A method and system for determining the results of blood clotting and blood clot inhibition experiments

Abstract

This invention discloses a method and system for determining the results of blood coagulation and blood coagulation inhibition experiments, comprising: placing a 96-well plate for the blood coagulation or blood coagulation inhibition experiment into an opaque, sealed cavity, embedding it in a slot on an LED backlight panel, turning on the LED backlight panel and tilting it to a predetermined angle to cause red blood cells to move accordingly at the bottom of the wells, and after standing for a predetermined time, obtaining an image containing all wells; extracting multiple features from each well in the image and inputting them into the corresponding basic classification model for independent judgment; inputting the image and the output results of the multiple basic classification models into three trained artificial intelligence models to determine the state of each well; and obtaining the blood coagulation or blood coagulation inhibition experiment results based on the states of all wells. This invention has low user learning costs and low equipment costs; it can automatically and quickly complete the simultaneous determination of multiple samples; and it uses fusion feature engineering technology to highlight features, resulting in higher accuracy.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of medical testing and artificial intelligence, and is particularly applicable to influenza A-H1N1 and B-Victoria types. More specifically, it relates to a system and method for measuring the results of hemagglutination and hemagglutination inhibition tests for influenza. Background Technology

[0002] The hemagglutination assay (HA) and hemagglutination inhibition assay (HI) for influenza are classic and core serological and virological experimental techniques for influenza virus detection, identification, and research. The two complement each other and are indispensable tools for influenza prevention and control, laboratory monitoring, and basic research. Their importance is reflected in the entire chain of work, including influenza virus detection, typing, antibody evaluation, vaccine development and efficacy verification. They are also the standard detection methods of the World Health Organization (WHO) Influenza Surveillance Network.

[0003] Currently, experimental results mainly rely on human visual recognition, which is labor-intensive. Limited by technical capabilities and individual experience, each experimenter's recognition is highly subjective, leading to unstable results. Therefore, there is an urgent need to develop a device that can replace manual labor and improve the efficiency and accuracy of this experiment.

[0004] Traditional AI algorithms in image recognition can be used to distinguish between HA and HI results. However, in real-world scenarios, HA and HI results exhibit significant heterogeneity, rendering traditional algorithms relatively inaccurate. Therefore, developing a recognition scheme that closely matches this experiment and improves accuracy is of practical value. Summary of the Invention

[0005] This invention addresses the problems of hemagglutination and hemagglutination inhibition experiments relying on manual interpretation, high subjectivity, and poor consistency of results, as well as the shortcomings of existing image recognition algorithms in distinguishing heterogeneous results. It proposes a method and system for determining the hemagglutination and hemagglutination inhibition experimental results of human A-H1N1 and B-Victoria influenza types.

[0006] The technical solution of this invention is:

[0007] In a first aspect, the present invention provides a method for determining the results of blood coagulation and blood coagulation inhibition experiments, comprising:

[0008] Step 1: Place the 96-well plate to be tested for blood coagulation or blood coagulation inhibition in an opaque closed cavity and embed it in a slot on an LED backlight plate. The LED backlight plate is controllably tiltable and an industrial camera is located vertically above the LED backlight plate.

[0009] Step 2: Turn on the LED backlight and tilt it to a predetermined angle so that the red blood cells move accordingly at the bottom of the well. After standing for a predetermined time, take a picture of the 96-well plate with the industrial camera to obtain an image containing all the wells.

[0010] Step 3: Extract multiple features for each hole in the image and input them into the corresponding basic classification model for independent judgment;

[0011] Step 4: Input the image and the output results of multiple basic classification models into three trained artificial intelligence models. Use the integration rule that at least two of the three artificial intelligence models output non-agglutination results to determine the state of each well. Based on the state of all wells, obtain the blood coagulation or blood coagulation inhibition experimental results.

[0012] Furthermore, the opaque, enclosed cavity is equipped with an LED backlight panel, a fixed support rod, a telescopic support rod, and an industrial camera; wherein,

[0013] The LED backlight panel uses a white LED light source and is fixed in the center of the bottom of the cavity. The upper surface of the LED backlight panel has a slot for placing a 96-hole plate. The fixed support rod is connected to one side of the bottom of the LED backlight panel, and the telescopic support rod is connected to the other side of the bottom of the LED backlight panel. It is driven by a stepper motor and is used to tilt the LED backlight panel and the 96-hole plate above it to a predetermined angle.

[0014] The industrial camera is fixedly installed in an opaque, enclosed cavity and positioned directly above the 96-hole plate when the LED backlight is tilted to a predetermined angle. The industrial camera is connected to an external computer.

[0015] Further, in step 3, the first type of feature is extracted for each hole location in the image, and input into the first basic classification model for determination, including:

[0016] Step 31: For each aperture, obtain the blue component light intensity gradient and the green component light intensity gradient along the radial path from the upper edge of the aperture to the near-center region.

[0017] Step 32: Fit the blue component light intensity gradient and the green component light intensity gradient using an exponential decay method to obtain their respective fitting coefficients;

[0018] Step 33: Determine the judgment result of the first type of feature based on the relationship between the fitting coefficient of the blue component and the preset blue threshold, and the relationship between the fitting coefficient of the green component and the preset green threshold.

[0019] Further, in step 32, the intensity gradients of the blue component and the green component are fitted using the following exponential decay form to obtain their respective fitting coefficients. :

[0020]

[0021] in, Indicates color, For color Intensity component in gradient at, Indicates the position of a pixel on the vertical path. The radius of the hole position. This represents the maximum aggregation radius of red blood cells at the center of the pore.

