Method and device for AI training and analysis of hydrothorax and ascites detection

Abstract

According to the hydrothorax and ascites detection AI training and analysis method and device, a hydrothorax and ascites sample is preprocessed to obtain a microscopic examination sample, and the microscopic examination sample is tiled; shooting to obtain a microscopic examination sample image; identifying and labeling tangible objects in the microscopic examination sample image to obtain a labeled picture; and AI training is carried out by using the labeled picture to obtain a hydrothorax and ascites detection AI feature data set A. An AI recognition algorithm of the urine AI feature data set is adopted, the chest and ascites tangible object image A is output, manual recheck recognition and labeling are carried out, and a labeling picture B is obtained; carrying out AI training by using the labeled picture B to obtain a hydrothorax and ascites detection AI feature data set B; and by adopting an AI recognition algorithm of the hydrothorax and ascites detection AI feature data set B or A, outputting a hydrothorax and ascites tangible object image C, carrying out manual recheck recognition and labeling, obtaining a labeling picture D, and carrying out AI training to obtain a hydrothorax and ascites detection AI feature data set C. Through AI training, tangible objects in the hydrothorax and ascites turbid liquid can be accurately classified and identified.

Description

Technical Field

[0001] This application belongs to the field of tangible object analysis technology based on magnified microscopic images, and specifically relates to an AI training method for tangible object recognition in pleural and peritoneal fluid, a method for detecting and analyzing pleural and peritoneal fluid, and a device for pleural and peritoneal fluid detection, analysis, computation, processing, and data storage. Background Technology

[0002] Pleural and peritoneal effusions refer to fluid accumulation in the pleural and peritoneal cavities. Pleural and peritoneal effusions may be caused by tumors or inflammation, such as pleurisy, tuberculous pleurisy, and pulmonary tuberculosis. A routine pleural effusion examination mainly involves extracting pleural fluid from the patient's pleural cavity for examination. Through appearance, cell count, and classification, the nature of the pleural effusion can be determined, which is of great significance for diagnosing the cause, such as determining whether there is an infection or malignant tumor.

[0003] Transudate pleural and peritoneal effusions are usually clear and transparent, colorless or pale yellow, and do not coagulate upon standing. Exudate effusions, due to the presence of red blood cells, white blood cells, bacteria, or chyle, are turbid and vary in color, appearing deep yellow in cases of jaundice. If the pleural and peritoneal effusions contain blood, they may appear red, dark red, or brownish-red. Generally, purulent bacterial infections present as yellowish-purulent, opaque, and viscous pleural and peritoneal effusions with a foul odor, often indicating anaerobic bacterial infection. Cholesterol pleural and peritoneal effusions are yellowish-white and contain a large amount of refractive cholesterol crystals; chylous effusions are milky white and are seen in filariasis, lymphomas, or tuberculosis. Examination of pleural and peritoneal effusions also includes specific markers such as tumor markers.

[0004] In existing technologies, the detection of pleural and peritoneal effusions is typically based on the body fluid detection mode of conventional hematology analyzers. Cell classification in pleural and peritoneal effusions is developed based on the principles of impedance cytometry, flow cytometry, and nucleic acid fluorescence staining. However, the detection tubing of flow cytometers is complex, making it difficult to utilize more advanced AI computing power to improve their productivity.

[0005] The types of formed elements in pleural and peritoneal effusions are more numerous and diverse than those in blood. Conventional methods such as flow cytometry are inaccurate for identifying and classifying these elements, as similarly sized elements can easily be confused. Often, supplementary pleural and peritoneal effusion smears are necessary. However, these smears only allow for localized observation and classification, providing qualitative analysis but not accurate quantitative analysis. For example, when nucleated cells, particularly strongly fluorescent cells (HFCs), are present in the pleural and peritoneal effusions that are undetectable by instruments, a high proportion may indicate the presence of tumor cells, necessitating further microscopic examination of the smears.

[0006] The applicant has filed a series of Chinese patents, such as

[0007] 1. CN2020112669290, "Cellular analysis methods and systems and quantitative methods and systems";

[0008] 2. CN2020112669182, "Imaging methods, systems and kits for cell suspension samples";

[0009] 3. CN2022104799126, "A method for rapid focusing and a method for detecting microscopic images in a microscopic image acquisition device";

[0010] 4. CN2023100423151, "Blood Imaging Analysis System and Method";

[0011] A brand-new AI recognition technology solution is used to measure the content of solid matter in blood, urine, and pleural and peritoneal fluid samples.

[0012] However, how to obtain diverse pleural and peritoneal fluid samples, train corresponding AI models, and continuously iterate and improve the accuracy and efficiency of AI recognition based on these are technical problems to be solved.

[0013] Bacteria: Based on their morphology, they are mainly classified into cocci, bacilli, and spirilla. Cocci: Common coccal morphologies include scattered single cocci, diplococci, streptococci, tetrad cocci, and octadic cocci. Bacilli: Different bacilli vary greatly in size, length, and thickness; they can appear as rod-shaped, thread-like, paired, or short chain arrangements.

[0014] Clinical significance of bacteria: Bacterial pleural effusion is commonly seen in pleural cavity infections caused by foreign body migration, penetrating chest trauma, hematogenous or lymphatic transmission, esophageal perforation, parasitic migration, progression of pyramiditis, transmission of parapneumonia, tumors, rupture of lung abscesses, and iatrogenic causes (thoracic surgery or thoracentesis). Bacterial ascites is commonly seen in bacterial peritonitis caused by damage to the gastrointestinal structure, hepatobiliary infections, and liver abscesses.

