A method and system for reviewing and homologous matching of blood routine test results

Abstract

The present invention relates to the field of medical technology, and in particular to a method and system for reviewing and homologous matching of blood routine test results, the method comprising: obtaining blood routine test data of a target population to form a training data set and a test data set; constructing an anomaly detection model, formulating evaluation rules for applicable scenarios, re-examination decisions and homologous predictions; using the training data set to train the anomaly detection model; during testing, in the sample quality review scenario, the input of the anomaly detection model is the difference in indicators of each item between the current subject's blood routine sample and the previous blood routine sample of the subject and the data of related interference factors, and the output is whether re-examination is required; in the homologous matching scenario, the input of the anomaly detection model is the difference in indicators of each item between the current subject's blood routine sample and the blood routine sample of a suspected homologous patient and the data of related interference factors, and the output is whether it matches. The present invention can quickly and accurately determine whether a patient sample is qualified, and perform homologous matching for mismatched samples.

Description

Technical Field

[0001] The present invention relates to the field of medical technology, and more particularly to a method and system for reviewing and homologous matching of blood routine test results. Background Art

[0002] Routine blood tests are one of the most widely used tests in clinical medicine. Errors in test results before laboratory analysis may lead to clinical decision errors, which in turn may cause medical accidents. Common pre-analytical errors include patient identification errors, sample identification errors, blood collection errors, sample contamination, incomplete samples, and the presence of interfering substances in samples. Among them, sample confusion caused by various factors is a serious pre-analytical error.

[0003] In order to ensure the quality of routine blood tests, it is necessary to control the test results to eliminate the damage caused by instruments, reagents, operations, samples, etc. Among them, delta check is a common method that can judge the difference in specific items of two or more consecutive test results of the same patient, but it has certain limitations. In addition, the existing technology cannot trace back and match mismatched samples, that is, it cannot accurately determine whether two or more routine blood test data are from the same patient. Summary of the invention

[0004] In view of this, the present invention provides a method and system for reviewing and homologous matching of blood routine test results, which can automatically, quickly and accurately determine whether patient samples are qualified, and perform homologous matching on mismatched samples, thereby improving the efficiency and quality of medical services.

[0005] In order to achieve the above object, the present invention adopts the following technical solution:

[0006] In a first aspect, the present invention provides a method for reviewing and homologous matching of blood routine test results, comprising the following steps:

[0007] Obtain the blood routine test data of the target population, form a training data set and a test data set, and perform preprocessing;

[0008] A fully connected neural network was used to build an anomaly detection model that can extract features from two blood routine data and make anomaly judgments;

[0009] Formulate evaluation rules for applicable scenarios, re-examination decisions, and homology predictions;

[0010] During training, for sample quality review scenarios and homology matching scenarios, under the guidance of evaluation rules for re-inspection decision and homology prediction, the anomaly detection model is trained using the training data set;

[0011] During the test, in the sample quality review scenario, the input of the anomaly detection model is the difference between the current subject's blood routine sample and the previous blood routine sample and the relevant interference factor data, and the output is whether re-examination is needed;

[0012] In the homologous matching scenario, the input of the anomaly detection model is the difference in indicators of each item and related interference factor data between the current subject's blood routine sample and the suspected homologous patient's blood routine sample, and the output is whether it matches.

[0013] Furthermore, the training set generation process includes:

[0014] The routine blood test data of the subjects who were tested earlier than the set date and were over 10 years old were obtained as training samples. Each subject had at least 2 routine blood test data, and the positive data of the routine blood test data were obtained by pairing the routine blood test data of the same subject, while the negative data were obtained by pairing the routine blood test data of different subjects at the first visit. The corresponding items were subtracted to generate the final training data set.

[0015] Furthermore, the test set generation process includes:

[0016] The routine blood data of the subjects who were tested after the set date and were over 10 years old were obtained as test samples, and the last routine blood data of each subject was selected from the test samples to form data set I; the second to last routine blood data of each subject and the last routine blood data of other subjects were selected from the test samples to form data set II; the data in data set I and data set II were paired, and the corresponding items were subtracted to generate the final test data set.

