Training method of intelligent medical decision-making model based on pediatric clinical practice

Abstract

The present invention relates to the field of data mining technology, and in particular to a training method for an intelligent medical decision model based on pediatric clinics. The method comprises the following steps: obtaining historical pediatric clinical data and performing desensitization to obtain historical pediatric clinical desensitization data; performing data fusion based on historical pediatric clinical desensitization data to obtain historical pediatric clinical fusion data; performing pediatric health care detection on historical pediatric clinical fusion data to obtain abnormal pediatric health care data; calculating health recovery impact factors of abnormal pediatric health care data to obtain health recovery impact factors; performing pediatrician recommendation priority division on historical pediatric clinical desensitization data to obtain pediatrician recommendation priority data; and constructing an intelligent pediatric medical decision model based on pediatrician recommendation priority data and health recovery impact factors. The present invention can improve the decision-making efficiency of pediatric medicine.

Description

Technical Field

[0001] The present invention relates to the technical field of data mining, and in particular to a training method for an intelligent medical decision-making model based on pediatric clinics. Background Art

[0002] With the continuous development of the medical field, especially the rapid progress of pediatric medicine, the clinical decision-making process has become increasingly complex. Pediatricians face many challenges, including diverse disease manifestations, individualized treatment needs, and a changing medical environment. Modern pediatric diagnosis and treatment often requires the cooperation of multiple disciplines, such as pediatric internal medicine, pediatric surgery, pediatric neurology, etc. Effective multidisciplinary collaboration is essential for the study of complex cases. Doctors need to coordinate the opinions of experts from all parties and comprehensively consider the suggestions of different disciplines. Pediatric patients may show diverse physiological characteristics and reactions due to their age, constitution, and developmental stage. This difference means that the same physiological phenomenon may have significantly different manifestations in newborns and school-age children. Therefore, medical researchers need to have extensive clinical experience and knowledge reserves to adapt to the changes in research scenarios brought about by these changes in physiological characteristics. Traditional methods face challenges in processing efficiency and accuracy when dealing with multidisciplinary collaboration and complex simulated clinical situations. Summary of the invention

[0003] Based on this, it is necessary for the present invention to provide a training method for an intelligent medical decision-making model based on pediatric clinical practice to solve at least one of the above technical problems.

[0004] To achieve the above purpose, a training method for an intelligent medical decision-making model based on pediatric clinical practice includes the following steps:

[0005] Step S1: Acquire historical pediatric clinical data, and desensitize the sensitive clinical patient information of the historical pediatric clinical data, thereby obtaining historical pediatric clinical desensitized data; fuse the pediatric disease classification clinical data according to the historical pediatric clinical desensitized data, thereby obtaining historical pediatric clinical fusion data;

[0006] Step S2: dividing the historical pediatric clinical fusion data into pediatric health care stages, thereby obtaining pediatric clinical health care stage data; performing abnormal health care detection based on the pediatric clinical health care stage data, thereby obtaining abnormal pediatric health care data;

[0007] Step S3: correcting the abnormal pediatric health care data for the impact of epidemic diseases, thereby obtaining abnormal pediatric health care corrected data, and calculating the health recovery impact factor based on the abnormal pediatric health care corrected data, thereby obtaining the health recovery impact factor;

[0008] Step S4: extracting clinician features from historical pediatric clinical desensitization data to obtain clinician data, and integrating pediatric clinical experience from the clinician data to obtain pediatrician clinical experience data; dividing pediatrician recommendation priorities based on health recovery influencing factors and pediatrician clinical experience data to obtain pediatrician recommendation priority data;

[0009] Step S5: Construct a pediatric clinician recommendation model based on the pediatrician recommendation priority data and health recovery influencing factors, and iterate and adjust the pediatric clinician recommendation model based on historical pediatric clinical desensitization data to obtain an intelligent pediatric medical decision-making model.

