Blood type auxiliary studying and judging method

Abstract

The invention relates to a blood type auxiliary study and judgment method, which comprises the following steps: performing mixing, centrifugation and reaction treatment on a weak agglutination classified blood sample in blood type detection operation, performing precipitation on the sample, and analyzing the light transmittance to obtain the light transmittance variance of the sample and the reactivity and reaction acceleration of the reaction sample. And performing threshold comparison and judgment by combining a light transmission change rate upper limit calibration threshold, a light transmission change acceleration upper limit calibration threshold and a local turbidity gradient variance upper limit calibration threshold under dynamic adjustment of a temperature adjustment factor reflected by the real-time temperature, so as to determine the optimal precipitation time of the sample, effectively control the precipitation time to be too large, and simultaneously ensure the precipitation accuracy of the sample. It is guaranteed that the sample can fully react under the optimal precipitation duration, then the blood detection process is accurately controlled, blood type study and judgment are assisted, and the method has wide application prospects in practical application.

Description

Technical Field

[0001] This invention relates to a method for assisting in blood type determination, belonging to the field of blood type auxiliary detection technology. Background Technology

[0002] In existing blood typing technologies, automated interpretation systems typically analyze results based on pre-defined agglutination characteristic models. The basic principle is to compare the reaction patterns at the reaction wells with a standard template; if the sample reaction morphology conforms to predetermined rules or template characteristics, it is determined to be either agglutinated or non-agglutinated. However, in actual testing, weakly agglutinated samples exhibit complex and diverse forms, often displaying atypical or irregular shapes, making them difficult to accurately identify using established rule models. These samples are highly susceptible to misidentification or missed detection, leading to errors in blood typing or the undetected presence of unexpected antibodies, directly impacting transfusion safety.

[0003] Currently available blood typing equipment generally employs the rule-matching-based interpretation logic described above. However, as clinical demands for testing accuracy continue to rise, its limitations are becoming increasingly apparent. To compensate for the shortcomings of automated interpretation, testing personnel often need to visually review all reaction results to eliminate potential interpretation errors. However, visual interpretation is not only affected by individual subjective experience differences and lacks a unified objective standard, but also prone to visual fatigue during long-term, high-volume review operations, further reducing the accuracy and reliability of the review. Therefore, the existing blood typing interpretation model, which relies on a combination of fixed rules and manual review, is cumbersome, inefficient, and struggles to reliably identify difficult samples such as those with weak agglutination, becoming a key technical bottleneck restricting the improvement of blood typing accuracy and efficiency. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a blood type auxiliary identification method, which uses light transmittance to reflect the sample reaction acceleration rate and combines it with dynamic adjustment of calibration thresholds by changing temperature to determine the optimal precipitation time for weak agglutination classification blood samples, and then constructs a sample, trains to obtain an agglutination precipitation time model, and combines it with blood type detection operation to efficiently assist in blood type identification work.

[0005] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: The present invention designs a blood type auxiliary judgment method, which, based on a preset number of blood samples belonging to the weak agglutination category, obtains an agglutination precipitation time model by following steps A to B, and then performs the following step D for the blood sample to be tested to realize blood type detection;

[0006] Step A. Based on the various spatiotemporal combinations between preset temperature ranges and preset sample volumes, for each blood sample corresponding to the samples after mixing, centrifugation, and reaction processing in the blood typing operation under each spatiotemporal combination, perform precipitation and determine the optimal precipitation time, thereby constructing each sample, forming a sample library, and then proceed to Step B.

[0007] Step B. Based on each sample in the sample library, with the temperature range and the sample size of the blood sample as input, and the preset sedimentation time range corresponding to the optimal sedimentation time in the sample as output, train the target network to obtain the agglutination sedimentation time model.

[0008] Step D. Following the blood typing procedure, the target sample of the blood sample to be tested is obtained after sequentially performing mixing, centrifugation, and reaction processing. According to the agglutination and precipitation time model, precipitation processing is performed on the target sample, followed by centrifugation, to achieve blood typing of the blood sample to be tested.

[0009] As a preferred technical solution of the present invention: In step A, based on each spatiotemporal combination between each preset temperature range and each preset sample volume, for each spatiotemporal combination and each blood sample under the spatiotemporal combination, based on the environment of the temperature range in the spatiotemporal combination, the following steps A1 to A5 are performed to construct the sample.

[0010] Step A1. For the blood sample volume in the spatiotemporal combination, obtain the blood sample after mixing the appropriate dose of blood typing reagent in the test tube blood typing operation, as well as the sample after centrifugation and reaction treatment, and then perform precipitation and initialization on the sample. Then proceed to step A2;

[0011] Step A2. Detect the upper, middle, and lower positions of the sample area in the test tube, corresponding to the... transmittance at any given moment , , And the average value constitutes the sample corresponding to the first Overall light transmittance at any given moment and obtain , , The variances of the three Then proceed to step A3;

[0012] Step A3. Press Obtain the sample corresponding to the first Reaction rate at time 1 , and wait Not less than the preset first time number When, use the following formula:

[0013] ;

[0014] ;

[0015] Obtain the first sample Reaction time The corresponding reaction rate and obtaining the first sample Reaction time The corresponding reaction acceleration rate ,in, This indicates the preset second time step. Indicates the sample corresponding to the first The degree of reaction at any given moment express The duration of time elapsed at each moment Indicates the sample number Reaction time The corresponding reaction rate, express The duration of time elapsed at each moment This indicates the overall transmittance of the sample at time 0, and then proceeds to step A4;

[0016] Step A4. Based on the current ambient temperature ,according to To obtain the temperature adjustment factor And according to the following formula:

[0017] ;

[0018] ;

[0019] ;

