Prediction method and device for syphilis disease, equipment and medium

Abstract

The invention relates to the technical field of syphilis detection, in particular to a syphilis disease prediction method and device, equipment and a medium. The method comprises the following steps: extracting syphilis feature vectors as syphilis input data by using CIA, TPPA, RPR, WB and MTPPA monitoring data in local clinical data, constructing a training set with syphilis output data, participating in machine learning of a clinical data end to obtain a disease prediction model, and predicting the syphilis disease risk according to the model, so that the prediction accuracy can be improved, and the syphilis disease risk can be predicted. The error caused by single detection means is effectively reduced, the clinical environment adaptability difference of diagnosis management is reduced, and accurate diagnosis of syphilis is facilitated.

Description

Technical Field

[0001] The present application relates to the technical field of syphilis detection, and in particular to a method, device, equipment and medium for predicting syphilis. Background Art

[0002] Syphilis is a common sexually transmitted disease caused by Treponema pallidum (Treponema genus) infecting the human body. It is divided into acquired syphilis and congenital syphilis according to the source of infection. Among them, acquired syphilis refers to syphilis infection in adults through sexual intercourse, blood transfusion, and surgery. It is divided into three stages according to clinical manifestations. The main symptoms include invasion of skin and mucous membranes and involvement of internal organs. It is a more serious sexually transmitted disease; congenital syphilis is syphilis caused by pregnant women with syphilis not receiving timely treatment or improper treatment, and Treponema pallidum directly enters the fetus through the placenta.

[0003] Syphilis detection methods include dark field microscopy, direct fluorescent antibody test, serological test and cerebrospinal fluid test, etc. The most common detection method is serological test. Serological test is divided into specific antibody detection method and non-specific antibody detection method. Specific antibody detection methods mainly include enzyme immunoassay (EIA), chemiluminescent immunoassay (CIA), colloidal gold method, Treponema pallidum particle agglutination method (TPPA), immunoblotting (WB), etc.; non-specific antibody detection methods mainly include rapid plasma reagin card test (RPR), toluidine red unheated serum test (TRUST), Venereal Disease Research Laboratory (VDRL), etc.

[0004] However, among patients with syphilis who receive early treatment, up to 24% of patients may return to anergy several years after treatment (Romanowski B, et al. Serologic response to treatment of infectious syphilis. Ann Intern Med. 1991; 114 (12): 1005-9). On the other hand, non-treponemal antibody tests, such as RPR, TRUST, and VDRL, often decrease in titer after appropriate treatment and eventually become anergy over time, so they are often used to assess disease activity and treatment response. However, these tests may produce anergy results in the early or late stages of syphilis.

[0005] In recent years, implementation of various reverse syphilis serologic screening strategies in high-volume laboratories (i.e., performing a T. pallidum-specific antibody test followed by a non-T. pallidum-specific antibody test or another T. pallidum-specific antibody test) has helped to identify individuals who have previously been treated for syphilis and those who have not. However, this approach has also resulted in a large number of serum samples showing reactivity only on the EIA / CIA, referred to as isolated reactive EIA / CIA results. Isolated reactive EIA / CIA results may occur during the early or latent stages of syphilis because EIA / CIA tests are generally more sensitive than other conventional T. pallidum antibody tests (e.g., TPPA) or non-T. pallidum antibody tests. In addition, it has been observed that isolated reactive CIA specimens persist even after serological reversal (conversion from reactive to non-reactive) of conventional Treponema pallidum antibody detection tests such as TPPA and FTA, indicating the presence of true Treponema pallidum infection (Hunter MG, et al. Significance of isolated reactive treponemal chemiluminescence immunoassay results. J Infect Dis. 2013; 207(9): 1416-23). ​​However, the clinical significance of isolated reactive EIA / CIA results in asymptomatic patients remains uncertain, especially when there is limited literature on recent syphilis exposure or previous treatment. The nature of these results, whether they represent false reactivity, previous infection, or early syphilis, remains unknown. Therefore, patients with isolated reactive EIA / CIA results may require unnecessary investigations, additional treatment, or serological follow-up.

[0006] To address this issue, further analysis of isolated reactive EIA / CIA specimens is essential and needs to be confirmed using additional confirmatory tests. This will help improve the understanding and interpretation of discrepant results in specific situations and provide better guidance for patient management. The optimal testing pathway to confirm or exclude suspected syphilis in samples with isolated Treponema pallidum antibody EIA / CIA results remains unestablished. The Centers for Disease Control and Prevention (CDC) plans to conduct a study to identify isolated Treponema pallidum antibody-reactive CIA sera by immunoblotting (e.g., WB) to determine their reactivity with Treponema pallidum antigens and to determine the cause of unconfirmed Treponema pallidum antibody-reactive results (Centers for Disease C, Prevention. Discordant results from reverse sequence syphilis screening--five laboratories, United States, 2006-2010. MMWR Morb Mortal Wkly Rep. 2011; 60(5):133-7). UK national guidelines recommend the use of IgG immunoblotting as an additional confirmatory test when reactive screening results cannot be confirmed by standard confirmatory tests (Kingston M, French P, Higgins S, McQuillan O, Sukthankar A, Stott C, et al. UK national guidelines on the management of syphilis 2015. Int J STD AIDS. 2016; 27(6): 421-46). Many other studies have also emphasized the need to use WB as a confirmatory method for samples with inconsistent results (Xu S, Li H, Wu X, Guo J, Zhang J, Hu X. Confirmation value of Western blotting in detecting anti-treponema pallidum specific antibodies with suspicious results. Eur J Med Res. 2022; 27(1): 16).However, European guidelines believe that WB does not provide significant added value compared with other Treponema pallidum antibody detection tests (Janier M, Unemo M, Dupin N, Tiplica GS, Potocnik M, Patel R. 2020 European guideline on the management of syphilis. J Eur Acad Dermatol Venereol. 2021; 35(3): 574-88). Summary of the invention

[0007] The inventor of the present application combined isolated reactive CIA data, WB and MTPPA monitoring data, extracted multiple syphilis feature vectors that can characterize these data, used them as syphilis input data, and the corresponding syphilis output data as syphilis output data to construct a training set, combined with multi-center or hospital clinical data terminals for machine learning to build a disease prediction model, and predict the risk of disease based on the model. In this way, the embodiments of the present application provide a prediction method, device, equipment and medium for syphilis to improve the reliability of reaction signals in serum detection of syphilis in the prior art; and realize the process of data model construction to improve the detection and prediction accuracy of syphilis.

[0008] Based on this, the embodiments of the present application disclose at least the following technical solutions:

[0009] In a first aspect, an embodiment discloses a method for predicting syphilis, the method comprising: obtaining syphilis input data and syphilis output data in a local database, the syphilis input data comprising CIA data, TPPA data and WB data, the syphilis output data being the syphilis diagnosis and medical history record data of clinical patients; performing feature extraction on the syphilis input data to obtain a syphilis feature vector; constructing a local training set based on the syphilis feature vector and the syphilis output data; participating in machine learning at the clinical data end based on the local training set to obtain a disease prediction model; inputting the syphilis input data to be tested into the disease prediction model; calculating a risk value based on the disease prediction model; and predicting the risk of syphilis based on the risk value.

[0010] In a second aspect, an embodiment discloses a device for predicting syphilis, the device comprising:

[0011] An acquisition unit, used to acquire syphilis input data and syphilis output data in a local database, wherein the syphilis input data includes CIA data, TPPA data and WB data, and the syphilis output data is the syphilis diagnosis and past medical history record data of clinical patients;

[0012] An extraction unit, used for performing feature extraction on the syphilis input data to obtain a syphilis feature vector;

[0013] A construction unit, used for constructing a local training set according to the syphilis feature vector and the syphilis output data;

[0014] A training unit, used to participate in machine learning at the clinical data end based on the local training set to obtain a disease prediction model;

[0015] An input unit, used for inputting the syphilis input data to be tested into the disease prediction model;

[0016] A calculation unit, used for calculating a risk value according to the disease prediction model;

[0017] An output unit is used to predict the risk of syphilis according to the risk value.