[0015] Serous cavity effusion: The serous membrane of the serous cavity comes from the germinal leaf and is mainly composed of mesothelial connective tissue. It is divided into the visceral layer and the parietal layer. The narrow space between the two layers is the serous cavity, including the pleural cavity, peritoneal cavity, pericardial cavity, and testicular tunica vaginalis cavity. Normal serous cavities contain a small amount of fluid, which plays a lubricating role. Under pathological conditions, a large amount of fluid can be produced, which is serous cavity effusion.

[0016] In clinical practice, cytological examination of pleural effusion and ascites requires slow centrifugation of samples followed by staining and microscopic examination of the precipitate using either the Wright's or Diff-Quick staining method. This process is cumbersome and time-consuming. During bacterial staining and microscopic examination, the background of stained protein particles significantly affects bacterial interpretation, potentially leading to misdiagnosis or missed detection. This procedure demands a high level of technical skill from laboratory personnel. In clinical practice, bacterial detection in the staining and microscopic examination of serous cavity effusions is generally based on qualitative or semi-quantitative results (e.g., negative or positive; 1+, 2+, 3+, 4+) to assess the degree of bacterial infection, making accurate bacterial counting difficult. Summary of the Invention

[0017] In this application, the inventors propose using AI training on microscopic images of samples containing tangible objects to accurately classify and identify tangible objects in pleural and peritoneal fluid suspensions. The AI ​​training method in this application involves manually reviewing and labeling images, including labels for tangible objects and suspected tangible objects. This not only improves training efficiency but also provides researchers with research samples. In particular, the use of a network to aggregate training sample collection from various locations to a server allows experts to bypass on-site sample collection and perform labeling or research analysis remotely.

[0018] In this application, the inventors propose a method for detecting and analyzing pleural and peritoneal fluid, which can accurately obtain the content of tangible matter per unit volume in a microscopic sample = NUMS1 / V1, and perform accurate quantitative analysis.

[0019] The technical solution proposed in this application to solve the above-mentioned technical problems is an AI training method for pleural and peritoneal effusion detection. The pleural and peritoneal effusion samples are preprocessed to obtain microscopic samples; the microscopic samples are laid flat; the flat microscopic samples are photographed to obtain microscopic sample images; tangible objects in the microscopic sample images are identified and labeled to obtain labeled images; the labeled images are used for AI training to obtain pleural and peritoneal effusion detection AI feature dataset A.

[0020] The technical solution proposed in this application to solve the above-mentioned technical problems is an AI training method for detecting pleural and peritoneal effusion. The pleural and peritoneal effusion sample is preprocessed to obtain a microscopic sample; the microscopic sample is laid flat; the flat microscopic sample is photographed to obtain a microscopic sample image; an AI recognition algorithm is used to identify and classify the microscopic sample image, outputting a tangible image A of the pleural and peritoneal effusion. The aforementioned AI recognition algorithm uses a urine AI feature dataset; the urine AI feature dataset is obtained through urine tangible object annotation training; the aforementioned pleural and peritoneal effusion tangible object image A is manually reviewed, identified, and annotated to obtain an annotated image B; annotated image B includes tangible objects in the pleural and peritoneal effusion and tangible objects suspected to be present in the pleural and peritoneal effusion; the annotated image B is used for AI training to obtain a pleural and peritoneal effusion detection AI feature dataset B.

[0021] The technical solution proposed in this application to solve the above-mentioned technical problems is an AI training method for detecting pleural and peritoneal effusion. The pleural and peritoneal effusion sample is preprocessed to obtain a microscopic examination sample; the microscopic examination sample is laid flat; the flattened microscopic examination sample is photographed to obtain a microscopic examination sample image; an AI recognition algorithm is used to identify and classify the microscopic examination sample image, outputting an image C of tangible objects in the pleural and peritoneal effusion; the aforementioned AI recognition algorithm uses either the aforementioned AI feature dataset B or the aforementioned AI feature dataset A for detecting pleural and peritoneal effusion; the aforementioned image C of tangible objects in the pleural and peritoneal effusion is manually reviewed, identified, and labeled to obtain labeled images D; labeled images D include tangible objects in the pleural and peritoneal effusion and suspected tangible objects in the pleural and peritoneal effusion; the labeled images D are used for AI training to obtain the AI ​​feature dataset C for detecting pleural and peritoneal effusion.

[0022] Artificially cultured bacteria were added to the above-mentioned pleural and peritoneal fluid samples.

[0023] The above-mentioned pretreatment of pleural or peritoneal fluid samples includes: Step 10: Inoculating the pleural or peritoneal fluid containing bacteria onto a culture medium for bacterial culture to obtain bacterial colonies; Step 20: Extracting an appropriate amount of bacteria from the bacterial culture medium and adding it to the pleural or peritoneal fluid sample to obtain an intermediate sample containing a large number of bacteria; Step 30: Taking the intermediate sample and adding it to the staining reagent and mixing thoroughly to obtain a microscopic examination sample.

[0024] The above-mentioned AI training method for detecting pleural and peritoneal effusions includes any one of the following technical features: TN1: The above-mentioned pleural or peritoneal effusions containing bacteria are from one or more experimental individuals; TN2: The above-mentioned bacterial culture includes aerobic culture, carbon dioxide culture, microaerophilic culture and / or anaerobic culture; TN3: The above-mentioned bacterial culture medium includes: agar medium, nutrient broth medium and / or peptone water medium.