[0020] Obtain the upper limit calibration threshold for transmittance change rate. Upper limit calibration threshold for light transmission change acceleration Local turbidity gradient variance upper limit calibration threshold ,in, This indicates the preset upper limit threshold for the rate of change of light transmittance. This indicates the preset upper limit threshold for the acceleration of light transmission change. This indicates the preset upper limit threshold for the local turbidity gradient variance. This indicates the preset temperature compensation coefficient. Indicate the preset reference temperature, then proceed to step A5;

[0021] Step A5. Determine if conditions 1, 2, and 3 are satisfied simultaneously. If yes, obtain the results from time 0 to time 1. The duration of time at each moment , and according to To obtain the optimal sedimentation time Using the temperature range and sample size in the spatiotemporal combination, combined with the optimal sedimentation time Construct samples; among them, Indicates the preset safe retention time. This indicates the preset minimum settling time. This indicates the preset maximum sedimentation time; otherwise, it follows the preset first adjacent time interval, waiting to enter the next time step. Perform an update by incrementing by 1 and return to step A2;

[0022] Condition 1. Based on self- For a predetermined number of consecutive time points from a given moment in the historical time direction, the absolute value of the reaction rate corresponding to the reactivity of the sample at each moment is less than [value missing]. ;

[0023] Condition 2. ;

[0024] Condition 3. .

[0025] As a preferred embodiment of the present invention: In step A5, it is determined whether conditions 1, 2, and 3 are all not satisfied; if so, the process proceeds according to a preset first adjacent time interval, waiting to enter the next time step. Perform an increment update and return to step A2; otherwise, further determine whether conditions 1, 2, and 3 are simultaneously satisfied. If so, obtain the optimal settling time. And construct samples; otherwise, according to the preset second adjacent time interval, wait until entering the next time, for Perform an update by incrementing by 1 and return to step A2, where the preset second adjacent time interval is less than the preset first adjacent time interval.

[0026] As a preferred technical solution of the present invention: in step A5, if it is determined that conditions 1, 2 and 3 are satisfied at the same time, then proceed to step A6 as follows;

[0027] Step A6. Perform centrifugal rotation on the sample under a preset centrifugal force for a preset duration, and ensure that the centrifugal rotation under the preset centrifugal force will not damage the already agglomerated part of the sample, and then proceed to step A7.

[0028] Step A7. Obtain a top view of the sample and determine whether the proportion of the dark red area extending outward from the center is greater than the preset minimum agglomeration area proportion. If so, the optimal sedimentation time is obtained. And construct samples; otherwise, the construction of blood samples under spatiotemporal combination will fail.

[0029] As a preferred technical solution of the present invention: step D includes the following steps D-1 to D-5;

[0030] Step D-1. Following the blood typing procedure, mix the appropriate dose of blood typing reagent with the blood sample to be tested, and perform centrifugation and reaction processing to obtain the corresponding target sample. If the target sample is strongly agglutinated, proceed to step D-5; if the target sample is not strongly agglutinated, proceed to step D-2.

[0031] Step D-2. Based on the temperature range corresponding to the ambient temperature of the target sample and the amount of blood sample to be tested, execute the agglutination and precipitation time model to obtain the corresponding precipitation time range. The upper limit of the precipitation time range constitutes the optimal precipitation time. The optimal precipitation time is used to precipitate the target sample, and then proceed to step D-3.

[0032] Step D-3. Perform centrifugal rotation on the target sample under a preset centrifugal force for a preset duration, and ensure that the centrifugal rotation under the preset centrifugal force does not damage the already agglomerated part of the target sample, and then proceed to step D-4.

[0033] Step D-4. Obtain a top view of the target sample and determine whether there is a change in pixel value from dark to light extending outward from the center position. If yes, the target sample is determined to be weakly aggregated and proceed to step D-5; otherwise, the target sample is determined to be non-aggregated and proceed to step D-5.

[0034] Step D-5. Based on whether the target sample exhibits strong agglutination, weak agglutination, or no agglutination, further determine the blood type of the blood sample to be tested.

[0035] As a preferred embodiment of the present invention, step F is further included as follows:

[0036] Step F. For the target sample determined to be weakly agglutinated in Step D, obtain the blood sample volume of the target sample and the temperature range corresponding to the ambient temperature during Step D. Construct a corresponding spatiotemporal combination and determine whether the same spatiotemporal combination exists in the sample library. If so, execute Step A for the blood sample to be tested, construct the corresponding sample, and add it to the sample library; otherwise, no further processing is performed.

[0037] As a preferred technical solution of the present invention: the target network to be trained in step B includes an initial feature extraction module, a feature compression and attention mechanism module, a multi-scale dilated spatial pyramid pooling module, a multi-scale branching module, a feature pyramid fusion module, a regularization module, an addition fusion module, a global context extraction module, and a classification head module;

[0038] The initial feature extraction module's input is the input of the target network to be trained. The output of the initial feature extraction module is connected to the input of the feature compression and attention mechanism module, and the output of the feature compression and attention mechanism module is connected to the input of the multi-scale dilated spatial pyramid pooling module. The output of the multi-scale dilated spatial pyramid pooling module is connected to the input of the regularization module and the two inputs of the multi-scale branching module. The two outputs of the multi-scale branching module are connected to the two inputs of the feature pyramid fusion module. The outputs of the feature pyramid fusion module and the regularization module are connected to the two inputs of the addition fusion module. The output of the addition fusion module is connected to the input of the global context extraction module. The output of the global context extraction module is connected to the input of the classification head module. The output of the classification head module constitutes the output of the target network to be trained.