[0010] By acquiring a large amount of historical data, the present invention can ensure that the decision model has a sufficient sample size, thereby improving the representativeness and predictive ability of the model. Data desensitization processing can protect patient privacy and improve the security of data use. Data fusion helps to form a more comprehensive case perspective, making the understanding of different diseases more systematic, and contributing to more accurate diagnosis and treatment. The health status of case study patients is divided into stages, such as health examination, treatment process, recovery stage, etc. Identify abnormal conditions that occur during the nursing process, such as poor treatment effect or the emergence of new symptoms. By dividing the health care stage, the course of the case patient can be better studied and the treatment plan can be adjusted in time. Abnormal health care detection helps to identify problems in treatment early, intervene in time, and reduce the risk of worsening of the disease. Through the correction of the impact of epidemic diseases, it can ensure that the analysis results are not affected by the epidemic and improve the accuracy of model prediction. The calculation of the health recovery influencing factor helps to understand which factors have a significant impact on the recovery of case patients, so as to select a treatment research plan. Match the case according to the characteristics and experience of the doctor to ensure that the most suitable doctor is recommended to the case for research. Prioritizing the recommendation of experienced doctors can improve the scientificity and effectiveness of the treatment research plan. The constructed recommendation model can provide doctors with data-driven decision support and improve the scientificity and accuracy of decision-making. Through iteration and adjustment of the model, the performance and applicability of the model can be continuously improved to adapt to the ever-changing medical environment and case complexity. These steps can effectively improve the decision-making quality and efficiency of pediatric medicine and improve the treatment effect and experience of patients by systematically processing and analyzing pediatric clinical data, optimizing treatment plans and doctor recommendation processes. At the same time, with the support of intelligent models, personalized medical services can be achieved and the operational efficiency of the overall medical system can be improved.

[0011] Optionally, step S1 specifically includes:

[0012] Step S11: Acquire historical pediatric clinical data, and desensitize the historical pediatric clinical data to sensitive clinical patient information, thereby obtaining historical pediatric clinical desensitized data;

[0013] Step S12: classifying the historical pediatric clinical desensitization data into pediatric clinical disease categories, thereby obtaining historical pediatric clinical surgical data and historical pediatric clinical internal medicine data;

[0014] Step S13: performing pediatric clinical surgery data fusion according to historical pediatric clinical surgery data, thereby obtaining pediatric clinical surgery fusion data;

[0015] Step S14: performing pediatric clinical internal medicine data fusion according to historical pediatric clinical internal medicine data, thereby obtaining pediatric clinical internal medicine fusion data;

[0016] Step S15: Perform clinical data integration on the pediatric clinical surgery fusion data and the pediatric clinical internal medicine fusion data to obtain historical pediatric clinical fusion data.

[0017] The sensitive information desensitization of the present invention helps to protect the privacy of research patients while maintaining the analytical value of the data. This provides protection for the legal use of data and reduces the risk of data leakage. Classifying data into surgical and internal medicine categories helps to analyze and process data more finely. This can improve the efficiency of research on specific diseases and provide more targeted clinical research support for different departments. Surgical data fusion integrates information from different sources to make the data more comprehensive, thereby improving the accuracy and effectiveness of surgical treatment and research. Similarly, internal medicine data fusion brings together data from different sources to make the analysis of internal medicine diseases more comprehensive, which helps to improve the accuracy of internal medicine diagnosis and treatment. Integrating the fused data of surgery and internal medicine provides a comprehensive perspective to help comprehensively evaluate the overall health status of research patients, thereby optimizing overall clinical decisions and treatment plans.

[0018] Optionally, step S13 is specifically:

[0019] Step S131: extracting operation records from historical pediatric clinical surgical data to obtain pediatric clinical surgical operation data, and classifying the pediatric clinical surgical operation data by operation stage to obtain pre-operative surgical operation data, intra-operative surgical operation data, and post-operative surgical operation data;

[0020] Step S132: integrating the preoperative description features of the surgical target site on the preoperative data of the surgical operation, thereby obtaining the preoperative description data of the surgical target site, and integrating the preoperative target site features according to the preoperative description data of the surgical target site, thereby obtaining the preoperative target site data;