[0025] The AI ​​training method for detecting pleural and peritoneal fluid described above includes the following technical features in the preprocessing of pleural and peritoneal fluid samples: TA1: Adding a staining agent to the pleural and peritoneal fluid samples to stain them; TA2: Adding a staining agent to the pleural and peritoneal fluid samples to stain them; the staining agent is a dry powder staining agent; TA3: Adding reference particles to the pleural and peritoneal fluid samples, with the reference particles located at the bottom layer of the microscopic sample; TA4: When photographing a flat microscopic sample, the camera's focal plane is aligned with the center of the tangible object; TA5: Adding a staining agent to the pleural and peritoneal fluid samples to stain them; the staining agent includes a neomethylene blue component; TA6: Mixing multiple pleural and peritoneal fluid samples, allowing them to settle and separate into layers, and then taking the bottom layer to obtain the microscopic sample; TA7: Mixing multiple pleural and peritoneal fluid samples, centrifuging to separate into layers, and then taking the bottom layer to obtain the microscopic sample; TA8: Taking pleural and peritoneal fluid samples from organisms with increased tangible matter. TA9: Add staining agent to pleural and peritoneal fluid samples to stain the fluid; the staining agent is a dry powder granule staining agent; TA10: Add staining agent to pleural and peritoneal fluid samples to stain the fluid; the staining agent is a dry powder granule staining agent; the dry powder granule staining agent is obtained by quantitatively aliquoting and drying a liquid staining reagent; the liquid staining reagent includes reference particles, staining agent and water, wherein the target concentration range of the reference particles can be adjusted; TA11: The above-mentioned AI training method for pleural and peritoneal fluid detection can also be applied to AI training for serous cavity effusion detection, including AI training for detecting fluid in the pleural cavity, peritoneal cavity, pericardial cavity and testicular tunica vaginalis cavity; TA12: The above-mentioned microscopic sample images are sent to the AI ​​training server via the network; TA13: 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 via the network.

[0026] The technical solution of this application to solve the above-mentioned technical problems is a method for detecting and analyzing pleural and peritoneal fluid. The pleural and peritoneal fluid sample is preprocessed to obtain a microscopic sample; the microscopic sample is laid flat to obtain a microscopic sample image, and the image area S1 is selected; the volume corresponding to the image area S1 is V1; the above microscopic sample image is identified using an AI recognition algorithm to identify tangible matter in the pleural and peritoneal fluid in the selected area S1 of the image, and the number of tangible matter NUMS1 is obtained; the above AI recognition algorithm includes a pleural and peritoneal fluid detection AI feature dataset; in the microscopic sample, the content of tangible matter per unit volume = NUMS1 / V1.

[0027] The above-mentioned method for detecting and analyzing pleural and peritoneal fluid includes any one of the following technical features: TG1: The above-mentioned AI feature dataset for detecting pleural and peritoneal fluid includes AI feature dataset A, AI feature dataset B, or AI feature dataset C for detecting pleural and peritoneal fluid; TG2: The above-mentioned pleural and peritoneal fluid samples are pretreated and concentrated N times, and the content of tangible matter per unit volume in the pleural and peritoneal fluid sample = NUMS1 / (V1×N); TG3: Reference particles are added to the pleural and peritoneal fluid samples after pretreatment, and the reference particles are used to assist in focusing adjustment; TG4: A staining agent is added to the pleural and peritoneal fluid samples after pretreatment, and the staining agent stains the pleural and peritoneal fluid samples; TG5: A staining agent is added to the pleural and peritoneal fluid samples after pretreatment, and the staining agent stains the pleural and peritoneal fluid samples; The staining agent is a dry powder staining agent: TG6: Pleural and peritoneal fluid samples are pretreated and then stained with staining agent, which includes neomethylene blue; TG7: The above-mentioned formed matter includes cellular formed matter, which includes any one or more of the following: neutrophils, degenerative neutrophils, eosinophils, basophils, lymphocytes, macrophages, phagocytic cells, erythrocytes, shadow erythrocytes, mesothelial cells, mast cells, and tumor cells; TG8: The above-mentioned formed matter includes microbial formed matter, which includes any one or more of the following: bacilli, cocci, histoplasma capsulatum, and yeast; TG9: The above-mentioned formed matter includes parasitic formed matter, which includes any one or more of the following: microfilariae and Toxoplasma gondii.

[0028] A computing processing device includes any one of the following technical features: TEA1: used for all or part of the above-described AI training method for pleural and peritoneal effusion detection; TEA2: the memory of the computing processing device includes the above-described AI feature dataset A, AI feature dataset B, or AI feature dataset C for pleural and peritoneal effusion detection; TEA3: used for running all or part of the above-described pleural and peritoneal effusion detection and analysis method.

[0029] A data storage device includes any one of the following technical features: TEB1: storing all or part of the algorithm code for performing the above-described AI training method for pleural and peritoneal effusion detection; TEB2: storing the above-described AI feature dataset A, AI feature dataset B, or AI feature dataset C for pleural and peritoneal effusion detection; TEB3: storing all or part of the algorithm code for performing the above-described AI analysis method for pleural and peritoneal effusion detection.

[0030] A detection device includes any one of the following technical features: TEC1: used to run all or part of the above-described AI training method for pleural and peritoneal effusion detection; TEC4: used to run all or part of the above-described pleural and peritoneal effusion detection and analysis method.

[0031] The technical effects of the above-mentioned technical solution include: obtaining AI feature dataset A for pleural and peritoneal effusion detection by training AI through labeled images, which provides a basic feature data set for subsequent AI training.

[0032] The technical effects of the above-mentioned technical solution include: the AI ​​recognition algorithm uses a urine AI feature dataset for preliminary recognition, and in the absence of a pleural effusion detection AI feature dataset, it can share the accumulation of AI features for similar formed elements in urine as much as possible, thereby improving the efficiency of AI training.

[0033] The technical effects of the above-mentioned technical solution include: manually reviewing and labeling the tangible image A of pleural and peritoneal fluid based on the urine AI feature dataset to obtain labeled image B; using labeled image B for AI training to obtain pleural and peritoneal fluid detection AI feature dataset B; and efficiently completing the iteration of the feature dataset.