[0022] If the fitting coefficient of the blue component is greater than the preset blue threshold, the blue component of the first type of feature is determined to be non-aggregated;

[0023] If the fitting coefficient of the green component is greater than the preset green threshold, the green component of the first type of feature is determined to be non-agglomerated.

[0024] Further, in step 3, a second type of feature is extracted for each hole location in the image, and input into the second basic classification model for determination, including:

[0025] Step 34: Within the maximum aggregation radius of red blood cells at the center of each well, use a contour detection algorithm to determine whether there is a red deposition area contour.

[0026] If no contour is detected, the second type of feature is determined to be non-aggregated;

[0027] If a contour is detected, obtain the number of pixels on the major axis and the number of pixels on the minor axis of the contour;

[0028] Step 35: Calculate the ratio of the number of pixels on the major axis to the number of pixels on the minor axis, and compare it with a preset ratio threshold. If it is greater than the threshold, the second type of feature is determined to be aggregated; otherwise, it is determined to be non-aggregated.

[0029] Furthermore, in step 3, a third type of feature is extracted for each hole location in the image, and input into the third basic classification model for determination, including:

[0030] Step 36: Draw a perpendicular line at half the distance from the bottom edge to the center of each hole. The perpendicular line intersects the edge of the hole to form a horizontal scanning line segment. Collect blue light intensity sequence and green light intensity sequence point by point along the scanning line segment from the edge to the center.

[0031] Step 37: Count the number of consecutive points less than the preset low light intensity threshold in the blue light intensity sequence and the green light intensity sequence respectively. Based on the consecutive point threshold n and the preset element number threshold, determine the judgment result of the third type of feature.

[0032] If a low-intensity and basically continuous region appears in the blue light intensity sequence, then the third type of blue component is determined to be agglomeration.

[0033] If a low-intensity and basically continuous region appears in the green light intensity sequence, then the third type of green component is determined to be agglomeration.

[0034] Furthermore, step 37 includes:

[0035] Step 371: Obtain the light intensity sequence S ,include 1 pixel ;in, Indicates the pixel number. Represents light intensity sequence S The total number of pixels in the middle, , They represent the numbers respectively. The position of the pixel along the horizontal path and the light intensity;

[0036] Step 372, Traversal Take from 1 ,calculate dimensional vector ; where, linear mapping function The linear function passes through the point and , position The ordinate value on the corresponding straight line Construct into vectors ;

[0037] Step 373, Traversal Take from 1 , position Corresponding light intensity Establish dimensional vector ;

[0038] Step 374, Construction dimensional vector The first part of the vector The elements are and The difference, later The zero elements are used to pad the vector;

[0039] Step 375, Traversal Take from 1 , build 3D matrix ;

[0040] Step 376: Construct the collection ;in, Threshold for consecutive point counts; light intensity sequence S In, each pixel All corresponding vectors ,like There is continuity If any element is negative, then the pixel is... Include in set L ,gather L Represents light intensity sequence S Areas where light intensity suddenly drops; such as There is continuity in the middle If each element is negative, then... Include in set L ;

[0041] Step 377: Statistical Set L If the number of elements in the data exceeds a preset threshold... At this point, the third type of feature is determined to be agglomeration.

[0042] Furthermore, step 4 includes:

[0043] Step 41: Combine the multiple judgment values ​​corresponding to the multiple features with the ordered one-dimensional unfolded values ​​of the RGB channels of the hole position image to form a feature vector;

[0044] Step 42: Input the feature vectors into the trained artificial intelligence models, including support vector machines, random forests, and logistic regression models, respectively, to obtain their independent prediction results;

[0045] Step 43: When at least two of the three artificial intelligence models output non-agglomeration results, the hole location is finally determined to be non-agglomeration; otherwise, it is determined to be agglomeration.

[0046] Furthermore, the three artificial intelligence models, namely support vector machine, random forest, and logistic regression, are obtained by pre-training on a labeled dataset; the labeled dataset is composed of the consistent results obtained by multiple professionals through independent judgment.

[0047] Secondly, the present invention provides a system for determining the results of blood coagulation and blood coagulation inhibition experiments, comprising:

[0048] An opaque enclosed cavity includes an LED backlight panel and an industrial camera; the LED backlight panel is disposed inside the opaque enclosed cavity, and its upper surface is provided with a slot for placing a 96-hole plate, and the LED backlight panel is configured to be tilted controllably; the industrial camera is fixedly disposed inside the opaque enclosed cavity, and is located directly above the 96-hole plate when the LED backlight panel is tilted.

[0049] The control and processing unit is electrically connected to both the LED backlight panel and the industrial camera. The control and processing unit includes:

[0050] The image acquisition module is used to control the LED backlight panel to turn on and tilt to a predetermined angle. After a predetermined time, it controls the industrial camera to take a picture of the 96-hole plate to obtain an image containing all the holes.

[0051] The basic classification module contains multiple basic classification models, which are used to extract multiple types of features for each hole in the image, make independent judgments, and obtain multiple basic judgment results.

[0052] The integrated judgment module includes three trained artificial intelligence models. It is used to input the image and the multiple basic judgment results into the three trained artificial intelligence models, and to determine the state of each well by using the integrated rule that at least two of the three artificial intelligence models output non-agglutination results. Based on the state of all wells, the blood coagulation or blood coagulation inhibition experimental results are obtained.