[0021] Step S133: extracting intraoperative monitoring data according to the intraoperative data of the surgical operation, thereby obtaining intraoperative monitoring data, and classifying the intraoperative monitoring data, thereby obtaining intraoperative monitoring images and intraoperative monitoring physiological coefficient data;

[0022] Step S134: dynamic structural modeling of the target part is performed according to the intraoperative monitoring image, thereby obtaining the dynamic structural model of the target part during the operation, and the dynamic structural model of the target part during the operation is filled with physiological coefficient data in time series according to the intraoperative monitoring physiological coefficient data, thereby updating the dynamic structural model of the target part during the operation;

[0023] Step S135: integrating the intraoperative target part features according to the intraoperative target part dynamic structure model, thereby obtaining intraoperative target part data;

[0024] Step S136: extracting a postoperative target part image from the postoperative surgical data to obtain a postoperative target part image, and integrating postoperative target part features based on the postoperative target part image to obtain postoperative target part data;

[0025] Step S137: Perform time series feature fusion on the preoperative target site data, the intraoperative target site data, and the postoperative target site data, so as to obtain pediatric clinical surgical fusion data.

[0026] The present invention can systematically organize the data and make it structured by extracting the surgical records of historical pediatric clinical surgical data, so as to facilitate subsequent analysis. The surgical data is divided into preoperative, intraoperative and postoperative stages, which is helpful for detailed analysis at each stage and improves the accuracy and operability of data analysis. By integrating the preoperative description features of the surgical target site, the state of the surgical area can be fully understood, which helps to formulate a more accurate surgical plan. The integration of preoperative target site data can provide more detailed and specific preoperative preparation information, and improve the accuracy and scientificity of preoperative preparation. Extracting and classifying intraoperative monitoring data can grasp the situation during the operation in real time, which is helpful to make timely adjustments and respond to emergencies. Classifying and acquiring intraoperative monitoring images and physiological coefficient data makes the data more comprehensive, which is helpful for subsequent comprehensive analysis and processing. Modeling the dynamic structure of the target site according to the intraoperative monitoring image can accurately reflect the actual changes of the target site during the operation and provide more realistic surgical information. By filling the physiological coefficient data in time series, the dynamic model is more perfect, and it can be effectively monitored and adjusted in real-time surgery. Integrating intraoperative target site data can provide more comprehensive and accurate information to help surgeons optimize surgical operations and decisions. By integrating the features of intraoperative data, the surgical effect and the status of the target site can be evaluated in real time, and the monitoring capability can be enhanced. Extracting images of the postoperative target site and integrating the features can help evaluate the surgical effect and ensure the achievement of the surgical goal. The integration of postoperative data provides a basis for follow-up tracking and processing, and helps to evaluate postoperative recovery and possible complications. By fusing the temporal features of preoperative, intraoperative, and postoperative data, a comprehensive view of pediatric clinical surgical data can be obtained, which helps to conduct more comprehensive analysis and research. Temporal feature fusion helps to ensure the coherence and consistency of the data and provide a comprehensive understanding of the entire surgical process.

[0027] Optionally, step S14 is specifically:

[0028] Step S141: extracting medical symptom summaries and symptom examination records from historical pediatric clinical medical data, thereby obtaining medical symptom summary data and symptom examination data;

[0029] Step S142: integrating the internal medicine symptom description features according to the internal medicine symptom summary data, thereby obtaining the internal medicine symptom description data;

[0030] Step S143: extracting inspection images according to the symptom inspection data to obtain symptom inspection images, and dividing the symptom inspection images into image frames and image audio to obtain symptom inspection image frames and symptom inspection audio;

[0031] Step S144: extracting physiological coefficient data features from the symptom examination data, thereby obtaining physiological coefficient data of internal medicine symptoms;

[0032] Step S145: selecting a target inspection part for the symptom inspection image frame according to the medical symptom description data, thereby obtaining a target inspection part image frame;

[0033] Step S146: constructing a three-dimensional structural model of the inspection part according to the inspection image frame of the target part, and fusing the time series structural features of the three-dimensional structural model of the inspection part according to the symptom inspection audio and the physiological coefficient data of the internal medicine symptom, so as to obtain the structural model of the internal medicine inspection part;

[0034] Step S147: Integrate the characteristics of the internal medicine examination part into the internal medicine examination part structure model, so as to obtain pediatric clinical internal medicine fusion data.