[0017] Further, preprocessing the training data set and the test data set includes:

[0018] Remove samples whose values ​​are too large or too small from the normal range;

[0019] The data were standardized to make them conform to normal distribution.

[0020] Furthermore, when the anomaly detection model is trained and tested, its input parameters include: the difference in test indicators of each item of two routine blood tests, the number of the routine blood test instrument, whether the instrument is consistent, the time difference of the routine blood test, the age and gender of the subjects, whether the gender of the paired data is consistent, and the age difference of the paired data.

[0021] Furthermore, the analysis process of the anomaly detection model on the training data set includes:

[0022] The training data set consists of n samples, expressed as: , where each sample contains m features;

[0023] Extract features from the input layer vector x and map it to a low-dimensional hidden layer vector h using a fully connected neural network structure;

[0024] The feature extraction layer vector h is judged for abnormality and mapped to a scalar p using a fully connected layer and an activation function. p represents the probability that two blood routine data are from the same person or that the values ​​are reasonable. The closer it is to 1, the more likely it is a normal sample, and the closer it is to 0, the more likely it is an abnormal sample.

[0025] The final result is output according to the abnormal judgment layer scalar p. If p ≥ ℇ, it is judged as normal, otherwise it is judged as abnormal, where ℇ is the threshold.

[0026] Furthermore, the applicable scenarios of the anomaly detection model are:

[0027] The examinee must have undergone routine blood tests twice or more, and the interval between the last one is within 1 year;

[0028] The blood routine data of the subjects must not contain outliers. Outliers are those that are greater than the upper limit of the reference range and the interval between them is greater than 3 times the length of the reference range.

[0029] The examinees must be between 10 and 60 years old.

[0030] Furthermore, the evaluation rules for homology prediction include:

[0031] For each subject i, it is paired with a candidate and marked according to whether they are the same person to form a label , where j = 1, 2, ..., a, let the predicted probability of the homologous sample pair of the subject be:

[0032]

[0033] Will As the threshold, count the number of false positives in all sample pairs of the subject ,Right now:

[0034]

[0035] Among them, count represents the number of sample pairs that meet the conditions, and the number of false positives reflects the number of wrong candidates that the model needs to exclude before reaching the threshold;

[0036] according to The maximum value of sets the interval range b of the number of false alarms;

[0037] For all the subjects, the false positive number analysis is performed. Assuming the total number of subjects is N, the proportion of subjects whose false positive number is in the interval (0, b / a] is:

[0038]

[0039] Where I is an indicator function, which is 1 when the condition is met and 0 otherwise;

[0040] The b / a corresponding to r being 80% is defined as the proportion of suspected candidates that can be screened out by the model when applied, and this is used to evaluate the homology matching performance.

[0041] Furthermore, the re-inspection decision is: when the probability p is greater than 0.5, re-inspection is required.

[0042] In a second aspect, the present invention provides a blood routine test result review and homology matching system, comprising:

[0043] An acquisition module is used to obtain two blood routine data;

[0044] The analysis module is used to analyze the index difference and related interference factor data of two routine blood test data using the abnormality detection model generated by the above method, and output the results of whether re-examination is needed and whether the two routine blood test data are of the same source.

[0045] It can be seen from the above technical solutions that, compared with the prior art, the present invention has the following beneficial effects:

[0046] The use of the present invention can be divided into two scenarios: result review: to determine whether multiple blood tests of the same patient need to be retested; homology matching: to determine whether different samples come from the same patient, and to assist in finding the source range to assist in the pairing correction of samples. The present invention makes up for the limitations of the traditional delta check method such as low accuracy, single function, and inability to perform homology analysis by comprehensively analyzing the differences in all dimensions of blood tests. The method is simple and easy to implement, and does not require additional instruments or reagents.

