Newborn disease screening and health management integrated platform

Abstract

The invention relates to the field of medical health, and discloses a neonatal disease screening and health management integrated platform, which comprises a screening management layer used for adopting a push algorithm based on collaborative filtering and deep learning user interest modeling fusion, learning a user behavior sequence by using an LSTM model, capturing user interest characteristics, and performing health management on the user behavior sequence; pushing the propaganda and education data; the collection interpretation layer is used for collecting multi-source data of neonatal disease screening and health management and transmitting and storing the multi-source data; and the dynamic quality control layer is used for preprocessing the multi-source data, constructing a fusion model based on an attention mechanism, and evaluating the performance of the fusion model through multiple indexes. Newborn multi-source data is collected from multiple terminals and preprocessed after encrypted transmission and storage, a fusion model based on an attention mechanism is constructed, potential association among multiple indexes is mined, the AI interpretation technology of multi-index association analysis is applied, disease risks are evaluated and diagnosed in combination with a disease knowledge base, and the accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the field of medical health technology, and in particular to an integrated platform for neonatal disease screening and health management. Background Art

[0002] In today's rapidly developing medical environment, the medical and health field continues to integrate advanced information technology to improve the quality and efficiency of medical services. Technologies such as big data, artificial intelligence, and the Internet of Things have gradually penetrated into all aspects of medical care, bringing new opportunities for disease prevention, diagnosis, and treatment. In the field of neonatal medical health, disease screening and health management are crucial. It can detect potential health problems of newborns at an early stage and intervene and treat them in a timely manner, which is of great significance to reducing neonatal mortality and improving the quality of the population.

[0003] Traditional platforms rely more on single indicators and the experience of medical staff when conducting disease diagnosis and risk assessment, resulting in a high rate of missed diagnosis for complex cases. Summary of the invention

[0004] In view of the shortcomings of the existing technology, the present invention provides an integrated platform for neonatal disease screening and health management, which solves the problem that traditional platforms rely on single indicators and the experience of medical staff when conducting disease diagnosis and risk assessment, resulting in a high rate of missed diagnosis of complex cases.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: a newborn disease screening and health management integrated platform, including:

[0006] Screening management layer: It is used to adopt a push algorithm based on the fusion of collaborative filtering and deep learning user interest modeling, use the LSTM model to learn the user behavior sequence, capture the user interest characteristics, and push the propaganda materials;

[0007] Collection and interpretation layer: used to collect multi-source data for neonatal disease screening and health management, and transmit and store them;

[0008] Dynamic quality control layer: used to pre-process multi-source data, build a fusion model based on the attention mechanism, and evaluate the performance of the fusion model through multiple indicators;

[0009] Diagnosis and treatment layer: used to conduct risk assessment and diagnosis of neonatal diseases based on the fusion model, give corresponding treatment recommendations, and conduct follow-up management;

[0010] Electronic display layer: used to display treatment recommendations for neonatal diseases and receive user feedback and input.

[0011] Preferably, the screening management layer includes an education management module, a learning module, and a data management module. The education management module is used to store education materials according to dimension classification, and adopts a push algorithm based on the fusion of collaborative filtering and deep learning user interest modeling to push education. The dimensions include disease type and screening stage. The learning module is used to use the LSTM model to learn user behavior sequences, explore user needs and interests, and optimize the education material push strategy. The push strategy includes adjusting the push time, frequency and content. The data management module is used to manage the various data generated, including the updating of education materials and the storage and analysis of user behavior data.

[0012] Preferably, the push algorithm is: ,in is the prediction score obtained by the collaborative filtering algorithm, For the prediction score obtained based on the LSTM model, is the fusion coefficient, and the calculation formula of the LSTM model is as follows:

[0013] Forget Gate ,in is the value of the forget gate, is the Sigmoid function, is the forget gate weight matrix, is the concatenation vector of the previous hidden state and the current input, is the forget gate bias vector;

[0014] Input Gate ,in, is the value of the input gate, is the input gate weight matrix, is the input gate bias vector;

[0015] Candidate memory cells ,in, is a candidate memory unit, is the hyperbolic tangent function, is the candidate memory unit weight matrix, is the candidate memory unit bias vector;

[0016] Memory Unit ,in, is the current moment memory unit, It is the memory unit of the previous moment;

[0017] Output Gate ,in, is the value of the output gate, is the output gate weight matrix, is the output gate bias vector;

[0018] Output ,in, Output for the current moment.