[0035] The present invention can systematically organize clinical data from different patients by extracting symptom summaries and symptom examination records, making the data more operational. Providing basic data for subsequent analysis, including symptom descriptions and examination records, is of great significance for model training and analysis. Integrating symptom summary features from different sources can identify and summarize key symptom features and help establish a more accurate symptom description model. By integrating description features, the patient's symptom manifestations can be more comprehensively understood, thereby improving the accuracy and comprehensiveness of diagnosis. Extracting and processing image data from symptom examination data helps to better understand and analyze image information and obtain more detailed examination data. Dividing image and audio data into frames and audio can perform more detailed analysis in time and space, improving the utilization efficiency and accuracy of data. Extracting physiological coefficient data helps to obtain detailed information about the patient's physiological state and provide more objective data support for symptom analysis and diagnosis. The extraction of physiological data features can provide more reference information for clinicians, thereby supporting more accurate diagnosis and personalized treatment plans. Selecting the target examination site according to the symptom description can improve the relevance and accuracy of the examination image and optimize the image analysis process. It can focus on specific examination sites, saving analysis time and resources, and improving the efficiency and accuracy of diagnosis. By constructing a three-dimensional structural model of the target site, the spatial structure of the examination site can be more intuitively understood, supporting more in-depth analysis and research. Fusion of the temporal characteristics of imaging and physiological data helps to comprehensively consider different types of information and improve the accuracy and comprehensiveness of the overall analysis. Comprehensive integration of all processed data to form pediatric clinical internal medicine fusion data helps to establish a comprehensive patient health record. By fusing data, a holistic view of the patient's health status can be provided, supporting comprehensive analysis and decision-making, and improving clinical diagnosis and treatment plans.

[0036] Optionally, step S2 specifically includes:

[0037] Step S21: extracting clinical treatment time series features and clinical symptom description features from historical pediatric clinical fusion data, thereby obtaining pediatric clinical treatment time series data and clinical symptom description data;

[0038] Step S22: sorting and dividing the description time according to the clinical symptom description data, thereby obtaining the initial symptom description data and the final symptom description data;

[0039] Step S23: Describe and time-series associate the pediatric clinical treatment time series data according to the initial symptom description data, so as to obtain the initial diagnosis and treatment time data; describe and time-series associate the pediatric clinical treatment time series data according to the final symptom description data, so as to obtain the review treatment time data;

[0040] Step S24: dividing the pediatric clinical treatment time series data into time windows according to the initial diagnosis and treatment time data and the reexamination and treatment time data, thereby obtaining pediatric treatment stage data and pediatric treatment observation stage data;

[0041] Step S25: integrating the historical pediatric clinical fusion data in time windows according to the pediatric treatment stage data and the pediatric treatment observation stage data, thereby obtaining pediatric clinical health care stage data;

[0042] Step S26: performing abnormal health care detection according to the pediatric clinical health care stage data, thereby obtaining abnormal pediatric health care data.

[0043] The present invention extracts key features from historical data to ensure that all relevant treatment and symptom information is captured. By extracting time series features, doctors can understand the time course of treatment, while symptom description features can help identify the manifestation of the disease. Extracting features makes the data more structured, which is helpful for subsequent data analysis and modeling. This can lay the foundation for feature matching and association analysis in subsequent steps. Time sorting of symptom description data helps to clearly distinguish the first and last symptom descriptions. In this way, the evolution of the disease can be clearly understood. By dividing the data into first and last descriptions, the development trend of the disease can be better analyzed, the effectiveness of treatment and the natural course of the disease can be identified. By analyzing the association between the first symptom description and the treatment time series, the treatment time after the first visit can be identified, and then the timeliness and effect of the treatment can be evaluated. The data of the last symptom description helps determine the review time and provides doctors with information on whether the condition is effectively controlled. This time series association analysis can evaluate the effectiveness of the treatment plan and optimize future treatment plans. By dividing the data of the first visit and the review time into windows, different treatment stages (such as acute stage, recovery stage, etc.) can be clarified to help formulate personalized treatment plans. Providing a clear time frame for the patient's continuous observation phase ensures that changes in the patient's health status are tracked and recorded, thereby improving the quality of long-term care. Integrating data from the treatment phase and the observation phase can obtain comprehensive health care phase data. This data integration helps to systematically evaluate the effectiveness of the entire treatment process and the patient's recovery status. By integrating data from different stages, caregivers can more accurately adjust care strategies to suit the actual needs of patients. Detecting abnormal health care data helps to detect potential health problems or improper care early, so that timely measures can be taken to intervene. Anomaly detection can help identify possible loopholes or problems in the care process, improve the overall quality of care, and reduce health risks for patients.

