Urine detection target object identification AI training method, calculation processing device and storage device
By staining and concentrating the targets in the suspended state in the urine, the problem of low content of targets to be tested in the urine is solved, and effective AI model training and urine detection and recognition are achieved.
Patent Information
- Application Number
- CN202410136764.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2023-12-09
- Filing Date
- 2024-01-31
- Publication Date
- 2025-06-10
AI Technical Summary
The content of targets to be tested in the urine is low and difficult to collect, which makes it difficult to train AI models and cannot effectively identify targets in the urine.
By staining the suspended urine target, it is significantly different from the background. AI training was used for use with the stained pictures to obtain the urine detection target feature data set A, and the target concentration was increased through urine concentration.
It effectively supports AI detection of target objects in the urine that recognizes suspended state, improves the training efficiency and recognition accuracy of the AI model, overcomes the recognition error caused by different stains, and does not reduce urine concentration.
Smart Images
Figure CN120125868A_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the technical field of analysis of formed elements based on microscopic magnified images, and particularly relates to an AI training method for identifying urine test targets, as well as a computing processing and storage device. Background Art
[0002] Urine analysis includes various methods, such as urine protein, urine sugar, and microscopic examination. Routine urine examination with microscopic examination mainly involves examination under a microscope. Routine microscopic examination includes examining the morphology and quantity of cells, casts, and salt crystals in urine sediment under a microscope, the number of red blood cells, white blood cells, pus cell casts, etc. in urine under each high-power microscopic field.
[0003] In routine urine tests, it is difficult to detect by manually identifying the number of red blood cells, white blood cells, pus cell casts, etc. in urine under high-power microscopic fields. Many substances in urine cannot be identified and counted, and the technology is backward.
[0004] The applicant has proposed a series of Chinese patents, such as
[0005] 1. CN2020112669290, "Cell Analysis Method and System and Quantitative Method and System";
[0006] 2. CN2020112669182, "Imaging Method and System for Cell Suspension Samples and Kit";
[0007] 3. CN2022104799126, "Quick Focusing Method for Microscopic Image Acquisition Device and Microscopic Image Acquisition Method";
[0008] 4. CN2023109924201, "Urine Formed Element Analysis Method System and Reagent Preparation Method for Staining Configuration Device"
[0009] Use a brand-new technical solution to measure the content of urine or formed elements in urine.
[0010] Artificial intelligence image recognition technology is an image processing technology based on artificial intelligence algorithms. By analyzing and learning images, it realizes the recognition and understanding of image content.
[0011] Data assets refer to data resources recorded in physical or electronic forms that are owned or controlled by individuals or enterprises and can bring future economic benefits to enterprises. Data assets are datasets in the cyberspace that have data ownership rights (exploration rights, usage rights, ownership rights), are valuable, measurable, and readable.
[0012] The content of red blood cells, white blood cells, pus cells, casts, etc. in urine is relatively low, and the content of some components is very low, such as Figure 6Among them, the contents of many components are zero, but there are a large number of unknown impurities. However, for AI recognition, an AI model needs to be trained with samples. How to collect enough samples is the key to the application of the suspended-state AI recognition technology. Many researchers cannot obtain the target specimens to be detected in urine, so the AI cannot be trained and the project cannot be carried out.
[0013] Many targets to be detected not only have low contents and are difficult to collect, but also, for some targets to be detected in the normal population or species group, the content is zero. A large amount of urine from different groups needs to be collected, photographed, and manually identified and labeled. For many detection items with low contents, it is almost impossible to obtain a sufficient number of training samples. How to increase the concentration of the targets to be detected in urine and how to effectively train the urine recognition AI model have become the key to the industry's development. Summary of the Invention
[0014] In this application, the inventor proposes to stain the urine target substances in a suspended state so that the urine detection target substances are distinguishable from the background. Using the stained pictures for AI training, the obtained AI feature dataset A can effectively support the AI detection of identifying urine detection target substances in suspended urine. By concentrating the urine, the concentration of the urine detection target substances is increased, making it possible to find the urine detection target substances.