[0034] The technical effects of the above-mentioned technical solution include: manually reviewing and labeling the tangible images of pleural and peritoneal fluid (PPF) based on the AI ​​feature dataset B or AI feature dataset A for PPF detection, obtaining labeled images D, and using labeled images D for AI training to obtain the AI ​​feature dataset C for PPF detection, which can more efficiently complete multiple rounds of feature dataset iteration.

[0035] The technical effects of the above-mentioned technical solution include: the trained pleural and peritoneal effusion detection AI feature dataset A, the evolved pleural and peritoneal effusion detection AI feature datasets B and C can all be packaged into independent data that can be sold separately, reducing the engineering difficulty of the entire industry, enabling new technologies to be promoted and implemented as soon as possible, and at the same time, the AI ​​recognition capability can be continuously iterated and improved.

[0036] The technical effects of the above-mentioned technical solution include: adding artificially cultured bacteria to pleural and peritoneal fluid samples to obtain microscopic samples; and further improving the efficiency of acquiring feature data.

[0037] The technical effects of the above-mentioned solution include: inoculating pleural or ascites fluid containing bacteria onto a culture medium for bacterial culture to obtain bacterial colonies; extracting an appropriate amount of bacteria from the bacterial culture medium and adding it to the pleural or ascites fluid sample to obtain an intermediate sample containing a large number of bacteria; and taking the intermediate sample and mixing it thoroughly with staining reagent to obtain a microscopic examination sample. The bacteria exhibit good homology, and the obtained characteristic data more closely matches the original situation.

[0038] The technical effects of the above-mentioned technical solution include: pleural effusion or ascites containing bacteria comes from one or more experimental individuals, which can improve the efficiency of acquiring characteristic data.

[0039] The technical effects of the above-mentioned technical solutions include: the variety of bacterial culture methods can further improve the efficiency of acquiring feature data.

[0040] The technical effects of the above-mentioned solution include: the purpose of staining pleural and peritoneal fluid is to detect leukocytes and other formed matter in the fluid. Using the same staining method when training the AI ​​model can improve the accuracy of identification. The staining agent can be either a liquid staining agent or a dry powder staining agent.

[0041] The technical effects of the above-mentioned technical solution include: directly adding dry powder dye avoids the problem of increased pleural and peritoneal fluid volume caused by adding liquid dye.

[0042] The technical advantages of the above-mentioned technical solution include: the dry powder granule staining agent can be added to the reference granules and the staining agent at the same time, resulting in higher sample pretreatment efficiency.

[0043] The technical effects of the above-mentioned technical solution include: the addition of reference particles makes the microscopic imaging efficiency of pleural and peritoneal fluid higher, and solves the problem that it is difficult to quickly complete the focusing due to the relatively low content of tangible matter in pleural and peritoneal fluid.

[0044] The technical advantages of the above solution include: the dry powder granule staining agent is obtained by quantitatively aliquoting and drying the liquid staining reagent; the liquid staining reagent can be weighed on a large scale, achieving more accurate and uniform quantification of reference particles and staining agents. A single dose of dry powder granule staining agent used for extremely small samples is typically in the microliter range, corresponding to a very small amount of dry powder granule staining agent, making accurate weighing difficult and compromising the accuracy of reference particle and staining agent concentrations. Under the condition that accurate weighing of single doses of dry powder granule staining agent is difficult, the above method can obtain precise single doses of dry powder granule staining agent, achieving more accurate and uniform quantification of reference particles and staining agents, ensuring the accuracy of subsequent quantitative analysis benchmarks.

[0045] The technical effects of the above-mentioned technical solution include: the camera's focal plane is aligned with the center of the object, which can obtain the clearest image of the object, which is beneficial for subsequent recognition training.

[0046] The technical benefits of the above solution include: mixing multiple pleural and peritoneal fluid samples to obtain microscopic samples, thus improving the efficiency of obtaining tangible samples. Not all samples contain tangible matter; the above method can improve the efficiency of obtaining feature data.

[0047] The technical effects of the above-mentioned technical solution include: after mixing multiple pleural and peritoneal fluid samples and allowing them to settle and separate into layers or performing routine separation, the bottom layer is taken to obtain a microscopic sample for use as a microscopic sample, which further improves the efficiency of obtaining tangible feature data.

[0048] The technical effects of the above-mentioned technical solution include: taking pleural and peritoneal fluid samples from biological individuals with increased tangible matter, further improving the efficiency of acquiring characteristic data.

[0049] The technical effects of the above-mentioned technical solution include: sending microscopic sample images to the AI ​​training server via the network, manually identifying and labeling the segmented images via the network, and manually reviewing and labeling the identified output images via the network. This allows for the aggregation and utilization of more resources, improves the efficiency of feature data acquisition, and enhances the accuracy of AI training and analysis.

[0050] The technical effects of the above solution include: The evolved AI feature dataset B for pleural and peritoneal effusion detection, based on dataset A, is encapsulated as independent data, enabling iterative improvement of AI recognition capabilities. Information on suspected tangible objects provides research material for clinical practice.

[0051] The technical effects of the above-mentioned technical solution include: the AI ​​feature dataset for detecting pleural and peritoneal effusions is obtained by training after manually annotating images of tangible objects in the pleural and peritoneal effusions, which ensures greater accuracy in AI recognition.

[0052] The technical effects of the above technical solution include: defining a new parameter, the number of solids in a unit volume of pleural and peritoneal fluid = the number of solids in the pleural and peritoneal fluid sample volume / the pleural and peritoneal fluid sample volume.