[0044] Optionally, step S26 is specifically:

[0045] Step S261: extracting nursing stage physiological coefficient data from pediatric clinical health nursing stage data, thereby obtaining nursing stage physiological coefficient data;

[0046] Step S262: performing frequency domain conversion according to the physiological coefficient data of the nursing stage, thereby obtaining a spectrum of the physiological coefficient of the nursing stage;

[0047] Step S263: extracting vibration amplitude features from the physiological coefficient spectrum during the nursing stage, thereby obtaining physiological coefficient vibration amplitude data, and performing physiological coefficient spectrum vibration amplitude threshold statistics based on the physiological coefficient vibration amplitude data, thereby obtaining a spectrum vibration amplitude threshold;

[0048] Step S264: performing abnormal vibration amplitude stage spectrum recognition on the nursing stage physiological coefficient spectrum according to the spectrum vibration amplitude threshold, thereby obtaining the abnormal vibration amplitude stage spectrum;

[0049] Step S265: performing time domain conversion on the abnormal vibration amplitude stage spectrum to obtain abnormal pediatric health care stage data, and performing initial diagnosis abnormal correction on the abnormal pediatric health care stage data to obtain abnormal pediatric health care data.

[0050] By extracting the physiological coefficient data of the nursing stage, the present invention can separate key physiological indicators from the overall clinical health care data and reduce data noise. Clearly extracting the physiological coefficients related to the pediatric nursing stage helps to more accurately assess the health status of children at a specific nursing stage. The extracted data is more targeted and convenient for further analysis and processing, such as frequency domain conversion and other steps. Frequency domain conversion (such as Fourier transform) can convert time domain signals into frequency domain signals, revealing periodicity and frequency components in the data, which may not be directly identified by time domain analysis. Through frequency domain conversion, the spectral features in the signal can be identified to help understand the performance of physiological data at different frequencies, so as to better identify normal and abnormal patterns. Frequency domain analysis can usually provide more detailed features, which helps to improve the analysis accuracy of physiological coefficients, especially when processing complex signals. By extracting the vibration amplitude features, the main components in the spectrum can be identified, which helps to understand the variation pattern of physiological coefficients and their amplitude changes. According to the vibration amplitude data, the spectrum vibration amplitude threshold is statistically calculated to help establish a benchmark and identify abnormal situations that deviate significantly from the normal range. The vibration amplitude threshold can be used to set an early warning mechanism to help detect possible health problems or abnormalities early. By identifying abnormal vibration amplitude phase spectra, abnormalities in health data can be discovered in a timely manner, which is crucial for early intervention and diagnosis. Effectively focusing on abnormal data reduces the amount of data that needs to be processed and improves diagnostic efficiency. Identifying abnormal vibration amplitude phase spectra helps provide personalized care plans for each child and improves care outcomes. Converting abnormal vibration amplitude phase spectra back to time domain data helps restore physiological data that is closer to reality, making it easier to understand and analyze. Through initial diagnosis abnormality correction, errors and abnormalities in the data can be corrected to provide more accurate health care data. Corrected abnormal pediatric health care data can provide more reliable information support for medical staff, allowing them to make more scientific diagnosis and treatment decisions.