[0015] An AI training method for identifying urine detection target substances includes preprocessing urine to obtain a urine sample; spreading the urine sample flat to make the urine detection target substances suspended in the liquid, presenting a natural state; photographing the urine sample to obtain pictures; and using the obtained pictures for AI training to obtain the urine detection target substance feature dataset A.
[0016] For the above-mentioned AI training method for identifying urine detection target substances, the above-mentioned urine preprocessing includes urine sample staining preprocessing, and a staining agent is added to the urine in the urine sample staining preprocessing. For the above-mentioned AI training method for identifying urine detection target substances, the above-mentioned staining agent is a dry powder staining agent or a staining emulsion.
[0017] For the above-mentioned AI training method for identifying urine detection target substances, the above-mentioned dry powder staining agent or staining emulsion is obtained by drying after being prepared by adding water to the staining agent.
[0018] For the above-mentioned AI training method for identifying urine detection target substances, the above-mentioned urine preprocessing includes urine sample enrichment preprocessing.
[0019] For the above-mentioned AI training method for identifying urine detection target substances, the above-mentioned urine sample enrichment preprocessing is to let the urine sample stand still for layering or perform centrifugal layering, and take the sample of the layer where the urine detection target substances are located to obtain an enriched sample.
[0020] The above urine test target recognition AI training method, where the urine test target is a lipid droplet, the lipid droplet floats on the upper layer of the urine sample, and the upper layer sample is taken to obtain an enriched sample; the urine sample is photographed to photograph the upper layer of the sample to obtain a picture.
[0021] The above urine test target recognition AI training method, where the urine sample enrichment pretreatment is before the urine sample staining pretreatment; or the urine sample enrichment pretreatment is after the urine sample staining pretreatment.
[0022] The above urine test target recognition AI training method, where the AI training includes image preprocessing and AI training processing. The image preprocessing outputs the labeled urine test target, and based on the labeled urine test target, the AI training processing outputs the urine test target feature dataset.
[0023] The above urine test target recognition AI training method, using the urine test target feature dataset A as the feature dataset, using an AI software to recognize one or more pictures of the above picture to obtain a marked picture. The marked picture includes the labeled urine test target, performing manual review on the marked picture to form a manually reviewed marked picture, and using the manually reviewed marked picture for AI training to obtain the urine test target feature dataset B.
[0024] The above urine test target recognition AI training method, where the above urine test target includes one or more of casts, crystals, cells, pathogenic microorganisms, and large particle unknown substances.
[0025] A computing device, the above computing device includes the above urine test target recognition AI training method.
[0026] A computing device, the memory of the above computing device includes the above urine test target AI feature dataset A or urine test target AI feature dataset B.
[0027] A data storage device, the above data storage includes the above urine test target AI feature dataset A or urine test target AI feature dataset B.
[0028] The technical effects of the above technical solution include: By performing AI training on the suspended urine, the obtained AI feature dataset A can effectively support the recognition of suspended urine AI detection.
[0029] The technical effects of the above technical solution include: By staining the urine, cell-like target substances are stained, making it easier to highlight them from the sample.
[0030] The technical effects of the above technical solution include: when taking training pictures, using the same staining agent as in the test, overcoming the different recognition error rates of AIs trained with different staining agents, and even being unable to recognize and use them.
[0031] The technical effects of the above technical solution include: with the dry powder staining agent, the concentration of urine is not reduced.
[0032] The technical effects of the above technical solution include: through enrichment pretreatment, the probability of obtaining samples is increased.
[0033] The technical effects of the above technical solution include: taking the upper layer sample to obtain an enriched sample and taking an image of the upper layer target object.
[0034] The technical effects of the above technical solution include: by identifying unknown substances, by detecting more photos of unknown substances, and using different personnel to identify and classify unknown substances, the identification efficiency is improved.