[0047] The present invention comprehensively considers multiple key factors, such as age, gender, instrument, and test time, making the comparative review process of blood routine data more accurate and efficient. It can be widely used in the fields of medical fraud detection, replacement test problem solving, treatment effect observation, and medical service optimization. Through automated homology analysis, the accuracy and consistency of medical data can be guaranteed, the quality of medical care can be improved, the workload of doctors can be reduced, the patient's medical experience can be improved, and the development of the medical field can be promoted, benefiting society and patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0049] Figure 1 A flow chart of the method for reviewing blood routine test results and homologous matching provided by the present invention;

[0050] Figure 2 A schematic diagram of applicable scenarios of the anomaly detection model provided by the present invention;

[0051] Figure 3 The experimental flow chart provided by the present invention;

[0052] Figure 4 A schematic diagram of the model effectiveness evaluation results of the first set of experiments provided by the present invention;

[0053] Figure 5 Schematic diagram of model performance evaluation for the second set of experiments provided by the present invention. DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] like Figure 1 As shown, the embodiment of the present invention discloses a method for reviewing and homologous matching of blood routine test results, comprising the following steps:

[0056] S1. Obtain the routine blood test data of the target population, form a training data set and a test data set, and perform preprocessing.

[0057] S2. A fully connected neural network is used to build an anomaly detection model that can extract features from two blood routine data and perform anomaly judgment.

[0058] S3. Develop evaluation rules for applicable scenarios, re-examination decisions, and homology predictions;

[0059] During training, for sample quality review scenarios and homology matching scenarios, the anomaly detection model is trained using the training data set under the guidance of the evaluation rules for re-inspection decision and homology prediction;

[0060] S4. During the test, in the sample quality review scenario, the input of the anomaly detection model is the difference between the current subject's blood routine sample and the previous blood routine sample of the subject and the relevant interference factor data, and the output is whether re-examination is needed;

[0061] In the homologous matching scenario, the input of the anomaly detection model is the difference in indicators of each item between the current subject's blood routine sample and the blood routine sample of the suspected homologous patient and the data of related interference factors, and the output is whether there is a match.

[0062] The present invention constructs an anomaly detection model that can be used in two scenarios. When modeling, two samples of correct pairing and incorrect pairing are created, and the information is the difference of indicators of each item and related interference factor data, so that the model can distinguish normal sample pairs from abnormal sample pairs. At the application level, people choose the scene independently. If it is a re-examination scene, the difference between two consecutive samples of the person is input; if it is a homologous detection scene, the difference between the sample of the subject and the candidate is input.

[0063] The above steps are further described below.

[0064] S1. Dataset construction:

[0065] 1) The training set generation process includes:

[0066] The routine blood test data of the subjects who were tested earlier than the set date and were over 10 years old were obtained as training samples. Each subject had at least 2 routine blood test data, and the positive data of the routine blood test data were obtained by pairing the routine blood test data of the same subject, while the negative data were obtained by pairing the routine blood test data of different subjects at the first visit. The corresponding items were subtracted to generate the final training data set.

[0067] Positive data refers to normal samples, which are the difference between two consecutive samples of the same subject. Negative data refers to abnormal samples, which are the difference between two samples from different people. Positive and negative data are matched to form a set of data, that is, a positive data contains two blood routine data of the same person, and a negative data contains two data of different people. Therefore, the difference between the two blood routine data in this matched data is calculated to form the final training data set.

[0068] 2) The test set generation process includes:

[0069] The routine blood data of the subjects who were tested after the set date and were over 10 years old were obtained as test samples, and the last routine blood data of each subject was selected from the test samples to form data set I; the second to last routine blood data of each subject and the last routine blood data of other subjects were selected from the test samples to form data set II; the data in data set I and data set II were paired, and the corresponding items were subtracted to generate the final test data set.

[0070] 3) Data preprocessing:

[0071] Preprocessing of training and testing datasets includes:

[0072] ① Remove samples whose values ​​are too large or too small compared to the normal range due to measurement errors, equipment failures, human interference, etc.; too large can be defined as an absurd value that exceeds the reference range by 3 times, and too small can be defined as a non-positive number.