[0053] The technical effects of the above-mentioned technical solution include: identifying and calculating the number of tangible objects (NUMS1) in microscopic examination sample images, providing basic parameters for quantitative analysis of tangible objects to be identified at the bottom, and achieving high efficiency and accuracy based on AI recognition algorithms.

[0054] The technical effects of the above-mentioned technical solution include: based on classification and recognition, combined with image-based sample quantification, accurate counting of tangible objects is achieved, further advancing the accurate quantitative analysis of tangible objects in pleural and peritoneal fluid, and enabling precise quantitative analysis of tangible objects.

[0055] The technical effects of the above-mentioned technical solution include: the volume quantitative calculation process of the pleural and peritoneal fluid detection and 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, thus eliminating the error caused by the deviation of the sample tiling height H in the calculation, and making the quantitative analysis more accurate. Attached Figure Description

[0056] Figure 1 This is one of the schematic diagrams of the AI ​​training method for detecting pleural and peritoneal effusions;

[0057] Figure 2 This is the second step in the AI ​​training method for detecting pleural and peritoneal effusions.

[0058] Figure 3 This is the third step in the AI ​​training method for detecting pleural and peritoneal effusions.

[0059] Figure 4 This is one of the schematic diagrams illustrating the pretreatment methods for pleural and peritoneal fluid samples;

[0060] Figure 5 This is the second schematic diagram of the pretreatment method for pleural and peritoneal fluid samples;

[0061] Figure 6 This is the third schematic diagram of the pretreatment method for pleural and peritoneal fluid samples;

[0062] Figure 7 This is the fourth schematic diagram of the pretreatment method for pleural and peritoneal fluid samples;

[0063] Figure 8 This is the fifth schematic diagram of the pretreatment method for pleural and peritoneal fluid samples; it includes a schematic diagram of the method for obtaining dry powder granule staining agents.

[0064] Figure 9 This is the fifth schematic diagram of the pretreatment method for pleural and peritoneal fluid samples;

[0065] Figure 10 It is one of the magnified microscopic images obtained after the pleural and peritoneal fluid samples were laid flat during microscopic examination;

[0066] Figure 11 This is the second magnified microscopic image obtained after the pleural and peritoneal fluid samples were laid flat during microscopic examination.

[0067] Figure 12 It is one of the magnified microscopic images of the formed elements extracted from the sample;

[0068] Figure 13 This is the second magnified microscopic image of the formed elements extracted from the sample.

[0069] Figures 14 to 16 It is a comparison of magnified microscopic images obtained after pleural and peritoneal fluid microscopy samples are laid flat, before and after identification and labeling. Detailed Implementation

[0070] The contents of this application will be further described in detail below with reference to the accompanying drawings. It should be noted that the following description is of preferred embodiments of the present invention and does not constitute any limitation on the present invention. The description of the preferred embodiments of the present invention is merely an explanation of the general principles of the invention. The designations "first," "second," "A," and "B" used in this invention are for ease of explanation only and do not represent a temporal or spatial order. The combinations of letters and numbers "TA," "TB," and "H" used in this invention are for ease of explanation only, and their specific meanings are determined by the specific terms they represent.

[0071] like Figure 1 In one embodiment of an AI training method for detecting pleural and peritoneal effusion, the pleural and peritoneal effusion samples are preprocessed to obtain microscopic samples; the microscopic samples are laid flat; the flat microscopic samples are photographed to obtain microscopic sample images; tangible objects in the microscopic sample images are identified and labeled to obtain labeled images; the labeled images are used for AI training to obtain a pleural and peritoneal effusion detection AI feature dataset A.

[0072] like Figure 2 In one embodiment of an AI training method for detecting pleural and peritoneal effusion, the pleural and peritoneal effusion sample is preprocessed to obtain a microscopic sample; the microscopic sample is laid flat; the flat microscopic sample is photographed to obtain a microscopic sample image; an AI recognition algorithm is used to identify and classify the microscopic sample image, outputting a tangible image A of pleural and peritoneal effusion, wherein the AI ​​recognition algorithm uses a urine AI feature dataset; the urine AI feature dataset is obtained by urine tangible object annotation training; the pleural and peritoneal effusion tangible object image A is manually reviewed, identified, and annotated to obtain an annotated image B; the annotated image B includes tangible objects in pleural and peritoneal effusion and tangible objects suspected to be in pleural and peritoneal effusion; the annotated image B is used for AI training to obtain a pleural and peritoneal effusion detection AI feature dataset B.

[0073] like Figure 3 In one embodiment of an AI training method for detecting pleural and peritoneal effusion, the pleural and peritoneal effusion sample is preprocessed to obtain a microscopic sample; the microscopic sample is laid flat; the flat microscopic sample is photographed to obtain a microscopic sample image; an AI recognition algorithm is used to identify and classify the microscopic sample image, outputting an image C of tangible objects in the pleural and peritoneal effusion; the AI ​​recognition algorithm uses either the aforementioned AI feature dataset B or the AI ​​feature dataset A for detecting pleural and peritoneal effusion; the image C of tangible objects in the pleural and peritoneal effusion is manually reviewed, identified, and labeled to obtain labeled images D; labeled images D include tangible objects in the pleural and peritoneal effusion and tangible objects suspected to be in the pleural and peritoneal effusion; AI is trained using labeled images D to obtain the AI ​​feature dataset C for detecting pleural and peritoneal effusion.

[0074] like Figure 4 Pretreatment of pleural and peritoneal fluid samples involves adding a staining agent to the samples to stain them; the staining agent is a dry powder staining agent. For example... Figure 5Pretreatment of pleural and peritoneal fluid samples involves adding reference particles to the samples; these particles are located at the bottom layer of the microscopic specimen. When photographing a flat microscopic specimen, the camera's focal plane is aligned with the center of the object. For example... Figures 4 to 6 Staining agents were added to the pleural and peritoneal fluid samples to stain the fluid; the staining agents included neomethylene blue.