[0035] The technical effects of the above technical solution include: through multiple iterations of identification, the training efficiency of the AI model is accelerated. Description of the Drawings
[0036] Figure 1 It is a schematic diagram of the steps of an AI training method for identifying urine test target objects Figure 1 ;
[0037] Figure 2 It is a schematic diagram of the steps of an AI training method for identifying urine test target objects Figure 2 ;
[0038] Figure 3 It is a schematic diagram of the steps of an AI training method for identifying urine test target objects Figure 3 ;
[0039] Figure 4 It is a schematic diagram of the steps of an AI training method for identifying urine test target objects Figure 4 ;
[0040] Figure 5 It is a schematic diagram of the suspended state of the substance to be detected in urine;
[0041] Figure 6 It is a picture of the bottom layer of the suspended urine taken;
[0042] Figure 7 It is a picture of various crystals found in a large number of pictures;
[0043] Figure 8 It is a picture of various casts found in a large number of pictures;
[0044] Figure 9 It is a picture of various cells found in a large number of pictures;
[0045] Figure 10 are pictures of various lipid droplets found among a large number of pictures;
[0046] Figure 11 are pictures of various unknown physical objects found among a large number of pictures;
[0047] Figure 12 are pictures of bacilli found among a large number of pictures;
[0048] Figure 13 are pictures of cocci found among a large number of pictures;
[0049] Figure 14 are items and reference ranges for urine tests. Specific implementation manners
[0050] The following further details the content of the present application in conjunction with each attached drawing. It should be noted that the following is an illustration of the preferred embodiments of the present invention and does not constitute any limitation to the present invention. The illustration of the preferred embodiments of the present invention is only for the illustration of the general principles of the present invention. The numbers such as "first", "second", "A", and "B" involved in the present invention are only for the convenience of illustration and do not represent the sequence relationship in time or space. The combinations of letters and numbers such as "TA", "TB", and "H" involved in the present invention are only for the convenience of illustration, and the specific meanings are determined by the specific words they represent.
[0051] Such as Figure 1 , a method for training an AI for identifying urine test targets, preprocessing urine to obtain a urine sample; spreading the urine sample flat to make the urine test targets suspended in the liquid and present a natural state; photographing the urine sample to obtain pictures; using the obtained pictures for AI training to obtain a urine test target feature dataset A.
[0052] During the training process, the urine needs to be kept in the same state as the detection to take pictures in the detection state and obtain AI training pictures. Only in this way can the trained model have a high recognition rate. The pictures taken by traditional urine tests, such as the pictures taken in the cover glass mode, have compression deformation on the object to be detected and cannot be used as AI training pictures.
[0053] Such as Figure 2 , the above-mentioned method for training an AI for identifying urine test targets, the above-mentioned urine preprocessing includes urine sample staining preprocessing, and a staining agent is added to the urine in the urine sample staining preprocessing. In the above-mentioned method for training an AI for identifying urine test targets, the staining agent is a dry powder staining agent or a staining emulsion.
[0054] The background color of urine is transparent. Many substances to be detected, especially bacteria, cells, etc., are themselves transparent and in a suspended state. The pictures taken are almost merged with the background and cannot be effectively trained. If a staining agent is used during the detection process, the same staining agent needs to be used when taking training pictures. The AIs trained with different staining agents have different recognition error rates and may even be unable to recognize and use.
[0055] For the above-mentioned AI training method for identifying urine detection target substances, the above-mentioned dry powder staining agent or staining emulsion is obtained by drying after adding water to the staining agent.
[0056] In the urine detection scenario, in order not to reduce the concentration of urine, the same staining agent can also be used in the training scenario through the dry powder staining agent.
[0057] Such as Figure 2 、 Figure 3 For the above-mentioned AI training method for identifying urine detection target substances, the above-mentioned urine pretreatment includes urine sample enrichment pretreatment.
[0058] In the detection environment, in order to calculate the concentration of the detected substance, there is a limit on the enrichment degree. In the AI training scenario, a high enrichment degree method can be adopted, such as enriching urine by 100 times or 1000 times, to enrich the target substances with low content and improve the efficiency of AI training.