[0018] In a third aspect, an embodiment discloses a device for predicting syphilis, comprising:

[0019] Memory;

[0020] Processor; and

[0021] A syphilis prediction program stored in the memory and executable on the processor, wherein when the prediction program is executed by the processor, the steps of the syphilis prediction method described in the first aspect are implemented.

[0022] In a fourth aspect, an embodiment discloses a computer-readable storage medium having a syphilis prediction program stored thereon, which, when executed by a processor, implements the steps of the syphilis prediction method described in the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 A block diagram of a syphilis prediction device provided in an embodiment.

[0024] Figure 2 The present invention is a flow chart of a method for predicting syphilis provided in an embodiment.

[0025] Figure 3 This is a schematic flow chart of step S20 provided in an embodiment.

[0026] Figure 4 This is a flow chart of step S202 provided in the embodiment.

[0027] Figure 5 This is a schematic flow chart of step S20 provided in an embodiment.

[0028] Figure 6 This is a flow chart of step S204 provided in the embodiment.

[0029] Figure 7 This is a flow chart of step S40 provided in the embodiment.

[0030] Figure 8 This is a schematic diagram of the structure of a syphilis prediction device provided in an embodiment.

[0031] Fig. 9 Screenshot of clinical data storage information provided for the embodiment. DETAILED DESCRIPTION

[0032] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application is further described in detail below in conjunction with the examples. It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application. The reagents not described separately in detail in the present application are all conventional reagents and can be obtained from commercial channels; the methods not described in detail are all conventional experimental methods and can be obtained from the prior art.

[0033] Syphilis prediction device

[0034] The present application provides a device for predicting syphilis. Figure 1 It is a schematic diagram of the structure of the hardware operating environment involved in the embodiment of the present application. It should be noted that: Figure 1 The syphilis prediction device of the present application embodiment may be a PC, or a terminal device with a display function such as a smart phone, a smart TV, a tablet computer, or a portable computer.

[0035] like Figure 1 As shown, the syphilis prediction device may include: a processor 1001, such as a CPU, a network interface 1004, a user interface 1003, a memory 1005, and a communication bus 1002. Among them, the communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display), an input unit such as a keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). The memory 1005 may be a high-speed RAM memory, or a stable memory (non-volatile memory), such as a disk memory. The memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0036] Optionally, the syphilis prediction device may also include a camera, an RF (Radio Frequency) circuit, a sensor, an audio circuit, a WiFi module, etc. Those skilled in the art will appreciate that Figure 1 The structure of the device for predicting syphilis shown in the figure does not constitute a limitation on the device for predicting syphilis, and may include more or less components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0037] like Figure 1 As shown, the memory 1005 as a computer storage medium may include an operating system, a network communication module, a user interface module, and a syphilis prediction program.

[0038] exist Figure 1 In the syphilis prediction device shown, the network interface 1004 is mainly used to connect to the backend server and perform data communication with the backend server; the user interface 1003 is mainly used to connect to the client (user end) and perform data communication with the client; and the processor 1001 can be used to call the syphilis prediction program stored in the memory 1005 and perform the following operations:

[0039] Acquire syphilis input data and syphilis output data in a local database, wherein the syphilis input data includes CIA data, TPPA data and WB data, and the syphilis output data is the syphilis diagnosis and medical history record data of clinical patients;

[0040] Performing feature extraction on the syphilis input data to obtain a syphilis feature vector;

[0041] Constructing a local training set according to the syphilis feature vector and the syphilis output data;

[0042] Participating in machine learning on the clinical data end based on the local training set to obtain a disease prediction model;

[0043] Inputting the syphilis input data to be tested into the disease prediction model;

[0044] Calculating a risk value according to the disease prediction model;

[0045] The risk of syphilis is predicted according to the risk value.

[0046] In some embodiments, the step of extracting features from the syphilis input data to obtain a syphilis feature vector comprises:

[0047] Extract and generate a CIA feature from the CIA data, wherein the CIA feature has three record symbols: "+", "-" and "±";

[0048] Extracting and generating a TPPA feature quantity from the TPPA data, wherein the TPPA feature quantity has three record symbols of “+”, “-” and “±”;

[0049] Extract and generate a WB feature quantity from the WB data, wherein the WB feature quantity has three record symbols of "+", "-" and "±";

[0050] The CIA feature quantity, the TPPA feature quantity and the WB feature quantity are vectorized to obtain the syphilis feature vector.

[0051] In some embodiments, the step of vectorizing the CIA feature, the TPPA feature, and the WB feature to obtain the syphilis feature vector includes:

[0052] Identify the amount of CIA data for which the CIA feature quantities are respectively "+", "-" or "±";

[0053] Identify the TPPA data quantity for which the TPPA characteristic quantity is respectively "+", "-" or "±";

[0054] Identifying the amount of WB data for which the WB feature quantities are respectively "+", "-" or "±";

[0055] A plurality of syphilis feature vectors are generated based on the record symbol of the CIA feature quantity, the record symbol of the TPPA feature quantity, the record symbol of the WB feature quantity, the CIA data volume, the TPPA data volume, and the WB data volume.

[0056] In some embodiments, the syphilis input data further includes RPR data and MTPPA data, and the step of extracting features from the syphilis input data to obtain a syphilis feature vector further includes:

[0057] Extracting and generating an RPR feature from the RPR data, wherein the RPR feature has three recording symbols: "+", "-" and "±";

[0058] Extract and generate an MTPPA feature from the MTPPA data, wherein the MTPPA feature has three record symbols: "+", "-" and "±";

[0059] The CIA feature quantity, the TPPA feature quantity, the RPR feature quantity, the WB feature quantity and the MTPPA feature quantity are vectorized to obtain the syphilis feature vector.

[0060] In some embodiments, the step of performing vector processing on the CIA feature, the TPPA feature, the RPR feature, the WB feature, and the MTPPA feature to obtain the syphilis feature vector includes:

[0061] Identify the amount of CIA data for which the CIA feature quantities are respectively "+", "-" or "±";

[0062] Identify the TPPA data quantity for which the TPPA characteristic quantity is respectively "+", "-" or "±";

[0063] Identifying the amount of RPR data for which the RPR feature quantities are respectively "+", "-" or "±";

[0064] Identify the amount of MTPPA data for which the MTPPA characteristic quantities are respectively "+", "-" or "±";

[0065] Identifying the amount of WB data for which the WB feature quantities are respectively "+", "-" or "±";

[0066] A plurality of the syphilis feature vectors are generated based on the record code of the CIA feature, the record code of the TPPA feature, the record code of the RPR feature, the record code of the WB feature, the MTPPA data volume, and the WB data volume.

[0067] In some embodiments, the step of participating in the machine learning of the clinical data terminal based on the local training set to obtain the disease prediction model includes:

[0068] Obtain the initial model parameters of the model to be trained from the preset server;

[0069] Performing local training on the model to be trained according to the local training set and the initial model parameters to obtain model parameter update;

[0070] The model parameter update is transmitted to the server end, so that the server end aggregates the model parameter update transmitted by the clinical data end to obtain the aggregated model parameter, and when it is detected that the model to be trained is in an unconverged state, the aggregated model parameter is returned to each of the clinical data ends to continue iterative training, until it is detected that the model to be trained is in a converged state, the aggregated model parameter is used as the final parameter of the model to be trained, and the disease prediction model is obtained, and the disease prediction model is transmitted to each of the clinical data ends;

[0071] Receive the disease prediction model transmitted by the server.