[0019] Preferably, the collection and interpretation layer includes an acquisition and transmission module and a storage security module. The acquisition and transmission module is used to collect multi-source data of neonatal disease screening and health management from multiple terminals and encrypt and transmit them. The multiple terminals include medical institution information systems, laboratory testing equipment, and mobile terminals. The multi-source data includes basic information of the newborn, genetic information, test data, imaging materials, and parent feedback information. The storage security module is used to store the multi-source data in a manner combining a distributed database and a file system, wherein structured data is stored in a distributed relational database, and unstructured data is stored in a distributed file system. Corresponding technologies are used to encrypt the stored multi-source data, and the corresponding technologies include data encryption, identity authentication, access control, and security auditing.

[0020] Preferably, the dynamic quality control layer includes a data processing module, a model building module, and an evaluation module. The data processing module is used to pre-process the relevant data using intelligent sensor technology and edge computing technology to obtain processed data. The preprocessing includes cleaning, conversion, normalization, and smoothing. The model building module is used to construct a fusion model based on the long short-term memory network and the convolutional neural network based on the attention mechanism according to the processed data. The evaluation module is used to evaluate the performance of the fusion model through multiple indicators, and the multiple indicators include accuracy, recall rate, F1 value, AUC-ROC, AUC-PR, and the performance includes accuracy, stability, and generalization ability.

[0021] Preferably, the diagnosis and treatment layer includes a risk assessment module, a diagnosis module, a treatment recommendation module, and a follow-up management module. The risk assessment module is used to use AI interpretation technology of multi-indicator association analysis to conduct risk assessment of neonatal diseases according to a fusion model, and use a deep learning algorithm to analyze multiple indicators, combined with a disease knowledge base, to assess the risk probability of neonatal genetic metabolic diseases, and obtain an assessment result. The diagnosis module is used to use a combination of rule-based reasoning methods and machine learning classification algorithms to diagnose neonatal diseases according to the assessment results and processed data to obtain a diagnosis result. The treatment recommendation module is used to give corresponding treatment recommendations based on the diagnosis results, by querying the treatment plan knowledge base, and combining expert experience and clinical guidelines. The follow-up management module is used to collaborate with the mobile terminal and the PC terminal to formulate a follow-up plan, conduct regular follow-ups, record follow-up data, analyze changes in the health status of the newborn, and adjust the treatment recommendations according to changes in health status.

[0022] Preferably, the electronic display layer includes an information display module, an interaction module, a feedback processing module, and a management module. The information display module is used to use visualization technology to display treatment recommendations for neonatal diseases to users. The visualization technology includes bar charts, line charts, and pie charts. The interaction module is used to receive user feedback and input, and use natural language processing technology to understand and analyze the text input by the user. The feedback processing module is used to process user feedback and promptly feedback the processing results to the user. The management module is used to provide management functions through collaboration between the mobile terminal and the PC terminal. The management functions include sample tracking, intelligent follow-up, and structured medical records.

[0023] A method for neonatal disease screening and health management comprises the following steps:

[0024] S1. Screening management: Adopt a push algorithm based on collaborative filtering and deep learning user interest modeling, use the LSTM model to learn user behavior sequences, capture user interest characteristics, and push educational materials;

[0025] S2. Collection and interpretation: Collect multi-source data on neonatal disease screening and health management, and transmit and store them;

[0026] S3, dynamic quality control: pre-process multi-source data, build a fusion model based on the attention mechanism, and evaluate the performance of the fusion model through multiple indicators;

[0027] S4. Diagnosis and treatment: Risk assessment and diagnosis of neonatal diseases are carried out based on the fusion model, and corresponding treatment recommendations are given and follow-up management is carried out;

[0028] S5. Electronic display: Display treatment recommendations for neonatal diseases and receive user feedback and input.