[0053] The beneficial effects of this invention are:

[0054] This invention, while maintaining the classic detection procedures for HA and HI experiments, effectively solves the problems of poor accuracy and unstable results in manual interpretation, as well as the insufficient adaptability of traditional image recognition to heterogeneous samples, through various feature recognition and judgment methods. It provides a highly efficient, accurate, and automated technical support for laboratory detection of influenza viruses, specifically including the following advantages:

[0055] 1. Improved objectivity and consistency of result interpretation: This invention constructs a stable image acquisition environment using an opaque, sealed cavity and an industrial camera, combined with a controllable tilting LED backlight, enabling red blood cells to form clear positional features at the bottom of the well. Based on this, through multi-feature extraction and multi-model integrated judgment, subjective differences caused by manual interpretation are avoided, ensuring good consistency in results measured by different operators and at different times.

[0056] 2. Improve the accuracy of identifying heterogeneous results: Addressing the diverse and highly heterogeneous nature of HA and HI experimental results, this invention designs multiple feature extraction methods that deeply match the experimental phenomena. Specifically, these include: a first type of feature based on fitting the radial light intensity gradient within the aperture to capture the continuous changes in erythrocyte sedimentation; a second type of feature based on the contour morphology of the central region to identify geometric differences in whether agglutination occurs; and a third type of feature based on detecting low-intensity regions in the transverse light intensity sequence to locate the distribution of dark areas formed by blood cell sedimentation. These three types of features are independently judged by their respective basic classification models, and then input into three artificial intelligence models: Support Vector Machine, Random Forest, and Logistic Regression. An ensemble rule of determining non-agglutination based on the majority non-agglutination is used for comprehensive judgment. This multi-feature, multi-model architecture effectively addresses the challenge of heterogeneous experimental results and achieves higher accuracy than a single traditional image recognition algorithm.

[0057] 3. Automating the detection process and reducing manual workload: This invention forms a complete automated closed loop from image acquisition, feature extraction, model determination to result output. Operators only need to place the 96-well plate; subsequent steps are completed automatically by the system, significantly reducing manual operation and subjective judgment, improving experimental efficiency, and is especially suitable for influenza surveillance scenarios with large sample sizes and high detection frequency.

[0058] 4. Enhanced reliability and traceability of results: This invention ensures the consistency of the acquisition environment through an opaque, sealed cavity, and reduces the risk of misjudgment by a single model by combining multi-model integrated judgment, making the final results more robust and reliable. Simultaneously, the system can output a borehole status distribution map and a custom report, facilitating result archiving, verification, and subsequent analysis.

[0059] Other features and advantages of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0060] The above and other objects, features and advantages of the present invention will become more apparent from the more detailed description of exemplary embodiments of the invention in conjunction with the accompanying drawings, wherein the same reference numerals generally represent the same components in the exemplary embodiments of the invention.

[0061] Figure 1 A structural diagram of the system for measuring the results of blood coagulation and blood coagulation inhibition experiments is shown.

[0062] Figure 2 A schematic diagram of a single hole in a 96-hole plate according to an embodiment of the present invention is shown. Detailed Implementation

[0063] Preferred embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While preferred embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein.

[0064] Example 1

[0065] like Figure 1 , 2 As shown, this invention provides a method for determining the results of blood coagulation and blood coagulation inhibition experiments. The aim is to improve the accuracy and efficiency of experimental result determination through automation and intelligent means. The entire method includes multiple steps such as sample preparation, image acquisition, feature extraction, and classification, gradually realizing the transformation from raw data to final experimental results. Specifically:

[0066] Step 1: Place the 96-well plate 5 to be tested for blood coagulation or blood coagulation inhibition in the opaque sealed cavity 1 and embed it in the slot on the LED backlight plate 3. The LED backlight plate 3 is tiltable and an industrial camera 6 is located vertically above the LED backlight plate 3.

[0067] Specifically, the opaque sealed cavity 1 is a closed environment used to isolate external light interference. Its internal dimensions are adapted to a standard 96-well plate 5 to ensure stable lighting conditions during experiments. The LED backlight 3 inside the cavity uses a uniform white light source with adjustable brightness. The depth and shape of the slots on the upper surface of the backlight 3 perfectly match the bottom of the 96-well plate 5, ensuring that the plate will not slide or shift during tilting. The industrial camera 6 is fixedly mounted on the top of the cavity. Through its high-resolution lens, it can cover the entire field of view of the 96-well plate 5 for subsequent image acquisition.

[0068] In one possible implementation, the tilting function of the LED backlight panel 3 is achieved through a fixed support rod 7 and a telescopic support rod 4 connected to the bottom. The fixed support rod 7 is located on one side of the backlight panel, providing a stable support point, while the telescopic support rod 4 is located on the other side and is driven by a stepper motor, enabling precise control of the tilt angle of the backlight panel.

[0069] Step 2: Turn on the LED backlight 3 and tilt it to a predetermined angle to cause the red blood cells to move accordingly at the bottom of the wells. After standing for a predetermined time, take a picture of the 96-well plate 5 using the industrial camera 6 to obtain an image containing all the wells. For example, tilt it to a predetermined angle between 45° and 60° (preferably 60°) to make the 96-well plate tilted, and let it stand for 10s to 30s (preferably 20s) to allow the red blood cells to slide and settle along the well walls.