[0039] The multi-scale branching module includes a first branching module and a second branching module. The first branching module, from input to output, sequentially includes a first max-pooling layer, a depthwise separable convolutional layer, and a SE attention module. The pooling kernel size of the first max-pooling layer is [value missing]. The first branch module has a stride of 2. Its input to the first max-pooling layer forms the input to the first branch module, and its output to the SE attention module forms the output. The second branch module, from input to output, sequentially includes a second max-pooling layer, a depthwise separable convolutional layer, and an SE attention module. The pooling kernel size of the second max-pooling layer is [value missing]. With a step size of 4, the input of the second max pooling layer constitutes the input of the second branch module, and the output of the SE attention module constitutes the output of the second branch module; the input of the first branch module and the input of the second branch module constitute the two inputs of the multi-scale branch module, and the output of the first branch module and the output of the second branch module constitute the two outputs of the multi-scale branch module.

[0040] The feature pyramid fusion module includes an upsampling module, The convolutional module and the addition fusion module are used. The input of the upsampling module and one input of the addition fusion module constitute the two inputs of the feature pyramid fusion module. The input of the upsampling module is connected to the output of the second branch module in the multi-scale branch module, and one input of the addition fusion module is connected to the output of the first branch module in the multi-scale branch module. The output of the upsampling module is connected to... The input end of the convolution module, The output of the convolution module is connected to the other input of the addition and fusion module, and the output of the addition and fusion module constitutes the output of the feature pyramid fusion module.

[0041] As a preferred technical solution of the present invention: the initial feature extraction module includes, from the input end to the output end, a 2D convolutional layer, a normalization layer, a ReLU activation layer, a depthwise separable convolutional layer, a normalization layer, and a ReLU activation layer in sequence, wherein the input end of the 2D convolutional layer constitutes the input end of the initial feature extraction module, and the output end of the second ReLU activation layer constitutes the output end of the initial feature extraction module.

[0042] The feature compression and attention mechanism module, from input to output, sequentially includes a first bottleneck residual module, an SE attention module, a second bottleneck residual module, and another SE attention module. The input of the first bottleneck residual module constitutes the input of the feature compression and attention mechanism module, and the output of the second SE attention module constitutes the output of the feature compression and attention mechanism module. The structures of the first and second bottleneck residual modules are identical, and both structures, from input to output, sequentially include… Convolutional module Depth-splitable convolutional modules Convolutional module; first in sequence The input of the convolution module forms the bottleneck input of the residual module, which is the second one in sequence. The output of the convolution module becomes the bottleneck output of the residual module;

[0043] The multi-scale void space pyramid pooling module includes a splicing layer, The system consists of a convolutional module, a normalization layer, and four pre-branch paths, with the first pre-branch being connected in series from the input to the output. The convolutional module and the normalization layer, the second, third, and fourth pre-branch paths are connected in series from the input to the output. The convolutional module and normalization layer; the inputs of each front branch are connected to form the input of the multi-scale hollow spatial pyramid pooling module, the outputs of each front branch are connected to the inputs of the stitching layer, and the outputs of the stitching layer are connected in series. The input of the convolutional module is connected to the input of the normalization layer, and the output of the normalization layer constitutes the output of the multi-scale void spatial pyramid pooling module.

[0044] As a preferred technical solution of the present invention: the regularization module includes, from the input end to the output end, a DropBlock layer, a Dropout layer, a normalization layer, and a ReLU activation layer, wherein the input end of the DropBlock layer constitutes the input end of the regularization module, and the output end of the ReLU activation layer constitutes the output end of the regularization module.

[0045] As a preferred technical solution of the present invention: the global context extraction module includes, from the input end to the output end, a global average pooling layer, a flattening operation layer, a linear layer, a ReLU activation layer, and a linear layer in sequence, wherein the input end of the global average pooling layer constitutes the input end of the global context extraction module, and the output end of the linear layer constitutes the output end of the global context extraction module.

[0046] The classification head module includes a fully connected layer and a Softmax layer connected in series from the input end to the output end. The input end of the fully connected layer constitutes the input end of the classification head module, and the output end of the Softmax layer constitutes the output end of the classification head module.

[0047] The blood type auxiliary determination method described in this invention, compared with the prior art, has the following technical advantages:

[0048] This invention presents a blood type auxiliary determination method. For weakly agglutinated blood samples, after mixing, centrifugation, and reaction processing in blood typing, precipitation is performed. Transmittance analysis is used to obtain the transmittance variance, the reactivity of the sample, and the reaction acceleration rate. This is combined with the upper limit calibration thresholds for transmittance change rate, transmittance change acceleration, and local turbidity gradient variance under dynamic temperature adjustment factors, as reflected by real-time temperature. Threshold comparison is then performed to determine the optimal precipitation time for the sample. This effectively controls excessive precipitation time while ensuring sufficient reaction of the sample within the optimal precipitation time, thereby accurately controlling the blood testing process and assisting in blood type determination. This method has broad application prospects in practical applications. Attached Figure Description

[0049] Figure 1 This is a flowchart illustrating step A regarding sample construction in the blood type auxiliary determination method designed in this invention;

[0050] Figure 2 This is a schematic diagram of the overall structure of the target network to be trained in the blood type auxiliary judgment method designed in this invention;

[0051] Figure 3a This is a schematic diagram of the initial feature extraction module and the feature compression and attention mechanism module in the network to be trained in the design target of this invention;

[0052] Figure 3b This is a schematic diagram of the multi-scale hollow spatial pyramid pooling module in the network to be trained, which is the design target of this invention;

[0053] Figure 3c This is a schematic diagram of the multi-scale branch module in the network to be trained, which is the design target of this invention;

[0054] Figure 3dThis is a schematic diagram of the feature pyramid fusion module in the network to be trained, which is the target of this invention;

[0055] Figure 3e This is a schematic diagram of the regularization module in the network to be trained, which is the target of this invention.