[0029] The present invention provides an integrated platform for neonatal disease screening and health management. It has the following beneficial effects:

[0030] 1. The present invention collects multi-source data of newborns from multiple terminals, performs pre-processing after encrypted transmission and storage, and constructs a fusion model based on the attention mechanism to explore the potential correlation between multiple indicators. It uses AI interpretation technology of multi-indicator correlation analysis and combines the disease knowledge base to evaluate disease risks and make diagnoses to improve accuracy. During the follow-up process, the mobile terminal and the PC terminal collaborate to collect newborn health data, and feed back to the platform to optimize the fusion model and diagnostic strategy, re-evaluate and adjust treatment recommendations, form a continuously optimized system, effectively reduce the missed diagnosis rate of complex cases, and provide more reliable protection for the health of newborns.

[0031] 2. The present invention realizes the push of educational materials through a push algorithm based on collaborative filtering and deep learning, thereby improving parents' awareness and cooperation in disease screening, and the automated data collection and analysis process reduces manual operations and errors, improves detection efficiency, and in the diagnosis and treatment stage, fast and accurate diagnosis results help to rationally allocate medical resources, avoid unnecessary examinations and treatments, and improve the utilization efficiency of medical resources.

[0032] 3. The present invention uses the fusion algorithm and LSTM model to push educational materials through the screening management layer to improve parental cooperation. The collection and interpretation layer collects multi-source data of newborns from multiple terminals. The dynamic quality control layer builds a fusion model to evaluate the data. The diagnosis and treatment layer conducts risk assessment, diagnosis and gives treatment suggestions based on this. The electronic display layer displays information and receives feedback, thus using data as a bridge and cooperating with each other to form a continuously optimized and continuously circulating closed loop for the entire newborn disease screening and health management process, comprehensively protecting the health of newborns and improving the disease prevention and control effects and health management levels. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 This is an architecture diagram of an integrated platform for neonatal disease screening and health management proposed by the present invention;

[0034] Figure 2 This is a method flow chart of a newborn disease screening and health management method proposed by the present invention. DETAILED DESCRIPTION

[0035] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings 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.

[0036] Please refer to the attached Figure 1 The embodiment of the present invention provides an integrated platform for neonatal disease screening and health management, including:

[0037] Screening management layer: It is used to adopt a push algorithm based on the fusion of collaborative filtering and deep learning user interest modeling, use the LSTM model to learn the user behavior sequence, capture the user interest characteristics, and push the education materials; the screening management layer includes the education management module, the learning module, and the data management module. The education management module is used to store the education materials according to the dimensions, and adopts a push algorithm based on the fusion of collaborative filtering and deep learning user interest modeling for education push. The dimensions include disease type and screening stage. The learning module is used to use the LSTM model to learn the user behavior sequence, explore user needs and interests, and optimize the education material push strategy. The push strategy includes adjusting the push time, frequency and content. The data management module is used to manage the various data generated, including the update of education materials and the storage and analysis of user behavior data.

[0038] The push algorithm is ,in is the prediction score obtained by the collaborative filtering algorithm, is the prediction score obtained based on the LSTM model, is the fusion coefficient, and the calculation formula of the LSTM model is as follows:

[0039] Forget Gate ,in is the value of the forget gate, is the Sigmoid function, is the forget gate weight matrix, is the concatenation vector of the previous hidden state and the current input, is the forget gate bias vector;

[0040] Input Gate ,in, is the value of the input gate, is the input gate weight matrix, is the input gate bias vector;

[0041] Candidate memory cells ,in, is a candidate memory unit, is the hyperbolic tangent function, is the candidate memory unit weight matrix, is the candidate memory unit bias vector;

[0042] Memory Unit ,in, is the current moment memory unit, Memory unit for the previous moment;

[0043] Output Gate ,in, is the value of the output gate, is the output gate weight matrix, is the output gate bias vector;

[0044] Output ,in, Output for the current moment.