[0073] ② Standardize the data to make it conform to the normal distribution. Standardization means converting the data into a normal distribution with a mean of 0 and a variance of 1, thereby eliminating the dimension and scale differences of the data and making the data more standardized and unified.

[0074] S2. Model construction.

[0075] The anomaly detection model of the present invention is a binary classification model, which consists of a multi-dimensional input layer, a multi-layer fully connected layer, and a 1-dimensional output layer. The neural network is compiled using the Adam optimizer and the binary_crossentropy loss function, and the epochs is set to 100 and the batch_size is set to 50000 for model training.

[0076] S3, model training and rule setting.

[0077] 1) When training and testing the anomaly detection model, its input parameters include: the difference in test indicators of each item of two blood routine tests, the number of the blood routine test instrument, whether the instrument is consistent, the time difference of the blood routine test, the age and gender of the subjects, whether the gender of the paired data is consistent, and the age difference of the paired data.

[0078] The anomaly detection model analyzes the training data set in the following steps:

[0079] The training data set consists of n samples, expressed as: , where each sample contains m features;

[0080] The input layer vector x is feature extracted and mapped to a low-dimensional hidden layer vector h using a fully connected neural network structure, namely:

[0081] ;

[0082] in, and They represent the weight matrix and bias vector of the i-th hidden layer respectively, σ represents the activation function, and n represents the number of hidden layers.

[0083] The feature extraction layer vector h is judged for abnormality and mapped to a scalar p using a fully connected layer and an activation function. p represents the probability that two blood routine data are from the same person or that the value is reasonable. The closer it is to 1, the more likely it is a normal sample, and the closer it is to 0, the more likely it is an abnormal sample. When judging whether a person's sample needs to be retested, the sample is input into the model as a difference between the sample and the person's previous sample, and the output probability p of the model is observed. If it is greater than 0.5, it is considered unqualified. When looking for the source of a sample, the user inputs the sample into the model as a difference between the sample and the samples of all candidates in turn, and counts the distance between the output probability and 1.

[0084] The final result is output according to the abnormal judgment layer scalar p. If p ≥ ℇ, it is judged as normal, otherwise it is judged as abnormal, where ℇ is the threshold.

[0085] 2) The applicable scenarios of anomaly detection models are:

[0086] The examinee must have undergone routine blood tests twice or more, and the interval between the last one is within 1 year;

[0087] The blood routine data of the subjects must not contain outliers. Outliers are those that are greater than the upper limit of the reference range and the interval between them is greater than 3 times the length of the reference range.

[0088] The examinee must be between 10 and 60 years old;

[0089] The analysis needs of the subjects are consistent with the model's functional constraint scenarios.

[0090] 2) Review decision:

[0091] On the premise that the examinee has input the difference in indicators of each item between the routine blood sample and his or her previous routine blood sample and the data of related interference factors into the anomaly detection model, when the output probability value exceeds 0.5, re-examination is recommended.

[0092] 3) The evaluation rules for homology prediction include:

[0093] For each subject i, it is paired with a candidate and marked according to whether they are the same person to form a label , where j = 1, 2, ..., a, let the predicted probability of the homologous sample pair of the subject be:

[0094]

[0095] There are two types of data sets: normal samples, labeled 1; abnormal samples, labeled 0; here y=1 refers to the label of the normal sample, i and j represent the serial numbers of the subject and the candidate. The current goal is to design a homology judgment rule, to give a probability statistics to a judgment threshold that can distinguish whether it is homology. For a certain subject i, there are multiple probabilities when he matches with multiple candidates. First find the probability p(y i,j =1), and name it p i .