[0059] Figure 5 For the above-mentioned AI training method for identifying urine detection target substances, the above-mentioned urine sample enrichment pretreatment is to let the urine sample stand and stratify or centrifuge and stratify, and take the sample of the layer where the urine detection target substance is located to obtain the enriched sample.
[0060] Figure 5 For the above-mentioned AI training method for identifying urine detection target substances, the above-mentioned urine detection target substance is a lipid droplet, and the lipid droplet floats on the upper layer of the above-mentioned urine sample. Take the upper layer sample to obtain the enriched sample; the above-mentioned urine sample shooting is to shoot the upper layer of the sample to obtain a picture.
[0061] For the above-mentioned AI training method for identifying urine detection target substances, the above-mentioned urine sample enrichment pretreatment is before the above-mentioned urine sample staining pretreatment; or the above-mentioned urine sample enrichment pretreatment is after the above-mentioned urine sample staining pretreatment.
[0062] Figure 2 , Figure 3 For the above-mentioned AI training method for identifying urine detection target substances, the above-mentioned AI training includes image preprocessing and AI training processing. The image preprocessing outputs the labeled urine detection target substances, and the AI training processing outputs the urine detection target substance feature data set according to the labeled urine detection target substances.
[0063] Figure 4, for the above urine test target recognition AI training method, urine test target feature dataset A is used as the feature dataset, and an AI software is used to recognize one or more of the above pictures to obtain marked pictures. The marked pictures include the marked urine test targets. The marked pictures are manually reviewed to form manually reviewed marked pictures, and the AI training is performed using the manually reviewed marked pictures to obtain urine test target feature dataset B.
[0064] Figures 6 to 14 , for the above urine test target recognition AI training method, the above urine test targets include one or more of casts, crystals, cells, pathogenic microorganisms, and large particle unknown substances.
[0065] Such as Figure 11 , due to the limited knowledge of individual annotators, for suspected targets that cannot be recognized, they can be labeled as different suspected substances, such as suspected casts, suspected cells, and suspected crystals. During the training process, the AI will enrich these suspected substances, and other experienced annotators can modify the labels to speed up the labeling of the entire urine system.
[0066] Such as Figure 6 , in a picture, it is difficult for ordinary individuals to recognize all the objects to be tested in the picture, and collaborative identification and labeling are required. After the AI system obtains feature dataset A through the first round of recognition, subsequent recognition is based on feature dataset A, continuously obtaining feature dataset B, and performing multiple rounds of iteration, which can label and recognize various objects to be tested. Using this method, the AI recognition training and labeling become a systematic and long-term task, gradually establishing the AI recognition model and AI feature dataset of the urine system. Only in this way is it possible for the AI recognition of suspended urine to have industrial application and development.
[0067] Such as Figure 7 , various crystals that are recognized, these clearly labeled crystals, can be used to train the model for iterating the AI model and improving feature dataset B.
[0068] Such as Figure 8 , the casts recognized using feature dataset A can be manually judged to exclude suspected casts, improve the recognition rate of the model, and form a more effective feature dataset B.
[0069] Such as Figure 9 , the cells recognized using feature dataset A can be manually judged to exclude suspected cells, improve the recognition rate of the model, and form a more effective feature dataset B. At the same time, further classification can be carried out. For example, sperm cells are a sub-classification of cells formed by expert recognition and labeling among the existing recognized cells, and subsequent sperm recognition becomes a standard classification in feature dataset B.
[0070] Such as Figure 10, the lipid droplets identified by the feature dataset A are manually interpreted to improve the recognition rate of the model and form a more effective feature dataset B. The lipid droplets float on the upper layer in the urine sample, and it is necessary to take images of the upper liquid surface for recognition.
[0071] Such as Figure 11 , the suspected casts identified by the feature dataset A are marked with unclear visible substances during the recognition process, which can quickly gather the data information of the unclear visible substances. After being interpreted by medical experts, new detection parameters can be created.
[0072] Such as Figure 12 , the bacilli identified by the feature dataset A.