[0070] Specifically, the LED backlight is turned on to provide uniform background illumination, ensuring clear contrast of the blood sample within the well in the image. Tilting the well to a predetermined angle utilizes gravity to cause red blood cells to shift at the bottom of the well, forming a specific distribution pattern. For example, in an agglutinated state, red blood cells may aggregate at the center of the well bottom, while in a non-agglutinated state, they may disperse or deposit at the edges. The predetermined settling time is adjusted according to the experimental type and blood sample characteristics to allow the red blood cell displacement to stabilize.

[0071] Step 3: Extract multiple features for each hole in the image and input them into the corresponding basic classification model for independent judgment.

[0072] Specifically, three types of features are extracted from each well site sub-image. The first type of feature is based on the color component intensity gradient, focusing on analyzing the color change pattern from the edge to the center of the well site; the second type of feature is based on the contour morphology, focusing on the shape characteristics of the erythrocyte deposition zone in the central region of the well site; the third type of feature is based on the intensity distribution, analyzing the continuous change of intensity along a specific path of the well site. The extraction and determination process of each type of feature is described in detail below.

[0073] Extract the first type of feature for each hole location in the image and input it into the first basic classification model for determination; including:

[0074] Step 31: For each aperture, obtain the blue component light intensity gradient and the green component light intensity gradient along the radial path from the upper edge of the aperture to the near-center region.

[0075] Step 32: Fit the blue component light intensity gradient and the green component light intensity gradient using an exponential decay method to obtain their respective fitting coefficients. :

[0076]

[0077] in, Indicates color, For color Intensity component in gradient at, Indicates the position of a pixel on the vertical path. The radius of the hole position. This represents the maximum aggregation radius of red blood cells at the center of the pore.

[0078] Step 33: Determine the judgment result of the first type of feature based on the relationship between the fitting coefficient of the blue component and the preset blue threshold, and the relationship between the fitting coefficient of the green component and the preset green threshold;

[0079] If the fitting coefficient of the blue component is greater than the preset blue threshold, the blue component of the first type of feature is determined to be non-aggregated;

[0080] If the fitting coefficient of the green component is greater than the preset green threshold, the green component of the first type of feature is determined to be non-agglomerated.

[0081] Extract the second type of features for each hole location in the image, and input them into the second basic classification model for determination; including:

[0082] Step S34: Within the maximum aggregation radius of red blood cells at the center of each well, a contour detection algorithm is used to determine whether there is a red deposition area contour.

[0083] If no contour is detected, the second type of feature is determined to be non-aggregated;

[0084] If a contour is detected, obtain the number of pixels on the major axis and the number of pixels on the minor axis of the contour;

[0085] Step 35: Calculate the ratio of the number of pixels on the major axis to the number of pixels on the minor axis, and compare it with a preset ratio threshold. If it is greater than the threshold, the second type of feature is determined to be aggregated; otherwise, it is determined to be non-aggregated.

[0086] Extract the third type of feature for each hole location in the image, and input it into the third basic classification model for judgment, including:

[0087] Step 36: Draw a perpendicular line at half the distance from the bottom edge to the center of each hole. The perpendicular line intersects the edge of the hole to form a horizontal scanning line segment. Collect blue light intensity sequence and green light intensity sequence point by point along the scanning line segment from the edge to the center.

[0088] Step 37: Count the number of consecutive points in the blue and green light intensity sequences that are less than a preset low light intensity threshold, and then calculate the number of consecutive points based on the threshold n and the preset element count threshold. To determine the judgment result of the third type of feature;

[0089] If a low-intensity and basically continuous region appears in the blue light intensity sequence, then the third type of blue component is determined to be agglomeration.

[0090] If a low-intensity and basically continuous region appears in the green light intensity sequence, then the third type of green component is determined to be agglomeration.

[0091] Specifically, step 37 includes:

[0092] Step 371: Obtain the light intensity sequence S ,include 1 pixel ;in, Indicates the pixel number. Represents light intensity sequence S The total number of pixels in the middle, , They represent the numbers respectively. The position of the pixel along the horizontal path and the light intensity;

[0093] Step 372, Traversal Take from 1 ,calculate dimensional vector ; where, linear mapping function The linear function passes through the point and , position The ordinate value on the corresponding straight line , Construct into vectors ;

[0094] Step 373, Traversal Take from 1 , position Corresponding light intensity Establish dimensional vector ;

[0095] Step 374, Construction dimensional vector The first part of the vector The elements are and The difference, later The zero elements are used to pad the vector;

[0096] Step 375, Traversal Take from 1 , build 3D matrix ;

[0097] Step 376: Construct the collection ;in, Threshold for consecutive point counts; light intensity sequence S In, each pixel All corresponding vectors ,like There is continuity If any element is negative, then the pixel is... Include in set L ,gather L Represents light intensity sequence S An area of ​​sudden drop in light intensity appeared;

[0098] Step 377: Statistical Set L If the number of elements in the data exceeds a preset threshold... At this point, the third type of feature is determined to be agglomeration.

[0099] In this embodiment, steps 372 to 377 involve constructing vectors and matrices based on the light intensity sequence, analyzing regions of sudden light intensity drops, counting the number of consecutive low-intensity points and elements, and determining the third type of feature result. Specifically, these steps use a series of data processing methods to identify whether there are significant continuous low-intensity regions in the light intensity sequence. The number of elements in the sudden drop region is counted; if it exceeds a preset threshold, it is determined to be an agglomeration state. This invention supports custom threshold parameters. For example, the threshold for the number of consecutive points can be adjusted according to the length of the scan line segment; if the line segment is short, the threshold can be appropriately lowered. The threshold for the number of elements can be adjusted according to the experimental type; if the experiment has high sensitivity requirements for the agglomeration state, the threshold can be appropriately lowered. After processing, the system outputs the determination result of the third type of feature and records key data during the analysis process for review by experimental personnel.