[0096] Will As the threshold, count the number of false positives in all sample pairs of the subject ,Right now:

[0097]

[0098] Among them, count represents the number of sample pairs that meet the conditions, and the number of false positives reflects the number of wrong candidates that the model needs to exclude before reaching the threshold; Indicates the label of the abnormal sample (ie, y i，j =0), for a subject i, the probability set when he matches others is recorded as p(y i,j ≠1), we need to count these probabilities ≥p i The number of .

[0099] according to The maximum value of the false alarm number is set as an interval b, and the maximum value does not exceed a, such as 20, 40, 60, 80, 100, 150, 200...

[0100] For all the subjects, the false positive number analysis is performed. Assuming the total number of subjects is N, the proportion of subjects whose false positive number is in the interval (0, b / a] is:

[0101]

[0102] Where I is an indicator function, which is 1 when the condition is met and 0 otherwise;

[0103] The b / a corresponding to r being 80% is defined as the proportion of suspected candidates that can be screened out by the model when applied, and this is used to evaluate the homology matching performance.

[0104] S4. Model testing.

[0105] like Figure 2As shown, during the test, in the sample quality review scenario, the input of the anomaly detection model is the difference in indicators of each item between the current subject's blood routine sample and the previous blood routine sample of the subject, as well as the data of related interference factors (such as the number of the blood routine test instrument, whether the instrument is consistent, the time difference of the blood routine test, the age and gender of the subject, whether the gender of the paired data is consistent, and the age difference of the paired data, etc.), and the output is whether re-testing is needed;

[0106] In the homologous matching scenario, the input of the anomaly detection model is the difference in indicators of each item between the current subject's blood routine sample and the blood routine sample of the suspected homologous patient, as well as the data of related interference factors (such as the blood routine test instrument number, whether the instruments are consistent, the time difference of the blood routine test, the age and gender of the subject, whether the gender of the paired data is consistent, and the age difference of the paired data, etc.), and the output is whether there is a match.

[0107] In one embodiment, the present invention further provides a blood routine test result review and homology matching system, comprising:

[0108] An acquisition module is used to obtain two blood routine data;

[0109] The analysis module is used to analyze the index difference and related interference factor data of two routine blood test data using the abnormality detection model generated by the above method, and output the results of whether re-examination is needed and whether the two routine blood test data are of the same source.

[0110] Next, to further verify the effectiveness of the method of the present invention, two sets of experiments were conducted, including:

[0111] 1. Extract all blood routine test data from a hospital and pair them to form homologous matching sample pairs. The data set includes a positive experimental group and a negative control group, which are paired with blood routine test data of the same subject and different subjects respectively. The specific construction rules are as follows:

[0112] 1a) Training data construction plan: People aged 10 years and above were selected as training data. Each person had at least 2 blood routine data, and the blood routine data were tested before September 1, 2022. Positive data were obtained by pairing the blood routine data of the same person, and negative data were obtained by pairing the blood routine data of different people at the first visit. The total number of positive data was 42,582, and the number of negative data was 9,520,066.

[0113] 1b) Test data construction plan: Extract blood routine data from September 1, 2022 to December 1, 2022. The specific construction process includes the following steps: First, select the last blood routine data of each patient from these data to form data set 1, which contains a total of 135 data; then, select the second to last blood routine data of each patient from these data, and the last blood routine data of other patients to form data set 2, which contains a total of 4448 data; finally, pair the data in data set 1 and data set 2 to obtain the final test set.

[0114] 1c) Preprocess the data to make it more suitable for subsequent model training and prediction, and improve the accuracy and efficiency of the model. Preprocessing includes the following:

[0115] Remove outliers: Delete data points that are significantly different from the normal range or inconsistent with other samples to prevent them from affecting the learning effect of the model.

[0116] Standardization: The Z-score standardization method is used to standardize the data. The mean of the feature is subtracted from each data point, and then divided by the standard deviation of the feature to convert it into a normal distribution with a mean of 0 and a standard deviation of 1, eliminating the dimensional differences between different features.