[0075] like Figure 7 and Figure 8 : Take pleural and peritoneal fluid samples from organisms with increased tangible matter; such as Figure 7 Multiple pleural and peritoneal fluid samples are mixed, allowed to settle and separate into layers or centrifuged to enrich the layers, and the bottom layer is taken to obtain a microscopic sample.

[0076] like Figures 5 to 7 Staining agents are added to pleural and peritoneal fluid samples to stain the fluid; the staining agents can be liquid or dry powder / granule staining agents. Figure 6 Staining agents were added to pleural and peritoneal fluid samples to stain the fluid. The staining agent was a dry powder granule staining agent, which was obtained by quantitatively aliquoting and drying the liquid staining reagent. The liquid staining reagent included reference particles, staining agent and water, wherein the target concentration range of the reference particles could be adjusted.

[0077] like Figure 8 Artificially cultured bacteria were added to the aforementioned pleural and peritoneal fluid samples.

[0078] like Figure 9 The bacterial pleural or peritoneal fluid containing bacteria is inoculated onto a culture medium for bacterial culture to obtain bacterial colonies; an appropriate amount of bacteria is extracted from the bacterial culture medium and added to the pleural or peritoneal fluid sample to obtain an intermediate sample containing a large number of bacteria; the intermediate sample is then added to a staining reagent and thoroughly mixed to obtain a microscopic sample.

[0079] The aforementioned pleural effusion or ascites containing bacteria is derived from one or more experimental individuals; the aforementioned bacterial culture includes aerobic culture, carbon dioxide culture, microaerophilic culture, and / or anaerobic culture; the aforementioned bacterial culture medium includes: agar medium, nutrient broth medium, and / or peptone water medium.

[0080] The aforementioned AI training method for detecting pleural and peritoneal effusions can also be applied to AI training for detecting serous cavity effusions, including AI training for detecting fluids in the pleural cavity, peritoneal cavity, pericardial cavity, and testicular tunica vaginalis.

[0081] In some embodiments not shown in the accompanying figures, the microscopic sample images are sent to an AI training server via a network; the segmented images are then manually identified and labeled via the network.

[0082] A method for detecting and analyzing pleural and peritoneal effusions includes: preprocessing pleural and peritoneal effusion samples to obtain microscopic samples; photographing the laid-out microscopic samples to obtain microscopic sample images, and selecting an image area S1; the volume corresponding to the image area S1 is V1; using an AI recognition algorithm to identify the above microscopic sample image, identifying tangible matter in the pleural and peritoneal effusions within the selected area S1 of the image, and obtaining the quantity of tangible matter NUMS1; the above AI recognition algorithm includes a pleural and peritoneal effusion detection AI feature dataset; in the microscopic sample, the content of tangible matter per unit volume = NUMS1 / V1.

[0083] In the method for detecting and analyzing pleural and peritoneal fluid, the AI ​​feature dataset for pleural and peritoneal fluid detection includes the aforementioned AI feature dataset A, AI feature dataset B, or AI feature dataset C for pleural and peritoneal fluid detection. The pleural and peritoneal fluid samples undergo pretreatment, concentrating the concentration N times. The content of formed matter per unit volume in the pleural and peritoneal fluid sample is calculated as NUMS1 / (V1×N). Reference particles are added to the pretreated pleural and peritoneal fluid samples to assist in focusing. A staining agent is added to the pretreated pleural and peritoneal fluid samples to stain them. The staining agent is a dry powder staining agent, including new... The product contains methylene blue; the aforementioned tangible matter includes cellular tangible matter, which includes any one or more of the following: neutrophils, degenerative neutrophils, eosinophils, basophils, lymphocytes, macrophages, phagocytic cells, erythrocytes, erythrocytes, mesothelial cells, mast cells, and tumor cells; the aforementioned tangible matter includes microbial tangible matter, which includes any one or more of the following: bacilli, cocci, histoplasma capsulatum, and yeast; the aforementioned tangible matter includes parasitic tangible matter, which includes any one or more of the following: microfilariae and Toxoplasma gondii.

[0084] A computing processing device is used to run all or part of the above-described AI training method for pleural and peritoneal effusion detection; the memory of the computing processing device includes the above-described AI feature dataset A, AI feature dataset B, or AI feature dataset C for pleural and peritoneal effusion detection; the computing processing device is used to run all or part of the above-described pleural and peritoneal effusion detection and analysis method.

[0085] A data storage device stores all or part of the algorithm code for the above-described AI training method for pleural and peritoneal effusion detection; stores the above-described AI feature dataset A, AI feature dataset B, or AI feature dataset C for pleural and peritoneal effusion detection; and stores all or part of the algorithm code for executing the above-described pleural and peritoneal effusion detection and analysis method.

[0086] A detection device is used to run all or part of the above-mentioned AI training method for pleural and peritoneal effusion detection; and to run all or part of the above-mentioned pleural and peritoneal effusion detection and analysis method.

[0087] like Figure 10One of the magnified microscopic images obtained after laying out samples from pleural and peritoneal fluid microscopy. For example... Figure 14 This is a comparison of magnified microscopic images obtained after laying out pleural and peritoneal fluid samples for microscopic examination, before and after identification and labeling. For example... Figure 10 and Figure 14 As can be seen, the content of formed matter in some pleural and peritoneal fluid samples is very low, almost impossible to identify with naked-eye microscopy. The circled areas in the image show the formed elements identified using an AI algorithm. Figure 15 and Figure 16 As shown, Figure 15 and Figure 16 The image on the left is the picture before the identification mark was made; the formed elements in it are difficult to identify with the naked eye. Figure 15 and Figure 16 The image on the right is the image after the identification of the markers. It can be seen that the trained algorithm can identify formed elements that are difficult to identify with the naked eye. Figure 15 The formed elements identified in the sample are bacilli. Figure 16 The formed element identified in the sample is cocci.