[0045] Specifically, by collecting educational materials on various types of neonatal disease screening and health management, such as popular science articles on different genetic metabolic diseases, screening process animations, health care videos, etc. When storing, classify them according to disease type, such as phenylketonuria, congenital hypothyroidism, etc., and then further subdivide the storage according to the screening stage, such as preparation before blood collection, interpretation of test results, etc. When the user (parent or medical staff) logs in to the platform, the learning module starts working. It will obtain the user's browsing history, click records, dwell time and other behavioral data on the platform. Use this data as input for learning through the LSTM model. Forget gate in the LSTM model The degree of retention of the memory unit information at the previous moment will be determined based on the input data and the hidden state at the previous moment. For example, if the user has recently frequently browsed materials about hearing screening, the forget gate may retain more memory information related to hearing screening to prevent the model from forgetting important content. The input gate controls the degree to which the current input data enters the memory unit, and the new browsing behavior data will be fused with the memory unit through the input gate. Generates new information that may be added to the memory cell, which works with the forget gate and input gate to update the memory cell Finally, through the output gate and the current memory cell Get the output , which represents the model's "understanding" of the user's current interest status. The learning module analyzes the user's interest in different educational content based on the output of the LSTM model. When it is found that the user pays more attention to content related to the early symptoms of a certain disease, the data management module will mark the information and feed it back to the education management module. The education management module adjusts the fusion coefficient in the push algorithm accordingly. When users are more interested in content-based personalized recommendations (i.e., the LSTM model prediction score better reflects the user's interest), the value of The weight of The value of , so that the pushed content is more in line with user needs. At the same time, the data management module will regularly update the education materials. For example, when new research results or diagnosis and treatment guidelines are released, the old education content will be replaced in time to ensure that the information obtained by users is accurate and up-to-date.

[0046] Collection and interpretation layer: used to collect multi-source data for neonatal disease screening and health management, and transmit and store them; the collection and interpretation layer includes an acquisition and transmission module and a storage security module. The acquisition and transmission module is used to collect multi-source data for neonatal disease screening and health management from multiple terminals and encrypt and transmit them. The multiple terminals include medical institution information systems, laboratory testing equipment, and mobile terminals. Multi-source data includes basic information of the newborn, genetic information, test data, imaging materials, and parent feedback information. The storage security module is used to store multi-source data in a combination of distributed databases and file systems. Among them, structured data is stored in a distributed relational database, and unstructured data is stored in a distributed file system. Corresponding technologies are used to encrypt the stored multi-source data. The corresponding technologies include data encryption, identity authentication, access control, and security auditing.

[0047] Specifically, the acquisition and transmission module is connected to the medical institution information system through a standard data interface, such as the HL7 (Health Level Seven) standard interface, to ensure that the basic information of the newborn can be accurately obtained, such as name, gender, date of birth, place of birth, etc., as well as past medical records and family medical history and other related information. For laboratory testing equipment, data acquisition programs are performed according to the interface types and data formats of different devices. For example, a fully automatic biochemical analyzer may use serial communication to read the blood biochemical index data detected by the device by writing an adapter program, such as blood sugar, bilirubin and other test data. For devices with digital imaging functions, such as hearing screeners and fundus cameras, the image data generated by them is collected. The mobile terminal collects data through applications. Parents can fill in feedback information such as the daily performance and feeding status of the newborn on the application, and can also upload photos or videos of the newborn as supplementary information. During the transmission process, the collected data is encrypted using the SSL / TLS (Secure Sockets Layer / Transport Layer Security) encryption protocol to prevent the data from being stolen or tampered with during network transmission. The storage security module stores structured data, such as basic information of newborns and test results, in a distributed relational database such as CockroachDB, which has high scalability and strong consistency. It stores data on multiple nodes through distributed storage technology to improve data availability and fault tolerance. Unstructured data, such as imaging materials and text feedback information uploaded by parents, are stored in distributed file systems such as Ceph. Ceph uses the CRUSH (Controlled Replication Under Scalable Hashing) algorithm to achieve efficient data storage and retrieval. To ensure data security, the AES (Advanced Encryption Standard) encryption algorithm is used to encrypt the stored data to ensure the security of the data on the storage medium. At the same time, by setting up complex identity authentication mechanisms, such as multi-factor identity authentication (password, SMS verification code, fingerprint recognition, etc.), only authorized users can access the data. The role-based access control (RBAC) policy assigns different access rights according to the user's role, such as doctor, nurse, parent, etc. For example, a doctor can view and modify a patient's medical record data, while a parent can only view some information related to his or her child. The security audit function records all access operations to data, including access time, visitor identity, operation content, etc., so as to track and analyze when security issues arise.