[0100] The extraction and determination of the third type of features are mainly based on the continuous characteristics of light intensity distribution, which can effectively capture the typical features of red blood cell deposition areas, especially when the red blood cell distribution is relatively concentrated, the determination results have high reliability. However, in cases of high image noise or uneven red blood cell distribution, the light intensity sequence may lead to misjudgment, so it is necessary to combine it with other features for comprehensive analysis.

[0101] Step 4: Input the image and the output results of multiple basic classification models into three trained artificial intelligence models. Use the integration rule that at least two of the three artificial intelligence models output non-agglutination results to determine the state of each well. Based on the state of all wells, obtain the blood coagulation or blood coagulation inhibition experimental results.

[0102] Specifically, this step aims to synthesize the preliminary judgment results of multiple features extracted in the preceding steps, perform secondary analysis using multiple independently trained artificial intelligence models, and combine ensemble rules to derive the final state of each well. This fully leverages the analytical advantages of different models, reduces potential biases from a single model, and thus improves the accuracy and reliability of the judgment. Finally, based on the state distribution of all 96 wells, the system automatically generates the overall results of the blood coagulation or blood coagulation inhibition experiments for researchers' reference.

[0103] In one embodiment, the implementation process of step 4 includes three sub-steps: feature integration, model prediction, and result integration.

[0104] Step 41 involves combining multiple judgment values ​​corresponding to various feature classes with the ordered one-dimensional expansion values ​​of the color channels of the borehole image to form a feature vector. The judgment values ​​corresponding to the various feature classes include the preliminary judgment results of the first, second, and third feature classes from the preceding steps. Each feature class may contain multiple sub-items; for example, the first feature class includes judgment results for the blue and green components, representing the aggregated or non-aggregated states numerically. Furthermore, the ordered one-dimensional expansion values ​​of the color channels of the borehole image refer to arranging the pixel values ​​of the red, green, and blue color channels of the borehole sub-image into a one-dimensional array in a predetermined order, which is then added to the feature vector as the original image features. This feature integration method of the present invention can simultaneously preserve the detailed information of the image and the analysis results of feature extraction, providing comprehensive data support for subsequent model prediction.

[0105] Step 42: Input the feature vectors into the trained AI models, including Support Vector Machine (SVM), Random Forest, and Logistic Regression, respectively, to obtain their independent prediction results. Specifically, the SVM model identifies the optimal hyperplane in the feature space, classifying data into aggregated and non-aggregated categories, suitable for situations where there are clear linear boundaries in the feature vectors; the Random Forest model constructs multiple decision trees and combines voting results for classification, effectively handling non-linear relationships and high-dimensional data in feature vectors; the Logistic Regression model outputs classification results through probability estimation, suitable for situations where the category distribution in the feature vectors is relatively balanced. The three models run independently, outputting their respective prediction results for the same feature vector, typically with 0 representing non-aggregation and 1 representing aggregation.

[0106] Step 43: If at least two of the three AI models output non-agglomeration results, the well location is ultimately determined to be non-agglomeration; otherwise, it is determined to be agglomeration. Specifically, this step adopts the ensemble rule of determining non-agglomeration based on majority non-agglomeration. That is, if at least two of the three models (Support Vector Machine, Random Forest, and Logistic Regression) predict non-agglomeration results, the well location is ultimately determined to be non-agglomeration; if two or more models predict agglomeration results, it is ultimately determined to be agglomeration. This effectively reduces the possibility of false agglomeration and improves the reliability of experimental results.

[0107] Example 2

[0108] This invention provides a system for determining the results of blood coagulation and blood coagulation inhibition experiments, comprising:

[0109] An opaque sealed cavity 1 includes an LED backlight panel 3 and an industrial camera 6; the LED backlight panel 3 is disposed inside the opaque sealed cavity 1, and its upper surface is provided with a slot for placing a 96-hole plate 5, and the LED backlight panel 3 is configured to be tilted controllably; the industrial camera 6 is fixedly disposed inside the opaque sealed cavity 1, and is located directly above the 96-hole plate 5 when the LED backlight panel 3 is tilted.

[0110] Control and processing unit 2 is electrically connected to LED backlight panel 3 and industrial camera 6 respectively. Control and processing unit 2 includes:

[0111] The image acquisition module is used to control the LED backlight panel 3 to turn on and tilt to a predetermined angle. After a predetermined time, it controls the industrial camera 6 to take a picture of the 96-hole plate 5 to obtain an image containing all the holes.

[0112] The basic classification module contains multiple basic classification models, which are used to extract multiple types of features for each hole in the image, make independent judgments, and obtain multiple basic judgment results.

[0113] The integrated judgment module includes three trained artificial intelligence models. It is used to input the image and the multiple basic judgment results into the three trained artificial intelligence models, and to determine the state of each well by using the integrated rule that at least two of the three artificial intelligence models output non-agglutination results. Based on the state of all wells, the blood coagulation or blood coagulation inhibition experimental results are obtained.

[0114] In practice:

[0115] I. Main Reagents and Consumables

[0116] Viral fluid: Influenza virus MDCK cells.