[0056] Figure 3f This is a schematic diagram of the global context extraction module and the classification head module in the target network to be trained in this invention;

[0057] Figure 4a This is a schematic diagram illustrating the results of using the method designed in this invention to detect simulated antibodies (anti-A and anti-B);

[0058] Figure 4b This is a schematic diagram showing the results of using the method designed in this invention to detect real weak antibodies (anti-B);

[0059] Figure 4c This is a schematic diagram illustrating the results of using the method designed in this invention to detect a weak antigen (anti-D).

[0060] Figure 5 This is a schematic diagram illustrating the results of using the design method of this invention to detect UAb;

[0061] Figure 6a This is a schematic diagram illustrating the results of using the method designed in this invention to detect samples from three known A2B subtype donors;

[0062] Figure 6b This is a schematic diagram illustrating the results of using the method designed in this invention to detect samples from three known subtype B donors. Detailed Implementation

[0063] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0064] This invention designs a blood type auxiliary identification method. In practical applications, based on a preset number of blood samples belonging to the weak agglutination category, the following steps A to B are performed to obtain an agglutination precipitation time model.

[0065] Step A. Based on the various spatiotemporal combinations between preset temperature ranges and preset sample volumes, for each blood sample corresponding to the samples after mixing, centrifugation, and reaction processing in the blood typing operation under each spatiotemporal combination, perform precipitation and determine the optimal precipitation time, thereby constructing each sample, forming a sample library, and then proceed to Step B.

[0066] In practical applications, the design specifically targets step A. Based on preset temperature ranges and preset sample sizes, various spatiotemporal combinations are implemented. For each spatiotemporal combination and each blood sample within that combination, the design is tailored to the environment of the temperature range within the spatiotemporal combination. Figure 1As shown, perform steps A1 to A5 to construct samples, thereby obtaining individual samples and forming a sample library.

[0067] Step A1. For the blood sample volume in the spatiotemporal combination, obtain the blood sample after mixing the appropriate dose of blood typing reagent in the test tube blood typing operation, as well as the sample after centrifugation and reaction treatment, and then perform precipitation and initialization on the sample. Then proceed to step A2.

[0068] Step A2. Detect the upper, middle, and lower positions of the sample area in the test tube, corresponding to the... transmittance at any given moment , , And the average value constitutes the sample corresponding to the first Overall light transmittance at any given moment and obtain , , The variances of the three Then proceed to step A3. The upper position of the sample area in the test tube refers to the position between the sample liquid level and the middle position in the test tube, and the lower position of the sample area refers to the position between the middle position and the bottom of the sample in the test tube.

[0069] Step A3. Press Obtain the sample corresponding to the first Reaction rate at time 1 , and wait Not less than the preset first time number When, use the following formula:

[0070] ;

[0071] ;

[0072] Obtain the first sample Reaction time The corresponding reaction rate and obtaining the first sample Reaction time The corresponding reaction acceleration rate ,in, This indicates the preset second time step. Indicates the sample corresponding to the first The degree of reaction at any given moment express The duration of time elapsed at each moment Indicates the sample number Reaction time The corresponding reaction rate, express The duration of time elapsed at each moment This indicates the overall transmittance of the sample at time 0, and then proceeds to step A4.

[0073] Step A4. Based on the current ambient temperature ,according to To obtain the temperature adjustment factor And according to the following formula:

[0074] ;

[0075] ;

[0076] ;

[0077] Obtain the upper limit calibration threshold for transmittance change rate. Upper limit calibration threshold for light transmission change acceleration Local turbidity gradient variance upper limit calibration threshold ,in, This indicates the preset upper limit threshold for the rate of change of light transmittance. This indicates the preset upper limit threshold for the acceleration of light transmission change. This indicates the preset upper limit threshold for the local turbidity gradient variance. This indicates the preset temperature compensation coefficient. The preset reference temperature is indicated, and then the process proceeds to step A5.

[0078] Step A5. Determine if conditions 1, 2, and 3 are satisfied simultaneously. If yes, obtain the results from time 0 to time 1. The duration of time at each moment , and according to To obtain the optimal sedimentation time Using the temperature range and sample size in the spatiotemporal combination, combined with the optimal sedimentation time Construct samples, then proceed to step A6; where, Indicates the preset safe retention time. This indicates the preset minimum settling time. This indicates the preset maximum sedimentation time.

[0079] Condition 1. Based on self- For a predetermined number of consecutive time points from a given moment in the historical time direction, the absolute value of the reaction rate corresponding to the reactivity of the sample at each moment is less than [value missing]. ;

[0080] Condition 2. ;

[0081] Condition 3. .

[0082] Corresponding to the scenario where conditions 1, 2, and 3 are simultaneously satisfied, the system continues to determine whether conditions 1, 2, and 3 are all unsatisfied. If so, it proceeds to the next moment according to the preset first adjacent time interval. Perform an increment update and return to step A2; otherwise, further determine whether conditions 1, 2, and 3 are simultaneously satisfied. If so, obtain the optimal settling time. And construct samples; otherwise, according to the preset second adjacent time interval, wait until entering the next time, for Perform an update by incrementing by 1 and return to step A2, where the preset second adjacent time interval is less than the preset first adjacent time interval.

[0083] Step A6. Perform centrifugal rotation on the sample under a preset centrifugal force for a preset duration, ensuring that the centrifugal rotation under the preset centrifugal force does not damage the already agglomerated portion of the sample, and then proceed to step A7.

[0084] Step A7. Obtain a top view of the sample and determine whether the proportion of the dark red area extending outward from the center is greater than the preset minimum agglomeration area proportion. If so, the optimal sedimentation time is obtained. And construct samples; otherwise, the construction of blood samples under spatiotemporal combination will fail.