[0048] Dynamic quality control layer: used to preprocess multi-source data, build a fusion model based on the attention mechanism, and evaluate the performance of the fusion model through multiple indicators; the dynamic quality control layer includes a data processing module, a model building module, and an evaluation module. The data processing module is used to preprocess relevant data using intelligent sensor technology and edge computing technology to obtain processed data. Preprocessing includes cleaning, conversion, normalization, and smoothing. The model building module is used to build a fusion model based on the attention mechanism and the long short-term memory network and the convolutional neural network according to the processed data. The evaluation module is used to evaluate the performance of the fusion model through multiple indicators, including accuracy, recall rate, F1 value, AUC-ROC, AUC-PR, and performance includes accuracy, stability, and generalization ability.

[0049] Specifically, the data processing module deploys sensors on the laboratory testing equipment to collect the operating data of the equipment, such as temperature, humidity, instrument calibration status, etc., and can also monitor the key data indicators in the testing process in real time. For example, in blood testing equipment, sensors can monitor the absorbance change curve of the sample. Using edge computing technology, the raw data is preliminarily processed at the data collection site. The cleaning operation identifies and removes duplicate values, error values, and outliers in the data. For example, when a value that is obviously deviated from the normal range and does not conform to physiological logic appears in the test data, such as a newborn's blood sugar value of 0 or too high to an unreasonable degree, it will be marked as an outlier and removed. The conversion operation is mainly to unify the data format for subsequent processing. Normalization processing maps the data to a specific interval, such as [0,1], so that data with different characteristics are comparable. Smoothing processing uses algorithms such as sliding average filtering to smooth data with large fluctuations, remove noise interference, and improve data stability. The model building module constructs an LSTM-CNN fusion model based on the attention mechanism based on the processed data. The LSTM part can process the time series characteristics of the data, such as analyzing the growth index change trend of newborns over a period of time. The CNN part is good at extracting local features of data, such as extracting lesion features in imaging data. The attention mechanism assigns weights according to the importance of the data, allowing the model to pay more attention to key information. For example, when analyzing newborn genetic test data, the attention mechanism will highlight the key gene site information related to genetic metabolic diseases. The evaluation module plays an important role in the model training and application process. During the training phase, a large amount of sample data with known results will be used to train the model. After the training is completed, the various evaluation indicators of the model are calculated using the test data set. The accuracy rate is used to measure the proportion of samples correctly predicted by the model; the recall rate focuses on the proportion of samples that are actually positive and correctly predicted; the F1 value is the harmonic mean of the accuracy rate and the recall rate, which comprehensively reflects the performance of the model. AUC-ROC (area under the receiver operating characteristic curve) evaluates the performance of the model at different classification thresholds by plotting the true positive rate and false positive rate curves at different thresholds. The closer the AUC-ROC value is to 1, the better the classification performance of the model. AUC-PR (area under the precision-recall curve) focuses more on evaluating the performance of the model in predicting positive samples. Through these indicators, the accuracy, stability and generalization ability of the model are comprehensively evaluated. When the model performance does not meet the preset standards, the model is adjusted and optimized.

[0050] Diagnosis and treatment layer: used to conduct risk assessment and diagnosis of neonatal diseases according to the fusion model, and give corresponding treatment suggestions and conduct follow-up management; the diagnosis and treatment layer includes risk assessment module, diagnosis module, treatment suggestion module and follow-up management module. The risk assessment module is used to adopt AI interpretation technology of multi-indicator association analysis, conduct risk assessment of neonatal diseases according to the fusion model, and use deep learning algorithm to analyze multiple indicators, and combine with disease knowledge base to assess the risk probability of neonates suffering from genetic metabolic diseases and obtain assessment results. The diagnosis module is used to diagnose neonatal diseases according to the assessment results and processed data by combining rule-based reasoning method and machine learning classification algorithm, and obtain diagnosis results. The treatment suggestion module is used to give corresponding treatment suggestions according to the diagnosis results, by querying the treatment plan knowledge base, and combining expert experience and clinical guidelines. The follow-up management module is used to collaborate with the mobile terminal and PC terminal to formulate a follow-up plan according to the diagnosis results and treatment suggestions of the newborn, and conduct regular follow-up, record follow-up data, analyze changes in the health status of the newborn, and adjust treatment suggestions according to changes in health status.