[0117] Red blood cells: Turkey blood, obtained from the slaughterhouse, was immediately mixed with Aspergillus solution at a 1:1 ratio, gently shaken, and brought back to the laboratory. Blood cells were evenly distributed into 15ml centrifuge tubes, each tube not exceeding 10ml. The tubes were leveled and centrifuged at 4°C and 2000rpm for 5 minutes. The supernatant plasma, Aspergillus solution, white film layer, and a small amount of red blood cells beneath the white film layer were removed. A suitable amount of PBS was added and gently mixed by pipetting. The ratio of turkey red blood cells to PBS was 1:2. The mixture was leveled again and centrifuged at 2000rpm for 5 minutes. The supernatant was removed, and the mixture was washed three times with PBS to prepare concentrated red blood cells. The prepared concentrated red blood cells were added to PBS at a concentration of 1% (v / v), gently shaken, and labeled with the preparer, preparation time, red blood cell type, and concentration. The mixture was stored at 4°C for 7 days.

[0118] Buffer reagents: 0.85% physiological saline, phosphate buffer (PBS, pH 7.4), sterilized and stored at room temperature.

[0119] Serum: Blood samples collected should be immediately separated into serum, aliquoted into centrifuge tubes and clearly labeled. Samples stored at 4°C should be tested within 5 days. Samples stored beyond 5 days should be frozen at low temperature. Violent shaking is prohibited during transportation and storage.

[0120] Receptor-destroying enzyme (RDE) (Lot340016, blue).

[0121] II. Auxiliary Instruments and Equipment

[0122] The entire experiment was conducted in the BSL-2 laboratory, equipped with 96-well V-type blood coagulation plates, 12-channel micropipettes, a constant temperature water bath, a high-speed centrifuge, and a vortex mixer.

[0123] III. Preparation of Samples to be Tested

[0124] 1. HA Experiment

[0125] 1.1 Obtain influenza virus solution.

[0126] 1.2 Place the 96-well plate stably on a sterile work surface and label the wells. Add PBS. Attach a 200 μL pipette with a filter tip to an 8-channel pipette and add 50 μL of PBS to the second well of the microplate. Continue adding 50 μL of PBS to each well until the last well. Add the virus to be tested. Attach a 200 μL pipette with a filter tip to a single-channel pipette and add 100 μL of the virus solution to the corresponding well in the first column of the microplate. Add 100 μL of PBS to the last row of wells as a red blood cell control. Attach a 200 μL pipette with a filter tip to a 12-channel pipette and add 50 μL of virus solution from each well in the first column to the corresponding well in the second column, mixing several times. Perform serial dilutions from the second to the twelfth column of the microplate. Discard 50 μL of liquid from each well in the last column. Attach a 200 μL dropper with a filter cartridge to the 12-channel pipette and aspirate 50 μL of red blood cell suspension into the sample well. Add 50 μL of 1% red blood cell suspension to each well, gently tap the microplate to thoroughly mix the red blood cells and virus, and incubate the reaction plate at room temperature for 30 minutes before testing.

[0127] 2. HI Experiment

[0128] 2.1 Serum Treatment: All sera were treated to remove complement and non-specific inhibitory factors. 20 µl of each serum sample was added, followed by 60 µl of RDE and incubated overnight at 37°C. The next morning, the water bath was adjusted to 56°C and incubated for 30 min. Then, 120 µl of PBS was added to each well and the solution was stored at 4°C. The ratio of serum:RDE:PBS = 1:3:6. Based on the hemagglutination titer determined by the HA assay, the standard viral strain was diluted with PBS to 4 hemagglutination units of antigen (4 HAU).

[0129] 2.2 Place the 96-well plate stably on a sterile operating table and label the wells. Add 25 µL of PBS to each well of the hemagglutination plate. Then, add 25 µL of the serum to be tested to each of the wells 1-12 in the first row. Using a multichannel pipette, take 25 µL from each of the wells 1-12 in the first row and serially dilute the serum 2-fold from the first to the eighth row, discarding the 25 µL from the last row. Add 25 µL of the prepared 4 hemagglutination units antigen to each well of the hemagglutination plate, tap the plate gently, and incubate at room temperature for 30 min. Take 8 unused wells and add 25 µL of PBS to each well. Take the corresponding positive control serum (positive control) and serially dilute the serum 2-fold, discarding the 25 µL from the last row. Add 25 µL of 4 hemagglutination units antigen to each well and incubate at room temperature for 30 min. Add 50 µL of 1% turkey blood cells to each well of the V-shaped hemagglutination plate, tap the plate gently to mix the red blood cells and virus thoroughly. Incubate at room temperature for 30 min before testing.

[0130] IV. Apparatus

[0131] like Figure 1 As shown, a closed cavity is constructed with a white diffuser panel lining its inner wall to reduce direct light reflection. Inside the cavity, one fixed support rod and one retractable support rod support a flicker-free white LED light source (24V power supply, 22V output voltage). Above the LED light source is a mounting position for a 96-hole plate, the angle of which can be adjusted to 60° vertically using the retractable support rod. Opposite to the LED light source, an industrial camera (120° distortion-free, 8K resolution, camera color temperature set to 6000K) is fixed above it.

[0132] The industrial camera connection control and processing unit 2 is mainly configured with an Intel I9-12900K CPU, 64GB of memory, and the software and algorithm development platform is VS Code, with Python as the programming language.

[0133] Model training, testing, and validation:

[0134] The agglutination and antiagglutination tests for A-H1N1 and B-Victoria influenza types use the same standard operating procedures. This example uses the agglutination test as an example.