[0085] Step B. Based on each sample in the sample library, with the temperature range and the sample size of the blood sample as inputs, and the preset sedimentation time range corresponding to the optimal sedimentation time in the sample as output, train the target network to obtain the agglutination sedimentation time model.

[0086] In practical applications, specific designs are made for the target network to be trained, according to... Figure 2 As shown, it includes an initial feature extraction module, a feature compression and attention mechanism module, a multi-scale hollow spatial pyramid pooling module, a multi-scale branching module, a feature pyramid fusion module, a regularization module, an addition fusion module, a global context extraction module, and a classification head module.

[0087] like Figure 2As shown, the input of the initial feature extraction module constitutes the input of the target network to be trained. The output of the initial feature extraction module is connected to the input of the feature compression and attention mechanism module, and the output of the feature compression and attention mechanism module is connected to the input of the multi-scale dilated spatial pyramid pooling module. The output of the multi-scale dilated spatial pyramid pooling module is connected to the input of the regularization module and the two inputs of the multi-scale branching module, respectively. The two outputs of the multi-scale branching module are connected to the two inputs of the feature pyramid fusion module, respectively. The outputs of the feature pyramid fusion module and the regularization module are connected to the two inputs of the addition fusion module, the output of the addition fusion module is connected to the input of the global context extraction module, the output of the global context extraction module is connected to the input of the classification head module, and the output of the classification head module constitutes the output of the target network to be trained.

[0088] The above modules are designed and constructed in specific structural ways for practical applications, such as... Figure 3a As shown, the initial feature extraction module includes, from input to output, a 2D convolutional layer, a normalization layer, a ReLU activation layer, a depthwise separable convolutional layer, a normalization layer, and a ReLU activation layer. The input of the 2D convolutional layer constitutes the input of the initial feature extraction module, and the output of the second ReLU activation layer constitutes the output of the initial feature extraction module.

[0089] like Figure 3a As shown, the feature compression and attention mechanism module, from input to output, includes a first bottleneck residual module, an SE attention module, a second bottleneck residual module, and an SE attention module. The input of the first bottleneck residual module constitutes the input of the feature compression and attention mechanism module, and the output of the second SE attention module constitutes the output of the feature compression and attention mechanism module. The structures of the first and second bottleneck residual modules are identical, and both the first and second bottleneck residual modules, from input to output, sequentially include... Convolutional module Depth-splitable convolutional modules Convolutional module; first in sequence The input of the convolution module forms the bottleneck input of the residual module, which is the second one in sequence. The output of the convolution module constitutes the output of the bottleneck residual module.

[0090] like Figure 3b As shown, the multi-scale void space pyramid pooling module includes a splicing layer, The system consists of a convolutional module, a normalization layer, and four pre-branch paths, with the first pre-branch being connected in series from the input to the output. The convolutional module and the normalization layer, the second, third, and fourth pre-branch paths are connected in series from the input to the output. The convolutional module and normalization layer; the inputs of each front branch are connected to form the input of the multi-scale hollow spatial pyramid pooling module, the outputs of each front branch are connected to the inputs of the stitching layer, and the outputs of the stitching layer are connected in series. The input of the convolutional module is connected to the input of the normalization layer, and the output of the normalization layer constitutes the output of the multi-scale void spatial pyramid pooling module.

[0091] like Figure 3e As shown, the regularization module includes a DropBlock layer, a Dropout layer, a normalization layer, and a ReLU activation layer from input to output. The input of the DropBlock layer constitutes the input of the regularization module, and the output of the ReLU activation layer constitutes the output of the regularization module.

[0092] like Figure 3c As shown, the multi-scale branching module includes a first branching module and a second branching module. The first branching module, from input to output, sequentially includes a first max-pooling layer, a depthwise separable convolutional layer, and a SE attention module. The pooling kernel size of the first max-pooling layer is [value missing]. The first branch module has a stride of 2. Its input to the first max-pooling layer forms the input to the first branch module, and its output to the SE attention module forms the output. The second branch module, from input to output, sequentially includes a second max-pooling layer, a depthwise separable convolutional layer, and an SE attention module. The pooling kernel size of the second max-pooling layer is [value missing]. With a step size of 4, the input of the second max pooling layer constitutes the input of the second branch module, and the output of the SE attention module constitutes the output of the second branch module; the input of the first branch module and the input of the second branch module constitute the two inputs of the multi-scale branch module, and the output of the first branch module and the output of the second branch module constitute the two outputs of the multi-scale branch module.

[0093] like Figure 3d As shown, the feature pyramid fusion module includes an upsampling module, The convolutional module and the addition fusion module are used. The input of the upsampling module and one input of the addition fusion module constitute the two inputs of the feature pyramid fusion module. The input of the upsampling module is connected to the output of the second branch module in the multi-scale branch module, and one input of the addition fusion module is connected to the output of the first branch module in the multi-scale branch module. The output of the upsampling module is connected to... The input end of the convolution module, The output of the convolution module is connected to the other input of the addition and fusion module, and the output of the addition and fusion module constitutes the output of the feature pyramid fusion module.

[0094] like Figure 3f As shown, the global context extraction module includes a global average pooling layer, a flattening operation layer, a linear layer, a ReLU activation layer, and another linear layer from the input to the output. The input of the global average pooling layer constitutes the input of the global context extraction module, and the output of the linear layer constitutes the output of the global context extraction module.

[0095] like Figure 3f As shown, the classification head module consists of a fully connected layer and a Softmax layer connected in series from the input end to the output end. The input end of the fully connected layer constitutes the input end of the classification head module, and the output end of the Softmax layer constitutes the output end of the classification head module.

[0096] Based on steps A to B above, after obtaining the agglutination and precipitation time model, in practical applications, the following step D is performed on the blood sample to be tested to achieve blood type detection.