[0072] In some embodiments, the disease prediction model is constructed as a convolutional neural network, which includes one convolutional layer and three fully connected layers.

[0073] Prediction of syphilis

[0074] Based on the above hardware structure, various embodiments of the method for predicting syphilis of the present application are proposed.

[0075] Figure 2 A method for predicting syphilis provided by an embodiment is shown. It should be noted that although a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in an order different from that herein. The execution subject of each embodiment of the method for predicting syphilis of the present application may be a terminal device such as a PC, a smart phone, a smart TV, a tablet computer, and a portable computer. For ease of description, the following embodiments are described with the clinical data terminal (i.e., the clinical data terminal) as the execution subject. The method for predicting syphilis includes:

[0076] Step S10, obtaining syphilis input data and syphilis output data in a local database, wherein the syphilis input data includes CIA data, TPPA data and WB data, and the syphilis output data is the syphilis diagnosis and medical history record data of clinical patients.

[0077] The data terminals of each hospital or clinical research center store the monitoring data of their own patients in their local databases, such as syphilis input data records, which record various physiological information of patients, such as CIA data, TPPA data, and WB data, etc. The collection of syphilis input data records can be measured by hospital equipment or reported by patients themselves. Each clinical data terminal can carry out machine learning cooperation, and train machine models by combining the monitoring data of each clinical data terminal, such as combining the monitoring data of syphilis patients, suspected syphilis patients, healthy people or close contacts of syphilis patients in the databases of each clinical data terminal to train syphilis prediction models, which are used to predict the syphilis of patients before doctors diagnose them, so as to provide reference for doctors when diagnosing patients. If the predicted risk value of syphilis is extremely high, the doctor can make further diagnosis of the patient to determine whether the patient really has syphilis. If the risk value is extremely low, the doctor can basically rule out the possibility of syphilis in the patient, and thus make other diagnosis for the patient.

[0078] The clinical data terminal obtains the syphilis input data and the corresponding recorded syphilis output data of the confirmed patients in the local database. The confirmed patients refer to the patients whose diseases have been diagnosed by doctors and confirmed by doctors. The local database stores the syphilis input data and syphilis output data of the confirmed patients. The syphilis output data may include the name or label of the disease, or the severity of the disease.

[0079] Step S20, extracting features from the syphilis input data to obtain a syphilis feature vector.

[0080] The clinical data end extracts features from the acquired syphilis input data to obtain syphilis feature vectors for each data sample. Among them, the data of each physiological information in the syphilis input data record (such as CIA data, TPPA data, and WB data) include some medical connections. The clinical data end can process each physiological information according to the medical connection, extract the medical features of each physiological information, vectorize the medical features, and obtain the syphilis feature vector of each patient. Specifically, for a patient's syphilis input data record, medical features are extracted from the syphilis input data record, and each medical feature is used as a vector element to form the patient's disease feature vector.

[0081] Step S30: constructing local training according to the syphilis feature vector and the syphilis output data.

[0082] The clinical data end constructs a local training sample set based on the disease feature vector and syphilis output data of each patient. Specifically, the local training sample set includes multiple training samples, each of which includes an input value and an output value. The clinical data end uses the disease feature vector of the patient as the input value and the syphilis output data of the patient as the output value to form a training sample. The training samples of all patients constitute the local training sample set.

[0083] Step S40, participating in machine learning at the clinical data end based on the local training set to obtain a disease prediction model.

[0084] After obtaining the local training sample set, the clinical data end uses the local training sample set to participate in machine learning. Specifically, in each iteration of machine learning training, the clinical data end uses the local training sample set to perform local training on the model to be trained, where the model to be trained can be a neural network model or a deep learning model. The final model obtained by machine learning is used as the disease prediction model.

[0085] It should be noted that each clinical data terminal can choose different machine learning models for modeling according to specific tasks and data conditions. The modeling methods include horizontal, vertical, migration and hybrid models. Specifically, the feature dimensions of the monitoring data of each clinical data terminal may be different. If the monitoring data of each clinical data terminal have highly overlapping feature dimensions, but the patient overlap is small, a horizontal machine learning model can be used. If the monitoring data of each clinical data terminal have highly overlapping patients, but the feature dimension overlap is small, a vertical machine learning model can be used. If there is not enough overlap in the samples and feature dimensions of the data set, a transfer learning model can be used.

[0086] In some embodiments, Figure 3 As shown, step S20 includes:

[0087] S201, extracting and generating a CIA feature from the CIA data, wherein the CIA feature has three record symbols of “+”, “-” and “±”; extracting and generating a TPPA feature from the TPPA data, wherein the TPPA feature has three record symbols of “+”, “-” and “±”; extracting and generating a WB feature from the WB data, wherein the WB feature has three record symbols of “+”, “-” and “±”;

[0088] S202: vectorize the CIA feature quantity, the TPPA feature quantity and the WB feature quantity to obtain the syphilis feature vector.

[0089] In some embodiments, Figure 4 As shown, step S202 includes:

[0090] S2021, identifying the amount of CIA data whose CIA feature quantities are respectively “+”, “-” or “±”, identifying the amount of TPPA data whose TPPA feature quantities are respectively “+”, “-” or “±”, and identifying the amount of WB data whose WB feature quantities are respectively “+”, “-” or “±”;

[0091] S2022. Generate a plurality of syphilis feature vectors according to the record symbol of the CIA feature quantity, the record symbol of the TPPA feature quantity, the record symbol of the WB feature quantity, the CIA data volume, the TPPA data volume, and the WB data volume.

[0092] In some embodiments, the syphilis input data also includes RPR data and MTPPA data. Figure 5 As shown, the step S20 includes:

[0093] S203, extract and generate CIA feature from the CIA data, the CIA feature has three record symbols of “+”, “-” and “±”, extract and generate TPPA feature from the TPPA data, the TPPA feature has three record symbols of “+”, “-” and “±”, extract and generate WB feature from the WB data, the WB feature has three record symbols of “+”, “-” and “±”, extract and generate RPR feature from the RPR data, the RPR feature has three record symbols of “+”, “-” and “±”, extract and generate MTPPA feature from the MTPPA data, the MTPPA feature has three record symbols of “+”, “-” and “±”;

[0094] S204. Perform vectorization processing on the CIA feature quantity, the TPPA feature quantity, the RPR feature quantity, the WB feature quantity, and the MTPPA feature quantity to obtain the syphilis feature vector.

[0095] In some embodiments, Figure 6 As shown, step S204 includes:

[0096] S2041, identifying the amount of CIA data whose CIA characteristic quantities are "+", "-" or "±", identifying the amount of TPPA data whose TPPA characteristic quantities are "+", "-" or "±", identifying the amount of RPR data whose RPR characteristic quantities are "+", "-" or "±", identifying the amount of MTPPA data whose MTPPA characteristic quantities are "+", "-" or "±", and identifying the amount of WB data whose WB characteristic quantities are "+", "-" or "±";

[0097] S2042, based on the record code of the CIA feature, the record code of the TPPA feature, the record code of the RPR feature, the record code of the MTPPA, the record code of the WB feature,

[0098] The CIA data volume, the MTPPA data volume, the RPR data volume, the MTPPA data volume, and the WB data volume generate a plurality of the syphilis feature vectors.