[0135] It is worth noting that when tilting the 96-well plate in the apparatus, the reading determination time is relatively short. Therefore, one skilled experimenter is required to operate the apparatus and take photos, while three senior experts are responsible for recording the readings. The photos are then cropped to obtain images of each well, as shown below. Figure 2 As shown, the result of each well site labeling is used as the dependent variable, where 1 is recorded when the expert determines it is non-agglomerative and 0 is recorded when it is determined to be agglomerative. The above experiment is repeated until sufficient data is obtained for training and testing, and the final model is obtained.

[0136] Sixteen 96-well plates were prepared, with eight used for A-H1N1 validation and eight for B-Victoria validation. Sixty-four randomly diluted A-H1N1 antigens were used to determine the A-H1N1 hemagglutination titer; 64 randomly diluted B-Victoria antigens were used to determine the B-Victoria hemagglutination titer. All experimental procedures except for result interpretation were performed.

[0137] The 16 96-well plates were placed sequentially in the device and tested using the algorithm described above. At the same time, experts took readings, and the median value of the blood coagulation titer recorded by three senior experts was used as the gold standard for comparison. The results are shown in Table 1.

[0138] Table 1. Accuracy of Artificial Intelligence Algorithms

[0139]

[0140] As shown in Table 1, the accuracy of this method exceeds 95%. It is worth mentioning that the measured values ​​of the five samples that were not accurately measured were only one unit different from the gold standard.

[0141] Example 2: Sample Detection and Comparison under Real-World Conditions

[0142] For the hemagglutination inhibition assay, 16 96-well plates were used, with 8 plates used for A-H1N1 real-world comparison and 8 plates used for B-Victoria real-world comparison. Serum samples from 128 individuals were collected; 64 samples were used to determine A-H1N1 antibody titers, and 64 samples were used to determine B-Victoria antibody titers. All experimental procedures except for result interpretation were performed.

[0143] The 16 96-well plates were placed in the device and tested using the algorithm described above. At the same time, three intermediate-level experimental staff and three senior experts took readings. The median value of the valence readings and records of the three senior experts was used as the gold standard. The results of this algorithm and the records of the three intermediate-level experimental staff were compared, as shown in Table 2.

[0144] Table 2. Sample testing and comparison under real-world conditions.

[0145]

[0146] Intermediate-level experimental personnel can represent the average level in the field. The results in Table 2 show that the accuracy of the artificial intelligence algorithm is higher than the average accuracy of intermediate-level experimental personnel, which proves that the device and supporting algorithm of the present invention have good effects, can meet the needs of daily work, and are worth promoting.

[0147] The various embodiments of the present invention have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments.

Claims

1. A method for determining the results of blood coagulation and blood coagulation inhibition experiments, characterized in that, include: Step 1: Place the 96-well plate (5) to be tested for blood coagulation or blood coagulation inhibition in the opaque closed cavity (1) and embed it in the slot on the LED backlight plate (3). The LED backlight plate (3) is tiltable and an industrial camera (6) is provided vertically above the LED backlight plate (3). Step 2: Turn on the LED backlight panel (3) and tilt it to a predetermined angle so that the red blood cells make corresponding displacement at the bottom of the well. After standing for a predetermined time, take a picture of the 96-well plate (5) with the industrial camera (6) to obtain an image containing all the wells. Step 3: Extract multiple features for each hole in the image and input them into the corresponding basic classification model for independent judgment; Step 4: Input the image and the output results of multiple basic classification models into three trained artificial intelligence models. Use the integration rule that at least two of the three artificial intelligence models output non-aggregated results to determine the state of each pore. Based on the status of all wells, obtain the results of blood coagulation or blood coagulation inhibition experiments; In step 3, the first type of feature is extracted for each hole in the image and input into the first basic classification model for judgment, including: Step 31, for each hole, the blue component light intensity gradient and the green component light intensity gradient are obtained on the radial path from the upper edge of the hole to the near center region; Step 32, the blue component light intensity gradient and the green component light intensity gradient are fitted in an exponential decay form to obtain their respective fitting coefficients; Step 33, the judgment result of the first type of feature is determined according to the relationship between the blue component fitting coefficient and the preset blue threshold, and the relationship between the green component fitting coefficient and the preset green threshold. In step 3, the second type of feature is extracted for each pore in the image and input into the second basic classification model for judgment, including: Step 34, within the maximum aggregation radius of red blood cells at the center of each pore, the contour detection algorithm is used to determine whether there is a red deposition area contour; if no contour is detected, the second type of feature is judged as non-aggregated; if a contour is detected, the number of pixels on the long axis and the number of pixels on the short axis of the contour are obtained; Step 35, the ratio of the number of pixels on the long axis to the number of pixels on the short axis is calculated and compared with a preset ratio threshold. If it is greater than the threshold, the second type of feature is judged as aggregated; otherwise, it is judged as non-aggregated. In step 3, the third type of feature is extracted for each hole in the image and input into the third basic classification model for judgment, including: Step 36, drawing a perpendicular line at half the distance from the bottom edge to the center of each hole, the perpendicular line intersecting the edge of the hole to form a horizontal scanning line segment, and collecting blue light intensity sequence and green light intensity sequence point by point along the scanning line segment from the edge to the center; Step 37, counting the number of consecutive points less than the preset low light intensity threshold in the blue light intensity sequence and the green light intensity sequence respectively, based on the consecutive point number threshold n and the preset element number threshold. The determination result of the third type of feature is determined; if a low-intensity and basically continuous region appears in the blue light intensity sequence, the blue component of the third type of feature is determined to be agglomeration; if a low-intensity and basically continuous region appears in the green light intensity sequence, the green component of the third type of feature is determined to be agglomeration.