[0097] Step D. Following the blood typing procedure, the target sample of the blood sample to be tested is obtained after sequentially performing mixing, centrifugation, and reaction processing. According to the agglutination and precipitation time model, precipitation processing is performed on the target sample, followed by centrifugation, to achieve blood typing of the blood sample to be tested.

[0098] In practical applications, the above step D is specifically designed to be executed as follows: steps D-1 to D-5.

[0099] Step D-1. Following the blood typing procedure, mix the appropriate dose of blood typing reagent with the blood sample to be tested, and perform centrifugation and reaction processing to obtain the corresponding target sample. If the target sample is strongly agglutinated, proceed to step D-5; if the target sample is not strongly agglutinated, proceed to step D-2.

[0100] Step D-2. Based on the temperature range corresponding to the ambient temperature of the target sample and the amount of blood sample to be tested, execute the coagulation and precipitation time model to obtain the corresponding precipitation time range. The upper limit of the precipitation time range constitutes the optimal precipitation time. The optimal precipitation time is used to precipitate the target sample, and then proceed to step D-3.

[0101] Step D-3. Perform centrifugal rotation on the target sample under a preset centrifugal force for a preset duration, ensuring that the centrifugal rotation under the preset centrifugal force does not damage the already agglomerated portion of the target sample, and then proceed to step D-4.

[0102] Step D-4. Obtain a top view of the target sample and determine whether there is a change in pixel value from dark to light extending outward from the center position. If yes, the target sample is determined to be weakly agglomerated and proceed to step D-5; otherwise, the target sample is determined to be non-agglomerated and proceed to step D-5.

[0103] Step D-5. Based on whether the target sample exhibits strong agglutination, weak agglutination, or no agglutination, further determine the blood type of the blood sample to be tested.

[0104] After completing the blood type testing of the blood sample to be tested based on the above steps D-1 to D-5, a sample bank update scheme is further designed, that is, the following step F is executed.

[0105] Step F. For the target sample determined to be weakly agglutinated in Step D, obtain the blood sample volume of the target sample and the temperature range corresponding to the ambient temperature during Step D. Construct a corresponding spatiotemporal combination and determine whether the same spatiotemporal combination exists in the sample library. If so, execute Step A for the blood sample to be tested, construct the corresponding sample, and add it to the sample library; otherwise, no further processing is performed.

[0106] The blood type auxiliary determination method of the present invention was applied to the results of detecting weak antibodies and antigens in Example 1, such as... Figure 4a This indicates the results of using the method designed in this invention to detect simulated antibodies (anti-A and anti-B), where 0 times represents undiluted known type O plasma; 2 times represents known type O plasma diluted 2 times with type AB plasma; 4 times represents known type O plasma diluted 4 times with type AB plasma; 8 times represents known type O plasma diluted 8 times with type AB plasma; 16 times represents known type O plasma diluted 16 times with type AB plasma; and 32 times represents known type O plasma diluted 32 times with type AB plasma.

[0107] like Figure 4b This indicates the results of using the method designed in this invention to detect real weak antibodies (anti-B), such as... Figure 4c This indicates the result of the method designed in this invention for detecting a weak antigen (anti-D); wherein, a normal type A donor sample is known, and if the type A donor sample is known to contain a weak anti-B, Anti-A represents anti-A reagent, Anti-B represents anti-B reagent, Cell-A represents A1 phenotype erythrocyte reagent, Cell-B represents B phenotype erythrocyte reagent, Cell-O represents O phenotype erythrocyte reagent, and Anti-D represents anti-D reagent.

[0108] The blood type auxiliary determination method of the present invention was applied to the results of UAb detection in Example 2, such as... Figure 5As shown, A2-UAB represents a known type A UAb donor sample that is UAb positive; O-UAb represents a known type O UAb donor sample that is UAb positive; B-subtype-UAb represents a known type B UAb donor sample that is UAb positive; and the blood type auxiliary determination method of the present invention is further applied to the results of blood subtype detection in Example 3, as shown... Figure 6a This represents three known A2B subtype donor samples, such as... Figure 6b This represents three known B-type donor samples.

[0109] As demonstrated by the above embodiments, the blood typing auxiliary method designed in this invention can quickly classify blood into strong agglutination, weak agglutination, and non-agglutination categories, with significant application effects. For blood samples classified as weakly agglutinated, this method, in conjunction with the blood typing reagent mixing, centrifugation, and reaction processing steps in the test tube blood typing procedure, further performs precipitation. Through transmittance analysis, it obtains the transmittance variance of the sample, as well as the reactivity and reaction acceleration of the sample. Combined with the upper limit calibration thresholds for transmittance change rate, transmittance change acceleration, and local turbidity gradient variance under dynamic temperature adjustment factors reflected by real-time temperature, threshold comparison is performed to determine the optimal precipitation time for the sample. This effectively controls excessive precipitation time while ensuring sufficient reaction of the sample within the optimal precipitation time, thereby accurately controlling the blood testing process and assisting in blood typing. It has broad application prospects in practical applications.

[0110] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A method for blood type auxiliary determination, characterized in that: Based on a preset number of blood samples belonging to the weak agglutination category, the agglutination precipitation time model is obtained by following steps A to B. Then, for the blood sample to be tested, the following step D is performed to achieve blood type detection. Step A. Based on the various spatiotemporal combinations between preset temperature ranges and preset sample volumes, for each blood sample corresponding to the samples after mixing, centrifugation, and reaction processing in the blood typing operation under each spatiotemporal combination, perform precipitation and determine the optimal precipitation time, thereby constructing each sample, forming a sample library, and then proceed to Step B. Step B. Based on each sample in the sample library, with the temperature range and the sample size of the blood sample as input, and the preset sedimentation time range corresponding to the optimal sedimentation time in the sample as output, train the target network to obtain the agglutination sedimentation time model. Step D. Following the blood typing procedure, the target sample of the blood sample to be tested is obtained after sequentially performing mixing, centrifugation, and reaction processing. According to the agglutination and precipitation time model, precipitation processing is performed on the target sample, followed by centrifugation, to achieve blood typing of the blood sample to be tested.