[0088] like Figure 11 The magnified microscopic image obtained after adding bacteria to the pleural and peritoneal fluid microscopic samples and then tiling them shows the relative distribution density of formed elements in the image. Figure 10 This makes it easier to perform recognition and AI training;

[0089] Figure 12 It is a magnified microscopic image of the formed elements of the selected portion. Figure 12 and Figure 13 In the image, the top row shows the raw image before recognition, and the bottom row shows the labeled image after recognition. Figure 12 The left column shows images of a single coccus before and after identification, the middle column shows diplococci, and the right column shows streptococci. Figure 13 It is a magnified microscopic image of the formed elements of the selected portion; Figure 13 In the image, the left column shows the bacilli before and after identification, the middle column shows bacilli arranged in pairs, and the right column shows bacilli arranged in chains.

[0090] A training method for an AI model of bacteria in pleural and peritoneal fluid: The method employs centrifugation enrichment and includes the following steps: Step 1: Centrifuge and enrich peritoneal fluid samples from animals suffering from bacterial peritonitis, remove the supernatant, and add the bottom sediment to a staining reagent; Step 2: Photograph the peritoneal fluid samples from animals suffering from bacterial peritonitis, use a urine-based bacterial AI model to identify and segment images from a bacterial database to the size of AI training images, obtain a large number of bacterial images, and establish a bacterial database; Step 3: Use the urine-based bacterial AI model and image algorithms to automatically label the images in the bacterial database; Step 4: Upload the processed images to an AI training server for training to obtain an AI recognition model capable of recognizing the bacteria database in pleural and peritoneal fluid; Step 5: Use the trained pleural and peritoneal fluid bacterial AI recognition model to identify and segment subsequent original images, automatically obtain bacterial information and coordinates, and automatically label them using image algorithms; Step 6: The AI ​​model automatically identifies and labels the database again, then uploads it to the AI ​​server for enhanced training of the AI ​​model to improve AI performance.

[0091] While the present invention has been described and illustrated with reference to preferred embodiments and several alternatives, the invention is not limited to the specific descriptions herein. Other alternatives or equivalent components may also be used to practice the invention.

Claims

1. A method for training an AI to detect pleural and peritoneal effusions, characterized in that, Pleural and peritoneal fluid samples were pretreated to obtain microscopic samples; The microscopic sample is laid flat; the flat microscopic sample is photographed to obtain an image of the microscopic sample; Tangible objects in microscopic sample images are identified and labeled to obtain labeled images; AI is trained using the labeled images to obtain AI feature dataset A for pleural and peritoneal effusion detection.

2. A method for training an AI to detect pleural and peritoneal effusions, characterized in that, Pleural and peritoneal fluid samples were pretreated to obtain microscopic samples; The microscopic sample is laid flat; the flat microscopic sample is photographed to obtain an image of the microscopic sample; An AI recognition algorithm is used to identify and classify microscopic sample images, outputting image A of tangible matter in the pleural and peritoneal fluid. The AI ​​recognition algorithm uses a urine AI feature dataset, which is obtained by training with labeled urine tangible matter. The image A containing tangible objects in the pleural and peritoneal fluid was manually reviewed, identified, and annotated to obtain an annotated image B. Annotated image B includes tangible objects in the pleural and peritoneal fluid and suspected tangible objects in the pleural and peritoneal fluid. AI was trained using labeled image B to obtain AI feature dataset B for pleural and peritoneal effusion detection.

3. A method for training an AI to detect pleural and peritoneal effusions, characterized in that, Pleural and peritoneal fluid samples were pretreated to obtain microscopic samples; The microscopic sample is laid flat; the flat microscopic sample is photographed to obtain an image of the microscopic sample; AI recognition algorithms are used to identify and classify microscopic sample images, and output images C of tangible matter in pleural and peritoneal fluid. The AI ​​recognition algorithm uses the AI ​​feature dataset B or AI feature dataset A for pleural and peritoneal effusion detection as described in claim 1 or claim 2. The image C containing tangible objects in the pleural and peritoneal fluid was manually reviewed, identified, and annotated to obtain an annotated image D; the annotated image D includes tangible objects in the pleural and peritoneal fluid and suspected tangible objects in the pleural and peritoneal fluid. AI was trained using labeled image D to obtain AI feature dataset C for pleural and peritoneal effusion detection.

4. The AI ​​training method for detecting pleural and peritoneal effusions according to claim 1, 2, or 3, characterized in that, Artificially cultured bacteria were added to the pleural and peritoneal fluid samples.

5. The AI ​​training method for detecting pleural and peritoneal effusions according to claim 4, characterized in that, The pretreatment of the pleural and peritoneal fluid samples includes: Step 10: Inoculate the pleural effusion or ascites containing bacteria onto a culture medium for bacterial culture and obtain bacterial colonies; Step 20: Extract an appropriate amount of bacteria from the bacterial culture medium and add it to the pleural fluid or ascites sample to obtain an intermediate sample containing a large number of bacteria. Step 30: Take the intermediate sample and add it to the staining reagent, mix thoroughly to obtain the microscopic sample.

6. The AI ​​training method for detecting pleural and peritoneal effusions according to claim 5, characterized in that, Includes any one of the following technical features: TN1: The pleural or peritoneal fluid containing bacteria was derived from one or more experimental individuals; TN2: The bacterial culture includes aerobic culture, carbon dioxide culture, microaerophilic culture and / or anaerobic culture; TN3: The bacterial culture medium includes: agar medium, nutrient broth medium and / or peptone water medium.