2. The method for determining the results of blood coagulation and blood coagulation inhibition experiments as described in claim 1, characterized in that, The opaque sealed cavity (1) is equipped with an LED backlight panel (3), a fixed support rod (7), a telescopic support rod (4), and an industrial camera (6); wherein, The LED backlight panel (3) uses a white LED light source and is fixed in the center of the bottom of the cavity. The upper surface of the LED backlight panel (3) is provided with a slot for placing the 96-hole plate (5). The fixed support rod (7) is connected to one side of the bottom of the LED backlight panel (3), and the telescopic support rod (4) is connected to the other side of the bottom of the LED backlight panel (3). It is driven by a stepper motor and is used to tilt the LED backlight panel (3) and the 96-hole plate (5) above it to a predetermined angle. The industrial camera (6) is fixedly installed in an opaque closed cavity (1) and is located directly above the 96-hole plate (5) with the LED backlight plate (3) tilted to a predetermined angle. The industrial camera (6) is connected to an external computer.

3. The method for determining the results of blood coagulation and blood coagulation inhibition experiments as described in claim 1, characterized in that... In step 32, the intensity gradients of the blue component and the green component are fitted using the following exponential decay form to obtain their respective fitting coefficients. : ； in, Indicates color, For color Intensity component in gradient at, Indicates the position of a pixel on the vertical path. The radius of the hole position. This represents the maximum aggregation radius of red blood cells at the center of the pore. If the fitting coefficient of the blue component is greater than the preset blue threshold, the blue component of the first type of feature is determined to be non-aggregated; If the fitting coefficient of the green component is greater than the preset green threshold, the green component of the first type of feature is determined to be non-agglomerated.

4. The method for determining the results of blood coagulation and blood coagulation inhibition experiments as described in claim 1, characterized in that, Step 37 includes: Step 371: Obtain the light intensity sequence S ,include m 1 pixel ;in, Indicates the pixel number. Represents light intensity sequence S The total number of pixels in the middle. , They represent the numbers respectively. The position of the pixel along the horizontal path and the light intensity; Step 372, Traversal Take from 1 ,calculate dimensional vector ; where, linear mapping function The linear mapping function passes through the point and , position The ordinate value on the corresponding straight line Construct into vectors ; Step 373, Traversal Take from 1 , position Corresponding light intensity Establish dimensional vector ; Step 374, Construction dimensional vector The first part of the vector The elements are and The difference, later The zero elements are used to pad the vector; Step 375, Traversal Take from 1 , build 3D matrix ; Step 376: Construct the collection matrix Each column The elements are consecutive Each element is negative. ;in, Threshold for continuous point counting; light intensity sequence S In, each pixel All correspond to vectors ,like There is continuity If any element is negative, then the pixel is... Include in set ,gather Represents light intensity sequence S An area of ​​sudden drop in light intensity appeared; Step 377: Statistical Set If the number of elements in the data exceeds a preset threshold... At this point, the third type of feature is determined to be agglomeration.

5. The method for determining the results of blood coagulation and blood coagulation inhibition experiments as described in claim 1, characterized in that, Step 4 includes: Step 41: Combine the multiple judgment values ​​corresponding to the multiple features with the ordered one-dimensional unfolded values ​​of the RGB channels of the hole position image to form a feature vector; Step 42: Input the feature vectors into the trained artificial intelligence models, including support vector machines, random forests, and logistic regression models, respectively, to obtain their independent prediction results; Step 43: When at least two of the three artificial intelligence models output non-agglomeration results, the hole location is finally determined to be non-agglomeration; otherwise, it is determined to be agglomeration.

6. The method for determining the results of blood coagulation and blood coagulation inhibition experiments as described in claim 5, characterized in that, The three artificial intelligence models, namely Support Vector Machine, Random Forest, and Logistic Regression, are trained using a pre-labeled dataset. The labeled dataset is composed of the consistent results obtained from independent judgments by multiple professionals.

7. A system for determining the results of blood coagulation and blood coagulation inhibition experiments, used to implement the method for determining the results of blood coagulation and blood coagulation inhibition experiments as described in any one of claims 1-6, characterized in that, include: An opaque sealed cavity (1) includes an LED backlight panel (3) and an industrial camera (6); the LED backlight panel (3) is disposed inside the opaque sealed cavity (1), and its upper surface is provided with a slot for placing a 96-hole plate (5), and the LED backlight panel (3) is configured to be tilted in a controllable manner; the industrial camera (6) is fixedly disposed inside the opaque sealed cavity (1), and is located directly above the 96-hole plate (5) when the LED backlight panel (3) is tilted; The control and processing unit (2) is electrically connected to the LED backlight panel (3) and the industrial camera (6), respectively. The control and processing unit (2) includes: The image acquisition module is used to control the LED backlight panel (3) to turn on and tilt to a predetermined angle. After a predetermined time, it controls the industrial camera (6) to take a picture of the 96-hole plate (5) to obtain an image containing all the holes. The basic classification module contains multiple basic classification models, which are used to extract multiple types of features for each hole in the image, make independent judgments, and obtain multiple basic judgment results. The integrated judgment module includes three trained artificial intelligence models. It is used to input the image and the multiple basic judgment results into the three trained artificial intelligence models, and to determine the state of each well by using the integrated rule that at least two of the three artificial intelligence models output non-agglutination results. Based on the state of all wells, the blood coagulation or blood coagulation inhibition experimental results are obtained.

Images