2. The blood type auxiliary determination method according to claim 1, characterized in that: In step A, based on the various spatiotemporal combinations between preset temperature ranges and preset sample volumes, for each spatiotemporal combination and each blood sample under the spatiotemporal combination, based on the environment of the temperature range in the spatiotemporal combination, the following steps A1 to A5 are performed to construct the sample. Step A1. For the blood sample volume in the spatiotemporal combination, obtain the blood sample after mixing the appropriate dose of blood typing reagent in the test tube blood typing operation, as well as the sample after centrifugation and reaction treatment, and then perform precipitation and initialization on the sample. Then proceed to step A2; Step A2. Detect the upper, middle, and lower positions of the sample area in the test tube, corresponding to the... transmittance at any given moment , , And the average value constitutes the sample corresponding to the first Overall light transmittance at any given moment and obtain , , The variances of the three Then proceed to step A3; Step A3. Press Obtain the sample corresponding to the first Reaction rate at time 1 , and wait Not less than the preset first time number When, use the following formula: ； ； Obtain the first sample Reaction time The corresponding reaction rate and obtaining the first sample Reaction time The corresponding reaction acceleration rate ,in, This indicates the preset second time step. Indicates the sample corresponding to the first The degree of reaction at any given moment express The duration of time elapsed at each moment Indicates the sample number Reaction time The corresponding reaction rate, express The duration of time elapsed at each moment This indicates the overall transmittance of the sample at time 0, and then proceeds to step A4; Step A4. Based on the current ambient temperature ,according to To obtain the temperature adjustment factor And according to the following formula: ； ； ； Obtain the upper limit calibration threshold for the rate of change of transmittance. Upper limit calibration threshold for light transmission change acceleration Local turbidity gradient variance upper limit calibration threshold ,in, This indicates the preset upper limit threshold for the rate of change of light transmittance. This indicates the preset upper limit threshold for the acceleration of light transmission change. This indicates the preset upper limit threshold for the local turbidity gradient variance. This indicates the preset temperature compensation coefficient. Indicate the preset reference temperature, then proceed to step A5; Step A5. Determine if conditions 1, 2, and 3 are satisfied simultaneously. If yes, obtain the results from time 0 to time 1. The duration of time at each moment , and according to To obtain the optimal sedimentation time Using the temperature range and sample size in the spatiotemporal combination, combined with the optimal sedimentation time Construct samples; among them, Indicates the preset safe retention time. This indicates the preset minimum settling time. This indicates the preset maximum sedimentation time; otherwise, it follows the preset first adjacent time interval, waiting to enter the next time step. Perform an update by incrementing by 1 and return to step A2; Condition 1. Based on self- For a predetermined number of consecutive time points from a given moment in the historical time direction, the absolute value of the reaction rate corresponding to the reactivity of the sample at each moment is less than [a certain value]. ; Condition 2. ; Condition 3. .

3. The blood type auxiliary determination method according to claim 2, characterized in that: In step A5, it is determined whether conditions 1, 2, and 3 are all not met. If so, the process proceeds to the next time step according to the preset first adjacent time interval. Perform an increment update and return to step A2; otherwise, further determine whether conditions 1, 2, and 3 are simultaneously satisfied. If so, obtain the optimal settling time. And construct samples; otherwise, according to the preset second adjacent time interval, wait until entering the next time, for Perform an update by incrementing by 1 and return to step A2, where the preset second adjacent time interval is less than the preset first adjacent time interval.

4. The blood type auxiliary determination method according to claim 2, characterized in that: In step A5, if it is determined that conditions 1, 2, and 3 are satisfied simultaneously, then proceed to step A6 as follows; Step A6. Perform centrifugal rotation on the sample under a preset centrifugal force for a preset duration, and ensure that the centrifugal rotation under the preset centrifugal force will not damage the already agglomerated part of the sample, and then proceed to step A7. Step A7. Obtain a top view of the sample and determine whether the proportion of the dark red area extending outward from the center is greater than the preset minimum agglomeration area proportion. If so, the optimal sedimentation time is obtained. And construct samples; otherwise, the construction of blood samples under spatiotemporal combination will fail.

5. The blood type auxiliary determination method according to claim 1, characterized in that: Step D includes the following steps D-1 to D-5; Step D-1. Following the blood typing procedure, mix the appropriate dose of blood typing reagent with the blood sample to be tested, and perform centrifugation and reaction processing to obtain the corresponding target sample. If the target sample is strongly agglutinated, proceed to step D-5; if the target sample is not strongly agglutinated, proceed to step D-2. Step D-2. Based on the temperature range corresponding to the ambient temperature of the target sample and the amount of blood sample to be tested, execute the agglutination and precipitation time model to obtain the corresponding precipitation time range. The upper limit of the precipitation time range constitutes the optimal precipitation time. The optimal precipitation time is used to precipitate the target sample, and then proceed to step D-3. Step D-3. Perform centrifugal rotation on the target sample under a preset centrifugal force for a preset duration, and ensure that the centrifugal rotation under the preset centrifugal force does not damage the already agglomerated part of the target sample, and then proceed to step D-4. Step D-4. Obtain a top view of the target sample and determine whether there is a change in pixel value from dark to light extending outward from the center position. If yes, the target sample is determined to be weakly aggregated and proceed to step D-5; otherwise, the target sample is determined to be non-aggregated and proceed to step D-5. Step D-5. Based on whether the target sample exhibits strong agglutination, weak agglutination, or no agglutination, further determine the blood type of the blood sample to be tested.