[0051] Specifically, the risk assessment module uses the data processed by the dynamic quality control layer and the fusion model as the basis, and uses deep learning algorithms, such as deep neural networks (DNN), to conduct a comprehensive analysis of multiple indicators such as genetic information and test data of newborns. For example, the gene sequencing data of newborns, metabolite concentration data in blood tests, clinical symptom information, etc. are used as input and input into the trained model. The model can discover the potential correlation between different indicators by learning a large amount of historical data. Combined with the disease knowledge base, which contains information such as the pathogenesis, symptoms, and diagnostic criteria of various genetic metabolic diseases, the risk probability of newborns suffering from specific genetic metabolic diseases is calculated. For example, for phenylketonuria, the model will comprehensively consider the test results of relevant mutation sites in the gene, the concentration of phenylalanine in the blood, and other factors to give a risk probability value. Based on risk assessment, the diagnosis module adopts a rule-based reasoning method to first make a preliminary judgment on the risk assessment results and processed data according to the existing clinical diagnostic standards and rules. For example, if some test indicators of the newborn exceed the normal range and meet the typical characteristics of a certain disease, it is preliminarily judged that the newborn may have the disease. Then, the preliminary judgment results are further verified and refined using machine learning classification algorithms, such as support vector machines (SVM) and random forests. Multiple data features are input into these algorithm models, and the final diagnosis results are obtained through the classification decision of the model. After obtaining the diagnosis results, the treatment recommendation module queries the treatment plan knowledge base, which collects standard treatment plans for various diseases, the latest research results, and the efficacy data of different treatment methods. Combined with expert experience and clinical guidelines, personalized treatment recommendations are formulated for newborns. For example, for newborns diagnosed with congenital hypothyroidism, appropriate thyroid hormone replacement therapy doses and monitoring plans are given according to the severity of their specific illness, age and other factors. The follow-up management module works collaboratively through the mobile terminal and the PC terminal. The doctor formulates a follow-up plan on the PC terminal based on the diagnosis results and treatment recommendations, including the follow-up time interval (such as weekly, monthly or quarterly) and the follow-up items (such as physical examination, laboratory tests, etc.). Parents receive follow-up reminders through the mobile terminal and record the daily health status of the newborn as required, such as height, weight, diet, medication reaction and other information. These follow-up data will be uploaded to the platform, and doctors will analyze them on the PC to observe changes in the health status of the newborn. If it is found that the growth and development indicators of the newborn do not meet expectations, or new symptoms appear, the treatment recommendations will be adjusted in time to ensure the effectiveness and safety of the treatment.

[0052] Electronic display layer: used to display treatment recommendations for neonatal diseases and receive user feedback and input. The electronic display layer includes an information display module, an interaction module, a feedback processing module, and a management module. The information display module uses visualization technology to display treatment recommendations for neonatal diseases to users. Visualization technology includes bar charts, line charts, and pie charts. The interaction module is used to receive user feedback and input, and uses natural language processing technology to understand and analyze the text input by users. The feedback processing module is used to process user feedback and promptly feedback the processing results to users. The management module is used to provide management functions through collaboration between the mobile terminal and the PC terminal. The management functions include sample tracking, intelligent follow-up, and structured medical records.

