the microscopic sample volume V2 corresponding to the selected area of the image = NUM_P_All / reference particle concentration Q; In the microscopic sample, the content of the target detection object per unit volume = NUM_S_All / V2.
13. The method for quantitative analysis of tangible elements according to claim 12, wherein: Include any one of the following technical features: TH1: When the reference particles in the selected image area Sn are identified, the number of reference particles at the boundary of the selected image area Sn is the number of cross-border reference particles NUM_P_B. Half of the number of cross-border reference particles NUM_P_B is counted in NUM_P_All. TH2: The sample includes any one of a blood sample, a stool sample, a urine sample or other biological samples; TH3: The target detection object includes any one of red blood cells, white blood cells, platelets, urine casts, urine crystals, urinary tract epithelial cells, urine microorganisms, urinary tract parasite eggs, intestinal parasite eggs, intestinal protozoa, intestinal microorganisms, intestinal epithelial cells, urine lipid droplets, and intestinal undigested matter.
14. The method for quantitative analysis of tangible elements according to claim 12, wherein: Include any one of the following technical features: TF1: the reference particle concentration Q ranges from 1400 particles / ul to 2000 particles / ul; TF2: The reference particles are floating particles with a specific gravity relative to water ranging from 0.9 to 0.95; TF3: The reference particles are bottom-sinking particles with a specific gravity relative to water ranging from 1.05 to 1.1; TF4: The reference particle diameter ranges from 1 μm to 10 μm; TF5: The reference particles preferably have a diameter of 5 μm.
15. A computing and processing device, characterized in that: Include any one of the following technical features: TEA1: used to run all or part of the AI training method for identifying epithelial cells according to any one of claims 1 to 7; TEA2: The memory of the computing and processing device includes the epithelial cell AI feature dataset A according to claims 1 to 7; TEA3: The memory of the computing and processing device includes the epithelial cell AI feature dataset B according to claim 5; TEA4: used to perform all or part of the epithelial cell analysis method according to any one of claims 8 to 11; TEA5: used to run all or part of the method for quantitative analysis of shaped ingredients according to any one of claims 12 to 14.
16. A data storage device, characterized in that Include any one of the following technical features: TEB1: stores program codes for executing all or part of the AI training method for identifying epithelial cells according to any one of claims 1 to 7; TEB2: stores the epithelial cell AI feature dataset A according to any one of claims 1 to 6; TEB3: stores the epithelial cell AI feature dataset B described in claim 7; TEB4: storing a program code for executing all or part of the epithelial cell analysis method according to any one of claims 6 to 7; TEB5: stores program codes for executing all or part of the method for quantitative analysis of shaped components according to any one of claims 8 to 14.
17. A detection device, characterized in that: Include any one of the following technical features: TEC1: used to run all or part of the AI training method for identifying epithelial cells according to any one of claims 1 to 7; TEC4: used to perform all or part of the epithelial cell analysis method according to any one of claims 8 to 11; TEC5: used to run all or part of the method for quantitative analysis of shaped ingredients according to any one of claims 12 to 14.