[0099] In some embodiments, the syphilis feature vector is characterized by a direction quantity and a distribution rate. The direction quantity is selected from the string "CIA+TPPA+WB+", "CIA+TPPA+WB-", "CIA+TPPA+WB±", "CIA+TPPA-WB+", "CIA+TPPA-WB-", "CIA+TPPA-WB±", "CIA-TPPA-WB+", "CIA-TPPA-WB-", "CIA-TPPA-WB±", "CIA+TPPA-RPR-MTPPA+", "CIA+TPPA-RPR-MTPPA-WB+", "CIA+TPPA-RPR-MTPPA-WB-", "CIA+TPPA-RPR-MTPPA-WB±",

[0100] “CIA+TPPA-RPR-WB+”, “CIA+TPPA-RPR-WB-MTPPA-”, “CIA+TPPA-RPR-WB-MTPPA+”, “CIA+TPPA-RPR-WB±MTPPA-”, “CIA+TPPA-RPR-WB±MTPPA+”. The distribution rate is selected from “CIA+TPPA+WB+”, “CIA+TPPA+WB-”, “CIA+TPPA+WB±”, “CIA+TPPA-WB+”, “CIA+TPPA-WB-”, “CIA+TPPA-WB±”, “CIA-TPPA-WB+”, “CIA-TPPA-WB-”, “CIA-TPPA-WB±”, “CIA+TPPA-RPR-MTPPA+”, “CIA+TPPA-RPR-MTPPA-WB+”, “CIA+TPPA-RPR-MTPPA-WB-”, “CIA+TPPA-RPR-MTPPA-WB±”,

[0101] The data volumes corresponding to “CIA+TPPA-RPR-WB+”, “CIA+TPPA-RPR-WB-MTPPA-”, “CIA+TPPA-RPR-WB-MTPPA+”, “CIA+TPPA-RPR-WB±MTPPA-” and “CIA+TPPA-RPR-WB±MTPPA+” are expressed as a proportion of the clinical quantity.

[0102] In the present application, the "+" character indicates a positive detection, "-" indicates a negative detection, and "±" indicates that it cannot be detected and determined. For example, "CIA+" indicates sample data detected as positive by CIA, "CIA-" indicates sample data detected as negative by CIA, and "CIA±" indicates sample data that cannot be detected and determined by CIA. For example, "TPPA+" indicates sample data detected as positive by TPPA, "TPPA-" indicates sample data detected as negative by TPPA, and "TPPA±" indicates sample data that cannot be detected and determined by TPPA. For example, "WB+" indicates sample data detected as positive by WB, "WB-" indicates sample data detected as negative by WB, and "WB±" indicates sample data that cannot be detected and determined by WB. For example, "RPR+" indicates sample data detected as positive by RPR, "RPR-" indicates sample data detected as negative by RPR, and "RPR±" indicates sample data that cannot be detected and determined by RPR. For example, “MTPPA+” indicates sample data detected as positive by MTPPA, “MTPPA-” indicates sample data detected as negative by MTPPA, and “MTPPA±” indicates sample data that cannot be determined by MTPPA detection.

[0103] For example, "CIA+TPPA+WB+" represents sample data that is detected as positive by CIA, TPPA and WB in sequence. The direction of the corresponding syphilis feature vector is represented by "CIA+TPPA+WB+", and the distribution rate is the proportion of the sample data to the total amount of training sample data used.

[0104] For example, "CIA+TPPA+WB-" represents sample data that is detected as positive by CIA, TPPA, and negative by WB. The direction of the corresponding syphilis feature vector is represented by "CIA+TPPA+WB-", and the distribution rate is the proportion of the sample data to the total amount of training sample data used.

[0105] For example, "CIA+TPPA+WB±" represents sample data that is detected as positive by CIA and TPPA respectively, and cannot be detected by WB. The direction of the corresponding syphilis feature vector is represented by "CIA+TPPA+WB±", and the distribution rate is the proportion of this sample data to the total amount of training sample data used.

[0106] For example, "CIA+TPPA-WB+" represents sample data that is detected as positive by CIA, negative by TPPA, and positive by WB. The direction of the corresponding syphilis feature vector is represented by "CIA+TPPA-WB+", and the distribution rate is the proportion of the sample data to the total training sample data used.

[0107] For example, "CIA+TPPA-WB-" represents sample data that is detected as positive by CIA, and negative by TPPA and WB. The direction of the corresponding syphilis feature vector is represented by "CIA+TPPA-WB-", and the distribution rate is the proportion of the sample data to the total amount of training sample data used.

[0108] For example, "CIA+TPPA-WB±" represents sample data that is detected as positive by CIA, negative by TPPA, and cannot be detected by WB. The direction of the corresponding syphilis feature vector is represented by "CIA+TPPA-WB±", and the distribution rate is the proportion of the sample data to the total amount of training sample data used.

[0109] For example, "CIA-TPPA-WB+" represents sample data that is detected as negative by CIA and TPPA and positive by WB. The direction of the corresponding syphilis feature vector is represented by "CIA-TPPA-WB+", and the distribution rate is the proportion of the sample data to the total amount of training sample data used.

[0110] For example, "CIA-TPPA-WB-" represents sample data that is tested negative by CIA, TPPA, and WB in sequence. The direction of the corresponding syphilis feature vector is represented by "CIA-TPPA-WB-", and the distribution rate is the proportion of the sample data to the total amount of training sample data used.

[0111] For example, "CIA-TPPA-WB±" represents sample data that is detected as negative by CIA, TPPA, and cannot be detected by WB. The direction of the corresponding syphilis feature vector is represented by "CIA-TPPA-WB±", and the distribution rate is the proportion of the sample data to the total amount of training sample data used.

[0112] For example, "CIA+TPPA-RPR-MTPPA+" represents sample data that is detected as positive by CIA, negative by TPPA and RPR, and positive by MTPPA. The direction of the corresponding syphilis feature vector is represented by "CIA+TPPA-RPR-MTPPA+", and the distribution rate is the proportion of the sample data to the total amount of training sample data used.

[0113] For example, "CIA+TPPA-RPR-MTPPA-WB+" represents sample data that is detected as positive by CIA, negative by TPPA, RPR and MTPPA, and positive by WB. The direction of the corresponding syphilis feature vector is represented by "CIA+TPPA-RPR-MTPPA-WB+", and the distribution rate is the proportion of the sample data to the total training sample data used.

[0114] For example, "CIA+TPPA-RPR-MTPPA-WB-" represents sample data that is detected as positive by CIA, and negative by TPPA, RPR, MTPPA, and WB. The direction of the corresponding syphilis feature vector is represented by "CIA+TPPA-RPR-MTPPA-WB-", and the distribution rate is the proportion of the sample data to the total amount of training sample data used.

[0115] For example, "CIA+TPPA-RPR-MTPPA-WB±" represents sample data that is detected as positive by CIA, negative by TPPA, RPR, and MTPPA, and cannot be detected by WB. The direction quantity of the corresponding syphilis feature vector is expressed as "CIA+TPPA-RPR-MTPPA-WB±", and the distribution rate is the proportion of this sample data to the total amount of training sample data used.

[0116] For example, "CIA+TPPA-RPR-WB+" represents sample data that is detected as positive by CIA, negative by TPPA and RPR, and positive by WB. The direction of the corresponding syphilis feature vector is represented by "CIA+TPPA-RPR-WB+", and the distribution rate is the proportion of the sample data to the total training sample data used.

[0117] For example, "CIA+TPPA-RPR-WB-MTPPA-" represents sample data that is detected as positive by CIA, negative by TPPA, RPR, WB and negative by MTPPA in sequence. The direction of the corresponding syphilis feature vector is represented by "CIA+TPPA-RPR-WB-MTPPA-", and the distribution rate is the proportion of the sample data to the total amount of training sample data used.

[0118] For example, "CIA+TPPA-RPR-WB-MTPPA+" represents sample data that is detected as positive by CIA, negative by TPPA, RPR, and WB, and positive by MTPPA. The direction of the corresponding syphilis feature vector is represented by "CIA+TPPA-RPR-WB-MTPPA+", and the distribution rate is the proportion of the sample data to the total training sample data used.