[0117] 2. Training the blood routine test data homology prediction model. This embodiment will face the situation of whether the age and gender information of the examinee is correct or not. Therefore, two groups of experiments are designed to deal with the blood routine test data homology matching under different circumstances. Figure 3 As shown, the specific process is as follows:

[0118] 2a) The first set of experiments: This experiment assumes that the age and gender information of the subjects is incorrect, that is, a certain proportion of the subjects provide incorrect or forged age and gender information. The difference in blood routine test indicators between the subject and the candidate, the blood routine test instrument, whether the blood routine test instrument is consistent, and the test time difference are used as input parameters and input into the blood routine homology prediction model to obtain the prediction result of whether the subject and the candidate are the same person.

[0119] In the experiment, a binary classification model was built using a fully connected neural network. The model consists of a 25-dimensional input layer, 5 fully connected layers, and a 1-dimensional output layer. The number of neurons in the 5 fully connected layers is 448, 704, 384, 128, and 128. The Adam optimizer and binary_crossentropy loss function were used to compile the neural network, and the epochs were set to 100 and the batch_size was set to 50,000 for model training.

[0120] 2b) The second group of experiments: This experiment assumes that the age and gender information of the examinee is completely correct, that is, there is no information error or falsification. The examinee's gender, age, time difference between the examinee and the candidate's blood routine test, blood routine test instrument, whether the blood routine test instrument is consistent, whether the gender is consistent, age difference and test index difference are used as input parameters and input into the blood routine homology prediction model to obtain the prediction result of whether the examinee and the candidate are the same person.

[0121] The experimental process also uses a fully connected neural network to build a binary classification model. The model consists of a 29-dimensional input layer, 4 fully connected layers, and a 1-dimensional output layer. The number of neurons in the 4 fully connected layers is 1024, 256, 704, and 128. The Adam optimizer and binary_crossentropy loss function are used to compile the neural network, and the epochs are set to 100 and the batch_size is set to 50000 for model training.

[0122] 3. Model performance evaluation. Using the preprocessed test data, the blood routine test data homology prediction model was used to analyze the probability that two blood routine test data belong to the same person, and the model performance was evaluated based on the evaluation rules. The model performance evaluation results of the two groups of experiments are shown in Figure 4 and Figure 5 .

[0123] The experimental results show that by introducing factors such as age and gender, 108 homologous candidates were successfully screened out from 10,000 candidates. This shows that the consideration of these factors helps to improve the screening accuracy of the model. Without taking factors such as age and gender into consideration, only using blood routine data, 1,124 homologous candidates were screened out, increasing the number of candidates. In summary, adding factors such as age and gender can improve the accuracy of homologous matching of blood routine data and effectively screen out potential homologous candidates. In practical applications, comprehensive consideration of multiple relevant factors can improve the accuracy and reliability of homologous matching.

[0124] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0125] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for reviewing and homologous matching of blood routine test results, characterized in that: The following steps are involved: Obtain the blood routine test data of the target population, form a training data set and a test data set, and perform preprocessing; A fully connected neural network was used to build an anomaly detection model that can extract features from two blood routine data and make anomaly judgments; Formulate evaluation rules for applicable scenarios, re-examination decisions, and homology predictions; During training, for sample quality review scenarios and homology matching scenarios, under the guidance of evaluation rules for re-inspection decision and homology prediction, the anomaly detection model is trained using the training data set; During the test, in the sample quality review scenario, the input of the anomaly detection model is the difference between the current subject's blood routine sample and the previous blood routine sample and the relevant interference factor data, and the output is whether re-examination is needed; In the homology matching scenario, the input of the anomaly detection model is the difference in indicators of each item and related interference factor data between the current subject's blood routine sample and the suspected homology patient's blood routine sample, and the output is whether it matches; The evaluation rules for homology prediction include: For each subject i, form a sample pair with a candidates and mark them according to whether they are the same person to form a label y i，j , where j = 1, 2, ..., a, and the predicted probability of the homologous sample pair of the subject is: p i =p'(y i,j =1) The p i As the threshold, we count the number of false positives n in all sample pairs of the subject. i ,Right now: n i =count(p'(y i,j ≠1)≥p i ) Among them, count represents the number of sample pairs that meet the conditions, and the number of false positives reflects the number of wrong candidates that the model needs to exclude before reaching the threshold; According to n i The maximum value of sets the interval range b of the number of false alarms; For all the subjects, the false positive number analysis is performed. Assuming the total number of subjects is N, the proportion of subjects whose false positive number is in the interval (0, b / a] is: Where I is an indicator function, which is 1 when the condition is met and 0 otherwise; The b / a corresponding to r being 80% is defined as the proportion of suspected candidates that can be screened out by the model when applied, and this is used to evaluate the homology matching performance.