[0073] Such as Figure 13 , the cocci identified by the feature dataset A are manually interpreted to improve the recognition rate of the model and form a more effective feature dataset B.
[0074] Such as Figure 14 , the applicant cooperates with pet hospitals to test and verify a large number of pet urine samples and train the AI model, gradually verifying the scientific nature of the AI training method and successfully finding sample photos of various parameters.
[0075] A computing device, the above computing device includes the above AI training method for urine detection target recognition.
[0076] A computing device, the memory of the above computing device includes the above urine detection target AI feature dataset A or urine detection target AI feature dataset B.
[0077] A data storage device, the above data storage includes the above urine detection target AI feature dataset A or urine detection target AI feature dataset B.
[0078] Although the present invention is described and illustrated according to preferred embodiments and several alternative solutions, the invention is not limited by the specific description in this specification. Other alternative or equivalent components can also be used to practice the present invention.
Claims
1. A urine detection target object recognition AI training method, characterized in that: The technical features include: Urine pretreatment to obtain urine samples; Spread the urine sample flat so that the urine test target is suspended in the liquid and presents a natural state; Take a picture of the urine sample; The obtained images are used for AI training to obtain the urine detection target feature data set A.
2. The urine detection target recognition AI training method according to claim 1, characterized in that: The technical features include: The urine pretreatment includes urine sample dyeing pretreatment, and the urine sample dyeing pretreatment includes adding a dye into the urine.
3. The urine detection target recognition AI training method according to claim 1, characterized in that: The invention comprises the following technical features: the dye is a dry powder dye, a dye emulsion or a liquid dye.
4. The urine detection target recognition AI training method according to claim 3, characterized in that: The invention comprises the following technical features: the dry powder dye or dye emulsion is obtained by adding water to the dye and then drying it.
5. The urine detection target recognition AI training method according to any one of claims 1 to 4, characterized in that: The method includes the following technical features: the urine pretreatment includes urine sample enrichment pretreatment.
6. The urine detection target recognition AI training method according to claim 5, characterized in that: The method includes the following technical features: the urine sample enrichment pretreatment is to stratify the urine sample statically or by centrifugation, and take the stratified sample where the urine detection target is located to obtain the enriched sample.
7. The urine detection target recognition AI training method according to claim 5, characterized in that: The target of urine detection is fat droplets, which float on the upper layer of the urine sample. The upper layer sample is taken to obtain an enriched sample; the urine sample is photographed by photographing the upper layer of the sample to obtain a picture.
8. The urine detection target recognition AI training method according to claim 5, characterized in that: The invention comprises the following technical features: the urine sample enrichment pretreatment is before the urine sample staining pretreatment; or the urine sample enrichment pretreatment is after the urine sample staining pretreatment.
9. The urine detection target recognition AI training method according to claim 1, characterized in that: The AI training includes image preprocessing and AI training processing. The image preprocessing outputs the marked urine detection target object. The AI training processing outputs the urine detection target object feature data set according to the marked urine detection target object.
10. The urine detection target recognition AI training method according to claim 1 or 5, characterized in that: A urine detection target feature data set A is used as a feature data set, and one or more images of the image are identified using AI software to obtain a marked image, wherein the marked image includes the marked urine detection target, and the marked image is manually reviewed to form a manually reviewed marked image. The manually reviewed marked image is used for AI training to obtain a urine detection target feature data set B.
11. The urine detection target recognition AI training method according to claim 1 or 5, characterized in that: The urine detection targets include one or more of casts, crystals, cells, pathogenic microorganisms, and large particles of unknown substances.
12. A computing device, characterized in that: The computing and processing device includes the AI training method for urine detection target object recognition as described in any one of claims 1 to 10.
13. A computing device, characterized in that: The memory of the computing and processing device includes the urine detection target AI feature data set A or the urine detection target AI feature data set B as described in any one of claims 1 to 9.
14. A data storage device, characterized in that: The data storage includes the urine detection target AI feature data set A or the urine detection target AI feature data set B as described in any one of claims 1 to 9.