[0119] For example, "CIA+TPPA-RPR-WB±MTPPA-" represents sample data that is detected as positive by CIA, negative by TPPA and RPR, unable to be detected by WB, and positive by MTPPA. The direction of the corresponding syphilis feature vector is represented by "CIA+TPPA-RPR-WB±MTPPA-", and the distribution rate is the proportion of the sample data to the total amount of training sample data used.

[0120] For example, "CIA+TPPA-RPR-WB±MTPPA+" represents sample data that is detected as positive by CIA, negative by TPPA and RPR, unable to be detected by WB, and negative by MTPPA. The direction of the corresponding syphilis feature vector is represented by "CIA+TPPA-RPR-WB±MTPPA+", and the distribution rate is the proportion of the sample data to the total amount of training sample data used.

[0121] In the present application, "CIA" means chemiluminescent immunoassay, "TPPA" means Treponema pallidum particle agglutination assay, "WB" means immunoblotting, "RPR" means rapid plasma reagin card test, and "MTPPA" means modified Treponema pallidum particle agglutination assay.

[0122] The detection principle of TPPA is to coat the purified bacterial components of Treponema pallidum (Nichols strain) on the carrier gelatin as sensitizing particles. When the Treponema pallidum antibodies in the sample to be tested react with the sensitizing particles to agglutinate, a cohesion effect is generated, thereby realizing the detection of the antibody titer in the sample based on the gel. The detection sample addition process and judgment of TPPA are shown in Table 3.

[0123] The TPPA kit mainly includes dissolving solution, serum diluent, sensitized particles (lyophilized powder), non-sensitized particles (lyophilized powder), and positive control serum (adjusted to a final dilution multiple of 1:320 with anti-Treponema pallidum (Nichols strain) rabbit immune serum).

[0124] The operation steps of the TPPA kit are shown in Table 1.

[0125] 1) Add 100 μL of serum diluent to the first well of the microplate, and add 25 μL of serum diluent to each well from the second to the fourth well.

[0126] 2) Add 25 μL of sample to well 1 and dilute in half to well 4.

[0127] 3) Add 25 μL of non-sensitized particles to the third well and add 25 μL of sensitized particles to the fourth well.

[0128] 4) Mix the plate with a plate mixer for 30 seconds at a strength that does not cause the contents of the microplate to splash out, cover it and leave it horizontally at room temperature (15-30°C). After 2 hours, record and observe the reaction image on the observation mirror, or measure it using an immunodilution determination device. In addition, even if it is left to stand until the next day, it will not affect the determination result.

[0129] Table 1TPPA

[0130]

[0131] As shown in Table 2, the operating steps of the MTPPA method are as follows: except for adding 25 μL of sensitized particles to the second well (instead of the fourth well), the other steps are the same as the TPPA.

[0132] Table 2MTPPA

[0133]

[0134] A specific embodiment of step S20 includes:

[0135] 1. Clinical Data

[0136] A prospective cohort study was conducted in Wuhan Tongji Hospital, and the clinical data of 119,002 samples screened for syphilis from January 1, 2015 to January 6, 2017 were included. Fig. 9 As shown, these clinical data include CIA data, TPPA data, RPR data, WB data, MTPPA data and syphilis output data.

[0137] The clinical data of 119,002 samples also recorded the age, gender and history of syphilis of each sample (marked by "Yes", "No" or "None"), ranging from 0 to 102 years old (median 42.0 years, IQR: 28.0-56.0 years), and males accounted for 50.1%. Among them, the CIA data of 2385 samples (2.0%) showed "CIA+" or "CIA-". Among them, the syphilis output data were marked with "Yes", "No" or "None".

[0138] 2. Generate syphilis feature vector

[0139] In some embodiments, 189 syphilis input data are obtained, including CIA data, TPPA data and WB data. Table 3 shows a plurality of syphilis feature vectors constructed according to the syphilis input data.

[0140] In Table 3, "CIA+TPPA+WB+" represents the sample data that is detected as positive by CIA, TPPA and WB in sequence, and the direction of the corresponding syphilis feature vector is represented by "CIA+TPPA+WB+", and the distribution rate is the proportion of this sample data to the amount of training sample data used.

[0141] "CIA+TPPA+WB-" represents sample data that is detected as positive by CIA, TPPA, and negative by WB. The direction of the corresponding syphilis feature vector is represented by "CIA+TPPA+WB-", and the distribution rate is the proportion of this sample data to the total amount of training sample data used.

[0142] "CIA+TPPA+WB±" represents sample data that are detected as positive by CIA and TPPA respectively, and cannot be detected by WB. The direction of the corresponding syphilis feature vector is represented by "CIA+TPPA+WB±", and the distribution rate is the proportion of this sample data to the total amount of training sample data used.

[0143] "CIA+TPPA-WB+" represents sample data that are detected as positive by CIA, negative by TPPA, and positive by WB. The direction of the corresponding syphilis feature vector is represented by "CIA+TPPA-WB+", and the distribution rate is the proportion of this sample data to the total amount of training sample data used.

[0144] "CIA+TPPA-WB-" represents sample data that is detected as positive by CIA, and negative by TPPA and WB. The direction of the corresponding syphilis feature vector is represented by "CIA+TPPA-WB-", and the distribution rate is the proportion of this sample data to the total amount of training sample data used.

[0145] "CIA+TPPA-WB±" represents the sample data that are detected as positive by CIA, negative by TPPA, and cannot be detected by WB. The direction of the corresponding syphilis feature vector is represented by "CIA+TPPA-WB±", and the distribution rate is the proportion of this sample data to the total amount of training sample data used.

[0146] "CIA-TPPA-WB-" represents sample data that are tested negative by CIA, TPPA and WB in sequence. The direction of the corresponding syphilis feature vector is represented by "CIA-TPPA-WB-", and the distribution rate is the proportion of this sample data to the total amount of training sample data used.

[0147] Table 3 Multiple syphilis feature vectors provided in the embodiment

[0148] Direction quantity Distribution rate CIA+TPPA+WB+ 67.2％ CIA+TPPA+WB- 14.8％ CIA+TPPA+WB± 18.0％ CIA+TPPA-WB+ 6.74％ CIA+TPPA-WB- 84.3％ CIA+TPPA-WB± 8.99％ CIA-TPPA-WB- 20.6％

[0149] In some embodiments, syphilis feature vectors are extracted from 85 sets of "CIA+TPPA-RPR-" clinical data. The 85 sets of "CIA+TPPA-RPR-" clinical data reflect 85 sets of isolated reactive CIA data.

[0150] In the present application, there are two methods for further determination of these isolated reactive CIA data. One of the methods is to perform MTPPA confirmation on these isolated reactive CIA data. If MTPPA is reactive, that is, it is marked as "CIA+TPPA-RPR-MTPPA+", it is stored as positive for syphilis infection. Thus, a syphilis feature vector in the training sample has the direction quantity of "CIA+TPPA-RPR-MTPPA+" and its distribution rate in the isolated reactive CIA data, and the corresponding syphilis output data is the syphilis diagnosis and medical history record data of clinical patients, that is, positive for syphilis infection. If MTPPA is unreactive, that is, it is marked as "CIA+TPPA-RPR-MTPPA-", WB is further used for confirmation. If WB is reactive, that is, it is marked as "CIA+TPPA-RPR-MTPPA-WB+", it is stored as syphilis infection. Thus, a syphilis feature vector in the training sample has the direction of "CIA+TPPA-RPR-MTPPA-WB+" and its distribution rate in the isolated reactive CIA data, and the corresponding syphilis output data is the syphilis diagnosis and medical history record data of the clinical patient, that is, syphilis infection is positive. If WB is unreactive, that is, marked as "CIA+TPPA-RPR-MTPPA-WB-", it is stored as no syphilis infection or syphilis cannot be diagnosed. Thus, a syphilis feature vector in the training sample has the direction of "CIA+TPPA-RPR-MTPPA-WB-" and its distribution rate in the isolated reactive CIA data, and the corresponding syphilis output data is the syphilis diagnosis and medical history record data of the clinical patient, that is, no syphilis infection or syphilis cannot be diagnosed. Thus, the syphilis feature vector shown in Table 4 is obtained. In Table 4 , the data of isolated reactive CIA of the standard in the direction of “CIA+TPPA-RPR-MTPPA-WB- / ±” were negative by MTPPA test, and negative or indeterminate by WB test, respectively.