[0053] Specifically, the information display module presents the treatment plan generated by the treatment suggestion module to the user in an intuitive and visual way. For the drug treatment part, a bar chart is used to show the dosage and frequency comparison of different drugs, so that parents can clearly understand the use of each drug. A line chart is used to show the changes in the growth and development indicators of newborns during the treatment process, such as the growth trend of height and weight over time, so that parents and medical staff can intuitively see the treatment effect. The proportion of disease-related factors, such as genetic factors and environmental factors of a certain disease, is displayed through a pie chart to help users quickly understand the influencing factors of the disease. The interactive module uses natural language processing technology such as word segmentation, part-of-speech tagging, and named entity recognition to analyze the text entered by the user. For example, a parent enters "My baby's milk intake has decreased recently, what should I do?" on the mobile terminal. The interactive module first performs word segmentation and splits the sentence into words such as "baby", "recently", "milk intake", "reduced", and "what to do". Then, the part of speech of each word is determined through part-of-speech tagging, and then the named entity recognition technology is used to identify that "reduced milk intake" is key information related to the health of newborns. Based on these analysis results, the question is passed to the feedback processing module. After receiving the question, the feedback processing module will assign it to the corresponding professional for processing according to the type of question and the content involved. If it is a common question, the answer may be directly obtained from the preset question library and fed back to the user; if it is a complex question, it will be transferred to the doctor or expert for answering. After the answer is completed, it will be fed back to the user in a timely manner through the mobile terminal or PC terminal. The management module realizes the sample tracking function through the mobile terminal and the PC terminal. When the sample is collected, a unique identification code is given to the sample. By scanning the identification code, parents and medical staff can view the transportation status, testing progress and other information of the sample in real time on the mobile terminal or PC terminal. The intelligent follow-up function will automatically send follow-up reminders to parents on the mobile terminal according to the plan formulated by the follow-up management module. The reminder content includes the time, items and precautions of the follow-up. For structured medical records, internationally accepted medical data standards such as HL7FHIR (Fast Healthcare Interoperability Resources) are adopted to store and manage all health-related information of newborns in a standard format. Medical staff can easily query, modify and share medical records on the PC terminal to improve the efficiency and quality of medical services.

[0054] Please refer to the attached Figure 2 , a method for neonatal disease screening and health management, comprising the following steps:

[0055] S1. Screening management: Adopt a push algorithm based on collaborative filtering and deep learning user interest modeling, use the LSTM model to learn user behavior sequences, capture user interest characteristics, and push educational materials;

[0056] S2. Collection and interpretation: Collect multi-source data on neonatal disease screening and health management, and transmit and store them;

[0057] S3, dynamic quality control: pre-process multi-source data, build a fusion model based on the attention mechanism, and evaluate the performance of the fusion model through multiple indicators;

[0058] S4. Diagnosis and treatment: Risk assessment and diagnosis of neonatal diseases are carried out based on the fusion model, and corresponding treatment recommendations are given and follow-up management is carried out;

[0059] S5. Electronic display: Display treatment recommendations for neonatal diseases and receive user feedback and input.

[0060] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An integrated platform for neonatal disease screening and health management, characterized in that: include: Screening management layer: It is used to adopt a push algorithm based on the fusion of collaborative filtering and deep learning user interest modeling, use the LSTM model to learn the user behavior sequence, capture the user interest characteristics, and push the propaganda materials; Collection and interpretation layer: used to collect multi-source data for neonatal disease screening and health management, and transmit and store them; Dynamic quality control layer: used to pre-process multi-source data, build a fusion model based on the attention mechanism, and evaluate the performance of the fusion model through multiple indicators; Diagnosis and treatment layer: used to conduct risk assessment and diagnosis of neonatal diseases based on the fusion model, give corresponding treatment recommendations, and conduct follow-up management; Electronic display layer: used to display treatment recommendations for neonatal diseases and receive user feedback and input.

2. The integrated platform for neonatal disease screening and health management according to claim 1, characterized in that: The screening management layer includes an education management module, a learning module, and a data management module. The education management module is used to store education materials according to dimension classification, and adopts a push algorithm based on the fusion of collaborative filtering and deep learning user interest modeling for education push. The dimensions include disease type and screening stage. The learning module is used to use the LSTM model to learn user behavior sequences, explore user needs and interests, and optimize the education material push strategy. The push strategy includes adjusting the push time, frequency and content. The data management module is used to manage the various data generated, including the update of education materials and the storage and analysis of user behavior data.

3. The integrated platform for neonatal disease screening and health management according to claim 2, characterized in that: The push algorithm is: ,in is the prediction score obtained by the collaborative filtering algorithm, For the prediction score obtained based on the LSTM model, is the fusion coefficient, and the calculation formula of the LSTM model is as follows: Forget Gate ,in is the value of the forget gate, is the Sigmoid function, is the forget gate weight matrix, is the concatenation vector of the previous hidden state and the current input, is the forget gate bias vector; Input Gate ,in, is the value of the input gate, is the input gate weight matrix, is the input gate bias vector; Candidate memory cells ,in, is a candidate memory unit, is the hyperbolic tangent function, is the candidate memory unit weight matrix, is the candidate memory unit bias vector; Memory Unit ,in, is the memory unit at the current moment, It is the memory unit of the previous moment; Output Gate ,in, is the value of the output gate, is the output gate weight matrix, is the output gate bias vector; Output ,in, Output for the current moment.