6. The blood type auxiliary determination method according to claim 1, characterized in that: It also includes step F as follows, Step F. For the target sample determined to be weakly agglutinated in Step D, obtain the blood sample volume of the target sample and the temperature range corresponding to the ambient temperature during Step D. Construct a corresponding spatiotemporal combination and determine whether the same spatiotemporal combination exists in the sample library. If so, execute Step A for the blood sample to be tested, construct the corresponding sample, and add it to the sample library; otherwise, no further processing is performed.

7. The blood type auxiliary determination method according to claim 1, characterized in that: The target network to be trained in step B includes an initial feature extraction module, a feature compression and attention mechanism module, a multi-scale dilated spatial pyramid pooling module, a multi-scale branching module, a feature pyramid fusion module, a regularization module, an additive fusion module, a global context extraction module, and a classification head module. The initial feature extraction module's input is the input of the target network to be trained. The output of the initial feature extraction module is connected to the input of the feature compression and attention mechanism module, and the output of the feature compression and attention mechanism module is connected to the input of the multi-scale dilated spatial pyramid pooling module. The output of the multi-scale dilated spatial pyramid pooling module is connected to the input of the regularization module and the two inputs of the multi-scale branching module. The two outputs of the multi-scale branching module are connected to the two inputs of the feature pyramid fusion module. The outputs of the feature pyramid fusion module and the regularization module are connected to the two inputs of the addition fusion module. The output of the addition fusion module is connected to the input of the global context extraction module. The output of the global context extraction module is connected to the input of the classification head module. The output of the classification head module constitutes the output of the target network to be trained. The multi-scale branching module includes a first branching module and a second branching module. The first branching module, from input to output, sequentially includes a first max-pooling layer, a depthwise separable convolutional layer, and a SE attention module. The pooling kernel size of the first max-pooling layer is [value missing]. The first branch module has a stride of 2. Its input to the first max-pooling layer forms the input to the first branch module, and its output to the SE attention module forms the output. The second branch module, from input to output, sequentially includes a second max-pooling layer, a depthwise separable convolutional layer, and an SE attention module. The pooling kernel size of the second max-pooling layer is [value missing]. With a step size of 4, the input of the second max pooling layer constitutes the input of the second branch module, and the output of the SE attention module constitutes the output of the second branch module; the input of the first branch module and the input of the second branch module constitute the two inputs of the multi-scale branch module, and the output of the first branch module and the output of the second branch module constitute the two outputs of the multi-scale branch module. The feature pyramid fusion module includes an upsampling module, The convolutional module and the addition fusion module are used. The input of the upsampling module and one input of the addition fusion module constitute the two inputs of the feature pyramid fusion module. The input of the upsampling module is connected to the output of the second branch module in the multi-scale branch module, and one input of the addition fusion module is connected to the output of the first branch module in the multi-scale branch module. The output of the upsampling module is connected to... The input end of the convolution module, The output of the convolution module is connected to the other input of the addition and fusion module, and the output of the addition and fusion module constitutes the output of the feature pyramid fusion module.

8. The blood type auxiliary determination method according to claim 7, characterized in that: The initial feature extraction module includes, from input to output, a 2D convolutional layer, a normalization layer, a ReLU activation layer, a depthwise separable convolutional layer, a normalization layer, and a ReLU activation layer. The input of the 2D convolutional layer constitutes the input of the initial feature extraction module, and the output of the second ReLU activation layer constitutes the output of the initial feature extraction module. The feature compression and attention mechanism module, from input to output, sequentially includes a first bottleneck residual module, an SE attention module, a second bottleneck residual module, and another SE attention module. The input of the first bottleneck residual module constitutes the input of the feature compression and attention mechanism module, and the output of the second SE attention module constitutes the output of the feature compression and attention mechanism module. The structures of the first and second bottleneck residual modules are identical, and both structures, from input to output, sequentially include… Convolutional module Depth-splitable convolutional modules Convolutional module; first in sequence The input of the convolution module forms the bottleneck input of the residual module, which is the second one in sequence. The output of the convolution module becomes the bottleneck output of the residual module; The multi-scale void space pyramid pooling module includes a splicing layer, The system consists of a convolutional module, a normalization layer, and four pre-branch paths, with the first pre-branch being connected in series from the input to the output. The convolutional module and the normalization layer, the second, third, and fourth pre-branch paths are connected in series from the input to the output. The convolutional module and normalization layer; the inputs of each front branch are connected to form the input of the multi-scale hollow spatial pyramid pooling module, the outputs of each front branch are connected to the inputs of the stitching layer, and the outputs of the stitching layer are connected in series. The input of the convolutional module is connected to the input of the normalization layer, and the output of the normalization layer constitutes the output of the multi-scale void spatial pyramid pooling module.

9. The blood type auxiliary determination method according to claim 7, characterized in that: The regularization module includes, from input to output, a DropBlock layer, a Dropout layer, a normalization layer, and a ReLU activation layer. The input of the DropBlock layer constitutes the input of the regularization module, and the output of the ReLU activation layer constitutes the output of the regularization module.

10. The blood type auxiliary determination method according to claim 7, characterized in that: The global context extraction module includes, from input to output, a global average pooling layer, a flattening operation layer, a linear layer, a ReLU activation layer, and another linear layer. The input of the global average pooling layer constitutes the input of the global context extraction module, and the output of the linear layer constitutes the output of the global context extraction module. The classification head module includes a fully connected layer and a Softmax layer connected in series from the input end to the output end. The input end of the fully connected layer constitutes the input end of the classification head module, and the output end of the Softmax layer constitutes the output end of the classification head module.

Images