[0151] Table 4 Multiple syphilis feature vectors provided in the embodiment

[0152] Direction quantity Distribution rate CIA+TPPA-RPR-MTPPA+ 25.8％ CIA+TPPA-RPR-MTPPA-WB+ 2.35％ CIA+TPPA-RPR-MTPPA-WB- / ± 71.16％

[0153] Another method is to perform WB confirmation on these isolated reactive CIA data. If the WB is reactive, that is, marked as "CIA+TPPA-RPR-WB+", it is stored as positive for syphilis infection. Thus, a syphilis feature vector in the training sample has a direction quantity of "CIA+TPPA-RPR-WB+" and its distribution rate in the isolated reactive CIA data, and the corresponding syphilis output data is the syphilis diagnosis and medical history record data of clinical patients, that is, positive for syphilis infection. If the WB is unreactive, that is, marked as "CIA+TPPA-RPR-WB-", MTPPA is further used for confirmation. If the WB is reactive, that is, marked as "CIA+TPPA-RPR-WB-MTPPA+", it is stored as syphilis infection. Thus, a syphilis feature vector in the training sample has the direction of "CIA+TPPA-RPR-WB-MTPPA-" and its distribution rate in the isolated reactive CIA data, and the corresponding syphilis output data is the syphilis diagnosis and medical history record data of clinical patients, that is, syphilis infection is positive. If WB is unreactive, that is, marked as "CIA+TPPA-RPR-MTPPA-WB-", it is stored as no syphilis infection or syphilis cannot be diagnosed. Thus, a syphilis feature vector in the training sample has the direction of "CIA+TPPA-RPR-MTPPA-WB-" and its distribution rate in the isolated reactive CIA data, and the corresponding syphilis output data is the syphilis diagnosis and medical history record data of clinical patients, that is, no syphilis infection or syphilis cannot be diagnosed. Thus, the syphilis feature vector shown in Table 5 is obtained. In Table 5, the data of isolated reactive CIA of the quantitative standard in the direction of "CIA+TPPA-RPR-WB- / ±MTPPA+" were negative or indeterminate by WB test, and positive by MTPPA test. The data of isolated reactive CIA of the quantitative standard in the direction of "CIA+TPPA-RPR-WB- / ±MTPPA-" were negative or indeterminate by WB test, and negative by MTPPA test.

[0154] Table 5 Multiple syphilis feature vectors provided in the embodiment

[0155] Direction quantity Distribution rate CIA+TPPA-RPR-WB+ 4.71％ CIA+TPPA-RPR-WB- / ±MTPPA+ 23.53％ CIA+TPPA-RPR-WB- / ±MTPPA- 71.16％

[0156] Differences in populations, the prevalence of Treponema pallidum, or the performance of screening and confirmatory tests may account for the differences in disease diagnosis. In addition, the lack of valid data on the diagnosis and management of syphilis limits the use of WB in clinical settings. Moreover, the analytical sensitivity of WB is lower compared with TPPA. In addition, an unreactive TPPA test does not exclude the presence of infectious syphilis, and WB may help identify syphilis in cases with conflicting serological results.

[0157] Therefore, through step S20 provided in the embodiment of the present application, by extracting multiple syphilis feature vectors and constructing a local training set with the syphilis output data, the disease prediction model can be effectively trained, the prediction accuracy of the model can be improved, the errors of these single detection methods can be effectively reduced, the differences in clinical environment adaptability of diagnostic management can be reduced, and contradictory results of serological tests can be retrieved.

[0158] In some embodiments, such as Figure 7 As shown, step S40 includes:

[0159] Step S401, obtaining the initial model parameters of the model to be trained from the preset server. The server transmits the initial model parameters of the model to be trained to each clinical data terminal participating in machine learning. The model parameters can be the weight parameters of the connection between the nodes of the neural network. The clinical data terminal obtains the initial model parameters of the model to be trained from the server.

[0160] Step S402, locally train the model to be trained according to the local training sample set and the initial model parameters to obtain updated model parameters. After obtaining the initial model parameters, the clinical data end locally trains the model to be trained according to the initial model parameters and the local training sample set to obtain updated model parameters. Specifically, the local training process may be that the clinical data end substitutes the initial model parameters into the model to be trained, takes a diseased feature vector in a local training sample as input, takes the syphilis output data as output, calculates the gradient value using the local training sample set, updates the initial model parameters according to the gradient value, and obtains updated model parameters, that is, obtains updated model parameters.

[0161] Step S403, the model parameter update is transmitted to the server end, so that the server end aggregates the model parameter updates transmitted by each clinical data end to obtain aggregated model parameters, and returns the aggregated model parameters to each clinical data end to continue iterative training when it is detected that the model to be trained is in an unconverged state, until it is detected that the model to be trained is in a converged state, and uses the aggregated model parameters as the final parameters of the model to be trained to obtain a disease prediction model, and transmits the disease prediction model to each clinical data end.

[0162] The clinical data end transmits the model parameter updates obtained from local training to the server. The server receives the model parameter updates transmitted by each clinical data end, aggregates each model parameter update to obtain the aggregate model parameters, and the aggregation process can be a weighted average of each model parameter update; after obtaining the aggregate model parameters, the server detects whether the model to be trained is in a convergence state. If it is detected that it is in a non-convergence state, the aggregate model parameters are transmitted to each clinical data end, and each clinical data end continues to perform local training based on the aggregate model parameters and the local training sample set; the cycle is repeated until the server detects that the model to be trained is in a convergence state, the training is terminated, and the latest aggregate model parameters are used as the final parameters of the model to be trained, that is, the final disease prediction model is obtained; the server transmits the obtained disease prediction model to each clinical data end, specifically, the final parameters can be transmitted to each clinical data end, or the final disease prediction model file can be transmitted to each clinical data end.

[0163] The server can detect whether the model to be trained is in a converged state by calculating the difference between the latest aggregate model parameters and the last aggregate model parameters. If the difference is less than a preset value, the model to be trained is determined to be in a converged state. If it is not less than the preset value, the model to be trained is determined to be in a non-converged state. It can also be determined whether the number of iterative training reaches the preset number. If it reaches the preset number, the model to be trained is determined to be in a converged state. It can also be determined whether the training time is greater than the preset time. If it is greater than the preset time, the model to be trained is determined to be in a converged state. The preset value, preset number of times and preset time can be set as needed.

[0164] In some embodiments, the clinical data end may encrypt the model parameter updates obtained through local training according to a preset encryption algorithm, and transmit the encrypted model parameter updates to the server end, which performs aggregation processing based on the encrypted model parameter updates. The preset encryption algorithm may be a homomorphic encryption algorithm.

[0165] Step S404, receiving the disease prediction model transmitted by the server. The clinical data end receives the disease prediction model transmitted by the server, that is, each clinical data end obtains the disease prediction model obtained by machine learning. The clinical data end can predict the patient's risk value based on the disease prediction model.