4. The integrated platform for neonatal disease screening and health management according to claim 1, characterized in that: The collection and interpretation layer includes an acquisition and transmission module and a storage security module. The acquisition and transmission module is used to collect multi-source data on neonatal disease screening and health management from multiple terminals and encrypt and transmit them. The multiple terminals include medical institution information systems, laboratory testing equipment, and mobile terminals. The multi-source data includes basic information, genetic information, test data, imaging materials, and parent feedback information of the newborn. The storage security module is used to store the multi-source data in a manner combining a distributed database and a file system, wherein structured data is stored in a distributed relational database, and unstructured data is stored in a distributed file system. The stored multi-source data is encrypted using corresponding technologies, and the corresponding technologies include data encryption, identity authentication, access control, and security auditing.

5. The integrated platform for neonatal disease screening and health management according to claim 1, characterized in that: The dynamic quality control layer includes a data processing module, a model building module, and an evaluation module. The data processing module is used to pre-process relevant data using intelligent sensor technology and edge computing technology to obtain processed data. The pre-processing includes cleaning, conversion, normalization, and smoothing. The model building module is used to construct a fusion model based on the long short-term memory network and the convolutional neural network based on the attention mechanism according to the processed data. The evaluation module is used to evaluate the performance of the fusion model through multiple indicators. The multiple indicators include accuracy, recall rate, F1 value, AUC-ROC, AUC-PR, and the performance includes accuracy, stability, and generalization ability.

6. The integrated platform for neonatal disease screening and health management according to claim 1, characterized in that: The diagnosis and treatment layer includes a risk assessment module, a diagnosis module, a treatment recommendation module, and a follow-up management module. The risk assessment module is used to use AI interpretation technology of multi-indicator association analysis to conduct risk assessment of neonatal diseases according to a fusion model, and use a deep learning algorithm to analyze multiple indicators, combined with a disease knowledge base, to assess the risk probability of neonatal genetic metabolic diseases, and obtain an assessment result. The diagnosis module is used to use a combination of rule-based reasoning methods and machine learning classification algorithms to diagnose neonatal diseases according to the assessment results and processed data to obtain a diagnosis result. The treatment recommendation module is used to give corresponding treatment recommendations based on the diagnosis results, by querying the treatment plan knowledge base, and combining expert experience and clinical guidelines. The follow-up management module is used to collaborate with the mobile terminal and the PC terminal to formulate a follow-up plan based on the diagnosis results and treatment recommendations of the newborn, and conduct regular follow-up, record follow-up data, analyze changes in the health status of the newborn, and adjust the treatment recommendations according to changes in health status.

7. The integrated platform for neonatal disease screening and health management according to claim 1, characterized in that: The electronic display layer includes an information display module, an interaction module, a feedback processing module, and a management module. The information display module is used to use visualization technology to display treatment recommendations for neonatal diseases to users. The visualization technology includes bar charts, line charts, and pie charts. The interaction module is used to receive user feedback and input, and use natural language processing technology to understand and analyze the text input by the user. The feedback processing module is used to process user feedback and promptly feedback the processing results to the user. The management module is used to provide management functions through collaboration between the mobile terminal and the PC terminal. The management functions include sample tracking, intelligent follow-up, and structured medical records.

8. A method for neonatal disease screening and health management, characterized in that: A newborn disease screening and health management integrated platform as described in any one of claims 1 to 7, comprising the following steps: S1. Screening management: Adopt a push algorithm based on collaborative filtering and deep learning user interest modeling, use the LSTM model to learn user behavior sequences, capture user interest characteristics, and push educational materials; S2. Collection and interpretation: Collect multi-source data on neonatal disease screening and health management, and transmit and store them; S3, dynamic quality control: pre-process multi-source data, build a fusion model based on the attention mechanism, and evaluate the performance of the fusion model through multiple indicators; S4. Diagnosis and treatment: Risk assessment and diagnosis of neonatal diseases are carried out based on the fusion model, and corresponding treatment recommendations are given and follow-up management is carried out; S5. Electronic display: Display treatment recommendations for neonatal diseases and receive user feedback and input.

Images