7. The AI ​​training method for detecting pleural and peritoneal effusions according to claim 4, characterized in that, Pretreatment of pleural and peritoneal fluid samples includes any one of the following technical features: TA1: Add staining agent to pleural and peritoneal fluid samples to stain the pleural and peritoneal fluid; TA2: Add staining agent to the pleural and peritoneal fluid samples to stain the fluid; the staining agent is a dry powder staining agent. TA3: Add reference particles to the pleural and peritoneal fluid samples. The reference particles are located at the bottom layer of the microscopically examined samples. TA4: When photographing a flat microscopic specimen, the camera's focal plane should be aligned with the center of the object. TA5: Add staining agent to pleural and peritoneal fluid samples to stain the fluid; the staining agent includes neomethylene blue. TA6: Mix multiple pleural and peritoneal fluid samples, allow them to settle and separate into layers, and then take the bottom layer to obtain a microscopic sample. TA7: Mix multiple pleural and peritoneal fluid samples, centrifuge to separate them into layers, take the bottom layer, and obtain a microscopic sample; TA8: Take pleural and peritoneal fluid samples from organisms with increased tangible matter. TA9: Add staining agent to pleural and peritoneal fluid samples to stain the fluid; the staining agent is a dry powder granular staining agent. TA10: Add staining agent to pleural and peritoneal fluid samples to stain the pleural and peritoneal fluid; the staining agent is a dry powder granule staining agent; the dry powder granule staining agent is obtained by quantitatively aliquoting and drying the liquid staining reagent; the liquid staining reagent includes reference particles, staining agent and water, wherein the target concentration range of the reference particles can be adjusted; TA11: The AI ​​training method for detecting pleural and peritoneal effusion can also be applied to AI training for detecting serous cavity effusion, including AI training for detecting fluid in the pleural cavity, peritoneal cavity, pericardial cavity, and testicular tunica vaginalis. TA12: The microscopic sample images are sent to the AI ​​training server via the network; TA13: 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.

8. A method for detecting and analyzing pleural and peritoneal effusions, characterized in that, Pleural and peritoneal fluid samples were pretreated to obtain microscopic samples; Take a picture of the laid-out microscopic sample to obtain a microscopic image of the sample, and select the image area S1; the volume corresponding to the image area S1 is V1; The AI ​​recognition algorithm is used to identify the microscopic sample image, and the tangible objects in the pleural and peritoneal water in the selected area S1 of the image are identified to obtain the number of tangible objects NUMS1. The AI ​​recognition algorithm includes an AI feature dataset for detecting pleural and peritoneal effusions. In microscopic examination of a sample, the content of tangible matter per unit volume = NUMS1 / V1.

9. The method for detecting and analyzing pleural and peritoneal effusions according to claim 8, characterized in that, Includes any one of the following technical features: TG1: The AI ​​feature dataset for pleural and peritoneal effusion detection includes any one of the AI ​​feature datasets for pleural and peritoneal effusion detection, such as A, B, or C, as described in any one of claims 1 to 3. TG2: The pleural and peritoneal fluid samples were pretreated and the concentration was concentrated N times. The content of tangible matter per unit volume in the pleural and peritoneal fluid samples = NUMS1 / (V1×N). TG3: The pleural and peritoneal fluid samples were pretreated and reference particles were added. The reference particles were used to assist in focusing. TG4: Pleural and peritoneal fluid samples are pretreated and then stained with a staining agent. TG5: Pleural and peritoneal fluid samples are pretreated and then stained with a dry powder staining agent. TG6: Pleural and peritoneal fluid samples were pretreated with staining agents, including neomethylene blue. TG7: The tangible matter includes cellular tangible matter, which includes any one or more of the following: neutrophils, degenerated neutrophils, eosinophils, basophils, lymphocytes, macrophages, phagocytic cells, erythrocytes, shadow erythrocytes, mesothelial cells, mast cells, and tumor cells. TG8: The tangible matter includes microbial tangible matter, which includes: bacilli, cocci, capsular histoplasmosis bacteria, yeast, any one or more; TG9: The tangible objects include parasitic tangible objects, which include: microfilariae, Toxoplasma gondii, any one or more.

10. A computing processing device, characterized in that, Includes any one of the following technical features: TEA1: Used to run all or part of the AI ​​training method for detecting pleural and peritoneal effusions as described in any one of claims 1 to 8; TEA2: The memory of the computing processing device includes the AI ​​feature dataset A, AI feature dataset B, or AI feature dataset C for pleural and peritoneal effusion detection as described in any one of claims 1 to 7; TEA3: Used to run all or part of the pleural and peritoneal fluid detection and analysis method according to any one of claims 8 to 9.

11. A data storage device, characterized in that, Includes any one of the following technical features: TEB1: Stores all or part of the algorithm code for performing the AI ​​training method for detecting pleural and peritoneal effusion as described in any one of claims 1 to 7; TEB2: Store the AI ​​feature dataset A, AI feature dataset B, or AI feature dataset C for pleural and peritoneal effusion detection as described in any one of claims 1 to 7; TEB3: Stores all or part of the algorithm code for performing the pleural and peritoneal effusion detection and analysis method according to any one of claims 8 to 9.

12. A detection device, characterized in that, Includes any one of the following technical features: TEC1: Used to run all or part of the AI ​​training method for detecting pleural and peritoneal effusions as described in any one of claims 1 to 9; TEC4: Used to run all or part of the pleural and peritoneal fluid detection and analysis method according to any one of claims 9 to 10.

Images