[0166] In this embodiment, the syphilis input data and syphilis output data of the confirmed patients in the local database are obtained through each clinical data terminal; the syphilis input data is feature extracted to obtain the disease feature vector of each patient; a local training sample set is constructed according to the disease feature vector and the disease of each patient; based on the local training sample set, the machine learning of each clinical data terminal is participated in to obtain the disease prediction model. By combining the data of each clinical data terminal and conducting federated training, a high-quality disease prediction model can be trained without leaking the privacy of the patients at the clinical data terminal, thereby playing an active auxiliary role in the doctor's diagnosis process.

[0167] Furthermore, the syphilis prediction method of the present application further includes, after step S40:

[0168] Step S50, input the syphilis input data to be tested into the disease prediction model. The syphilis input data is the syphilis input data and case text of the target user. For the target user to be diagnosed by a doctor, the clinical data terminal can obtain the syphilis input data and case text of the target user. The case text can be a handwritten or electronic case text transmitted by the doctor, which can record the doctor's preliminary diagnosis results of the target user.

[0169] Step S60, calculating the risk value according to the disease prediction model.

[0170] The clinical data end can extract the features of the target user's syphilis input data to obtain the target user's feature vector, input the feature vector into the disease prediction model, and the disease prediction model outputs the risk value of the target user. For example, when the clinical data end is trained based on the disease data of syphilis patients, suspected syphilis patients, healthy people or close contacts of syphilis patients, the disease prediction model can be used to predict the probability of patients suffering from syphilis, so as to assist doctors in diagnosing syphilis in patients.

[0171] In some embodiments, the disease prediction model is constructed from a convolutional neural network.

[0172] Step S70, predicting the risk of syphilis according to the risk value. Inputting the case text, the risk value and the syphilis input data of the target user into a pre-trained convolutional neural network to obtain a recommended diagnosis scheme for the doctor to use as a reference for diagnosis.

[0173] In this embodiment, after obtaining the risk value of the target user, if the risk value of the target user exceeds a certain threshold, such as the predicted risk value of syphilis of the target user is higher than a certain threshold, it means that the target user is likely to have syphilis. At this time, a diagnostic plan can be further recommended to the target user to help the doctor choose a diagnostic plan for the target user.

[0174] The clinical data end inputs the case text, the target user's risk value output by the disease prediction model, and the target user's syphilis input data record into the pre-trained convolutional neural network to obtain a recommended diagnosis plan. Among them, the convolutional neural network can output the number of the recommended diagnosis plan, and the clinical data end searches for the corresponding recommended diagnosis plan in the database based on the output number. The recommended diagnosis plan can record a feasible treatment method for the disease, such as medication. Doctors can formulate specific treatment plans for target users based on the recommended diagnosis plan output by the clinical data end.

[0175] The training process of the convolutional neural network can be: the convolutional neural network is composed of one convolutional layer and three fully connected layers; the clinical data end obtains a large number of case texts, risk values ​​and syphilis input data records of confirmed patients, performs feature extraction on the case texts and syphilis input data records, vectorizes the extracted features and risk values ​​as input vectors, and uses the manually labeled recommended diagnostic schemes as outputs for calculating loss values, performs multiple iterative training until the convolutional neural network is detected to converge, and a trained convolutional neural network is obtained.

[0176] In this embodiment, after obtaining the risk value of the target user, the case text, risk value and syphilis input data of the target user are input into a pre-trained convolutional neural network to obtain a recommended diagnosis plan to assist doctors in making specific treatment plans for the target user, thereby helping doctors improve diagnostic efficiency.

[0177] Syphilis prediction device

[0178] In addition, the present application also provides a syphilis prediction device, referring to Figure 8 The above-mentioned syphilis prediction device includes an acquisition unit 10, an extraction unit 20, a construction unit 30, a training unit 40, an input unit 50, a calculation unit 60, and an output unit 70.

[0179] The acquisition unit 10 is used to acquire syphilis input data and syphilis output data in a local database, wherein the syphilis input data includes CIA data, TPPA data and WB data, and the syphilis output data is the syphilis diagnosis and medical history record data of clinical patients.

[0180] The extraction unit 20 is used to perform feature extraction on the syphilis input data to obtain a syphilis feature vector.

[0181] The construction unit 30 is used to construct a local training set according to the syphilis feature vector and the syphilis output data.

[0182] The training unit 40 is used to participate in machine learning at the clinical data end based on the local training set to obtain a disease prediction model.

[0183] The input unit 50 is used to input the syphilis input data to be tested into the disease prediction model;

[0184] The calculation unit 60 is used to calculate the risk value according to the disease prediction model;

[0185] The output unit 70 is used to predict the risk of syphilis according to the risk value.

[0186] In some embodiments, the extraction unit 20 includes a first extraction module 201 and a first generation module 202 .

[0187] The first extraction module 201 is used to extract and generate CIA feature quantities from the CIA data, and the CIA feature quantities have three record symbols of "+", "-" and "±", extract and generate TPPA feature quantities from the TPPA data, and the TPPA feature quantities have three record symbols of "+", "-" and "±", and extract and generate WB feature quantities from the WB data, and the WB feature quantities have three record symbols of "+", "-" and "±".

[0188] The first generating module 202 is used to perform vectorization processing on the CIA feature quantity, the TPPA feature quantity and the WB feature quantity to obtain the syphilis feature vector.

[0189] In some embodiments, the extraction unit 20 includes a second extraction module 203 and a second generation module 204 .

[0190] The second extraction module 203 is used to extract and generate RPR feature quantities from the RPR data, and the RPR feature quantities have three record symbols of "+", "-" and "±", and to extract and generate MTPPA feature quantities from the MTPPA data, and the MTPPA feature quantities have three record symbols of "+", "-" and "±".

[0191] The second generating module 204 performs vectorization processing on the CIA feature quantity, the TPPA feature quantity, the RPR feature quantity, the WB feature quantity and the MTPPA feature quantity to obtain the syphilis feature vector.

[0192] In some embodiments, Figure 8 As shown, the training unit 40 includes a model acquisition module 401, a local training module 402, a transmission module 403 and a receiving module 404.

[0193] The acquisition module 401 is used to obtain the initial model parameters of the model to be trained from a preset server.

[0194] The local training module 402 is used to perform local training on the model to be trained according to the local training sample set and the initial model parameters to obtain model parameter updates.

[0195] The transmission module 403 is used to transmit the model parameter update to the server end, so that the server end can aggregate the model parameter updates transmitted by each clinical data end to obtain aggregated model parameters, and return the aggregated model parameters to each clinical data end to continue iterative training when it is detected that the model to be trained is in an unconverged state, until it is detected that the model to be trained is in a converged state, and use the aggregated model parameters as the final parameters of the model to be trained to obtain a disease prediction model, and transmit the disease prediction model to each clinical data end.

[0196] The receiving module 404 is used to receive the disease prediction model transmitted by the server.

[0197] Furthermore, the acquisition module 10 is also used to acquire the syphilis input data record and case text of the target user after obtaining the disease prediction model.

[0198] The input unit 50 is also used to input the syphilis input data record of the target user into the disease prediction model to obtain the risk value of the target user; input the case text, the risk value and the syphilis input data record of the target user into the pre-trained convolutional neural network to obtain a recommended diagnosis plan for the doctor as a diagnostic reference.

[0199] Furthermore, the convolutional neural network includes one convolutional layer and three fully connected layers.

[0200] The expanded contents of the specific implementation of the device for predicting syphilis of the present application are basically the same as the embodiments of the method for predicting syphilis mentioned above, and will not be elaborated here.