2. The method for reviewing and homologous matching blood routine test results according to claim 1, characterized in that: The generation process of the training data set includes: The routine blood test data of the subjects who were tested earlier than the set date and were over 10 years old were obtained as training samples. Each subject had at least 2 routine blood test data, and the positive data of the routine blood test data were obtained by pairing the routine blood test data of the same subject, while the negative data were obtained by pairing the routine blood test data of different subjects at the first visit. The corresponding items were subtracted to generate the final training data set.

3. The method for reviewing and homologous matching blood routine test results according to claim 2, characterized in that: The generation process of the test data set includes: The routine blood data of the subjects who were tested after the set date and were over 10 years old were obtained as test samples, and the last routine blood data of each subject was selected from the test samples to form data set I; the second to last routine blood data of each subject and the last routine blood data of other subjects were selected from the test samples to form data set II; the data in data set I and data set II were paired, and the corresponding items were subtracted to generate the final test data set.

4. The method for reviewing and homologous matching blood routine test results according to claim 1, characterized in that: Preprocessing the training data set and the test data set includes: Remove samples whose values ​​are too large or too small from the normal range; The data were standardized to make them conform to normal distribution.

5. The method for reviewing and homologous matching blood routine test results according to claim 1, characterized in that: When the anomaly detection model is trained and tested, its input parameters include: the difference in test indicators of each item of two routine blood tests, the number of the routine blood test instrument, whether the instrument is consistent, the time difference of the routine blood test, the age and gender of the subjects, whether the gender of the paired data is consistent, and the age difference of the paired data.

6. The method for reviewing and homologous matching blood routine test results according to claim 1, characterized in that: The analysis process of the anomaly detection model on the training data set includes: The training data set consists of n samples, represented as: x = {x1, x2, ...x n }, where each sample contains m features; Extract features from the input layer vector x and map it to a low-dimensional hidden layer vector h using a fully connected neural network structure; The feature extraction layer vector h is judged for abnormality, and is mapped to a scalar p using a fully connected layer and an activation function. p represents the probability of a reasonable value. The closer it is to 1, the more likely it is a normal sample, and the closer it is to 0, the more likely it is an abnormal sample. The final result is output according to the abnormal judgment layer scalar p. If p ≥ ε, it is judged as normal, otherwise it is judged as abnormal, where ε is the threshold.

7. The method for reviewing and homologous matching blood routine test results according to claim 1, characterized in that: The anomaly detection model is applicable to the following scenarios: The examinee must have undergone routine blood tests twice or more, and the interval between the last one is within 1 year; The blood routine data of the subjects must not contain outliers. Outliers are those that are greater than the upper limit of the reference range and the interval between them is greater than 3 times the length of the reference range. The examinees must be between 10 and 60 years old.

8. The method for reviewing and homologous matching blood routine test results according to claim 6, characterized in that: The re-inspection decision is: when the probability p is greater than 0.5, re-inspection is required.

9. A blood routine test result review and homology matching system, characterized in that: include: An acquisition module is used to obtain two blood routine data; The analysis module is used to analyze the index difference and related interference factor data of two routine blood test data using the abnormality detection model generated by the method described in any one of claims 1 to 8, and output the result of whether re-examination is needed and whether the two routine blood test data are of the same source.

Images