[0201] Computer readable storage medium

[0202] In addition, an embodiment of the present application further proposes a computer-readable storage medium, on which a syphilis prediction program is stored. When the syphilis prediction program is executed by a processor, the steps of the syphilis prediction method described above are implemented.

[0203] The expanded contents of the specific implementation methods of the syphilis prediction device and computer-readable storage medium of the present application are basically the same as the above-mentioned embodiments of the syphilis prediction method, and will not be elaborated here.

[0204] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or system including the element.

[0205] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0206] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0207] The above is only a preferred specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed in the present application should be covered within the protection scope of the present application.

Claims

1. A method for predicting syphilis, the method comprising: Acquire syphilis input data and syphilis output data in a local database, wherein the syphilis input data includes CIA data, TPPA data and WB data, and the syphilis output data is the syphilis diagnosis and medical history record data of clinical patients; Performing feature extraction on the syphilis input data to obtain a syphilis feature vector; Constructing a local training set according to the syphilis feature vector and the syphilis output data; Participating in machine learning on the clinical data end based on the local training set to obtain a disease prediction model; Inputting the syphilis input data to be tested into the disease prediction model; Calculating a risk value according to the disease prediction model; The risk of syphilis is predicted according to the risk value.

2. The prediction method according to claim 1, wherein the step of extracting features from the syphilis input data to obtain a syphilis feature vector comprises: Extract and generate a CIA feature from the CIA data, the CIA feature having three record symbols of "+", "-" and "±", extract and generate a TPPA feature from the TPPA data, the TPPA feature having three record symbols of "+", "-" and "±", and extract and generate a WB feature from the WB data, the WB feature having three record symbols of "+", "-" and "±"; The CIA feature quantity, the TPPA feature quantity and the WB feature quantity are vectorized to obtain the syphilis feature vector.

3. The prediction method according to claim 2, wherein the step of vectorizing the CIA feature quantity, the TPPA feature quantity and the WB feature quantity to obtain the syphilis feature vector comprises: Identify the CIA data volume whose CIA feature quantities are "+", "-" or "±", identify the TPPA data volume whose TPPA feature quantities are "+", "-" or "±", and identify the WB data volume whose WB feature quantities are "+", "-" or "±"; A plurality of syphilis feature vectors are generated based on the record symbol of the CIA feature quantity, the record symbol of the TPPA feature quantity, the record symbol of the WB feature quantity, the CIA data volume, the TPPA data volume, and the WB data volume.

4. The prediction method according to claim 2, wherein the syphilis input data further comprises RPR data and MTPPA data, and the step of extracting features from the syphilis input data to obtain a syphilis feature vector further comprises: Extracting and generating an RPR feature from the RPR data, the RPR feature having three record symbols of "+", "-" and "±", and extracting and generating an MTPPA feature from the MTPPA data, the MTPPA feature having three record symbols of "+", "-" and "±"; The CIA feature quantity, the TPPA feature quantity, the RPR feature quantity, the WB feature quantity and the MTPPA feature quantity are vectorized to obtain the syphilis feature vector.

5. The prediction method according to claim 4, wherein the step of vectorizing the CIA feature, the TPPA feature, the RPR feature, the WB feature, and the MTPPA feature to obtain the syphilis feature vector comprises: Identify the CIA data volume whose CIA characteristic quantities are "+", "-" or "±", identify the TPPA data volume whose TPPA characteristic quantities are "+", "-" or "±", identify the RPR data volume whose RPR characteristic quantities are "+", "-" or "±", identify the MTPPA data volume whose MTPPA characteristic quantities are "+", "-" or "±", and identify the WB data volume whose WB characteristic quantities are "+", "-" or "±"; A plurality of the syphilis feature vectors are generated based on the record code of the CIA feature, the record code of the TPPA feature, the record code of the RPR feature, the record code of the WB feature, the MTPPA data volume, and the WB data volume.

6. The prediction method according to any one of claims 1 to 5, wherein the syphilis feature vector is characterized by a direction quantity and a distribution rate; Optionally, the direction quantity is selected from the string "CIA+TPPA+WB+", "CIA+TPPA+WB-", "CIA+TPPA+WB±", "CIA+TPPA-WB+", "CIA+TPPA-WB-", "CIA+TPPA-WB±", "CIA-TPPA-WB+", "CIA-TPPA-WB-", "CIA-TPPA-WB±", "CIA+TPPA-RPR-MTPPA+", "CIA+TPPA-RPR-MTPPA-WB+", "CIA+TPPA-RPR-MTPPA-WB-", "CIA+TPPA-RPR-MTPPA-WB±", "CIA+TPPA-RPR-WB+", "CIA+TPPA-RPR-WB-MTPPA-", "CIA+TPPA-RPR-WB-MTPPA+", "CIA+TPPA-RPR-WB±MTPPA-", "CIA+TPPA-RPR-WB±MTPPA+" means; Optionally, the distribution ratio is selected from "CIA+TPPA+WB+", "CIA+TPPA+WB-", "CIA+TPPA+WB±", "CIA+TPPA-WB+", "CIA+TPPA-WB-", "CIA+TPPA-WB±", "CIA-TPPA-WB+", "CIA-TPPA-WB-", "CIA-TPPA-WB±", "CIA+TPPA-RPR-MTPPA+", "CIA+TPPA-RPR-MTPPA-WB+", "CIA+TPPA-RPR-MTPPA-WB-", "CIA+TPPA-RPR-MTPPA-WB±", The data volumes corresponding to "CIA+TPPA-RPR-WB+", "CIA+TPPA-RPR-WB-MTPPA-", "CIA+TPPA-RPR-WB-MTPPA+", "CIA+TPPA-RPR-WB±MTPPA-", and "CIA+TPPA-RPR-WB±MTPPA+" are expressed as the proportion of the clinical quantities.

7. The prediction method according to claim 1, wherein the step of participating in the machine learning of the clinical data terminal based on the local training set to obtain the disease prediction model comprises: Obtain the initial model parameters of the model to be trained from the preset server; Performing local training on the model to be trained according to the local training set and the initial model parameters to obtain model parameter update; The model parameter update is transmitted to the server end, so that the server end aggregates the model parameter update transmitted by the clinical data end to obtain the aggregated model parameter, and when it is detected that the model to be trained is in an unconverged state, the aggregated model parameter is returned to each of the clinical data ends to continue iterative training, until it is detected that the model to be trained is in a converged state, the aggregated model parameter is used as the final parameter of the model to be trained, and the disease prediction model is obtained, and the disease prediction model is transmitted to each of the clinical data ends; Receive the disease prediction model transmitted by the server.

8. A device for predicting syphilis, the device comprising: An acquisition unit, used to acquire syphilis input data and syphilis output data in a local database, wherein the syphilis input data includes CIA data, TPPA data and WB data, and the syphilis output data is the syphilis diagnosis and past medical history record data of clinical patients; An extraction unit, used for performing feature extraction on the syphilis input data to obtain a syphilis feature vector; A construction unit, used for constructing a local training set according to the syphilis feature vector and the syphilis output data; A training unit, used to participate in machine learning at the clinical data end based on the local training set to obtain a disease prediction model; An input unit, used for inputting the syphilis input data to be tested into the disease prediction model; A calculation unit, used for calculating a risk value according to the disease prediction model; An output unit is used to predict the risk of syphilis according to the risk value. A device for predicting syphilis, comprising: Memory; Processor; and A syphilis prediction program stored in the memory and executable on the processor, wherein when the prediction program is executed by the processor, the steps of the syphilis prediction method according to any one of claims 1 to 7 are implemented.

9. A computer-readable storage medium having stored thereon a syphilis prediction program, wherein when the syphilis prediction program is executed by a processor, the steps of the syphilis prediction method according to any one of claims 1 to 7 are implemented.

Images