[0051] Optionally, step S3 specifically includes:

[0052] Step S31: Obtain epidemic disease description data, and extract epidemic time features from the epidemic disease description data, thereby obtaining epidemic time data;

[0053] Step S32: performing time series correlation on the abnormal pediatric health care data according to the disease epidemic time data, thereby obtaining time-correlated abnormal pediatric health care data;

[0054] Step S33: extracting symptom description features from the time-related abnormal pediatric health care data, thereby obtaining time-related care symptom description data;

[0055] Step S34: Calculating the similarity of the symptom description according to the time-related nursing symptom description data and the epidemic disease description data, thereby obtaining the epidemic disease description similarity data, and identifying the epidemic disease nursing data for the time-related nursing symptom description data according to the epidemic disease description similarity data, thereby obtaining the epidemic disease nursing data;

[0056] Step S35: performing epidemic disease correction on the abnormal pediatric health care data according to the epidemic disease care data, thereby obtaining abnormal pediatric health care correction data;

[0057] Step S36: Calculate the health recovery impact factor based on the abnormal pediatric health care correction data to obtain the health recovery impact factor.

[0058] By extracting the time characteristics of epidemic diseases, the present invention can identify the epidemic patterns of diseases in different time periods, provide basic data for subsequent analysis, and help understand the seasonal or cyclical characteristics of diseases. By temporally associating abnormal pediatric health care data with disease epidemic time data, it is possible to identify anomalies in nursing data related to the time of epidemic diseases and improve the accuracy of anomaly detection. Extracting the symptom description features of time-related abnormal data helps to correspond abnormal data with specific disease symptoms and provide detailed information for subsequent similarity calculations. By calculating the similarity of symptom descriptions, the correlation between time-related nursing symptoms and epidemic diseases can be determined, which helps to identify which nursing data may involve epidemic diseases, thereby improving the accuracy of disease identification. Correcting abnormal pediatric health care data based on epidemic disease nursing data can reduce misdiagnosis or nursing omissions and improve nursing quality and accuracy. Calculating health recovery influencing factors helps to evaluate the impact of different correction measures on health recovery, provide a basis for optimizing health intervention strategies, and improve overall recovery effects.

[0059] Optionally, step S36 is specifically:

[0060] Step S361: extracting the heart rate characteristics of children and the body temperature characteristics of children from the historical pediatric clinical desensitization data, thereby obtaining the heart rate data and the body temperature data of children;

[0061] Step S362: constructing a health recovery degree evaluation system according to the heart rate data and the body temperature data of the child patient, and performing a health recovery degree evaluation on the heart rate data and the body temperature data of the child patient according to the health recovery degree evaluation system, thereby obtaining the health recovery degree data of the child patient;

[0062] Step S363: extracting the time distribution characteristics of the physiological coefficients of the child patient according to the abnormal pediatric health care correction data, thereby obtaining the physiological coefficient distribution data of the child patient;

[0063] Step S364: constructing a health recovery regression model according to the health recovery degree data of the child patient and the physiological coefficient distribution data of the child patient;

[0064] Step S365: Calculate the health recovery impact factor for the abnormal pediatric health care correction data according to the health recovery regression model, thereby obtaining the health recovery impact factor.

[0065] By extracting the heart rate and body temperature characteristics of child patients, the present invention can obtain key data about their physiological status, which helps doctors accurately monitor and analyze the health status of children. Establishing a health recovery assessment system and conducting an assessment enables the medical team to quantify the recovery process of child patients, thereby providing a scientific basis for further treatment and intervention. Extracting the time distribution characteristics of physiological coefficients helps understand the physiological change patterns of child patients, helps identify abnormal health conditions, and provides data support for targeted care. Constructing a health recovery regression model can combine heart rate and body temperature data with the distribution of physiological coefficients to provide a more comprehensive health recovery prediction and provide a basis for formulating personalized treatment plans. By calculating the health recovery influencing factors, the specific impact of different factors on children's health recovery can be quantified, the nursing intervention strategy can be optimized, and the overall treatment effect can be improved.

[0066] Optionally, step S4 is specifically:

[0067] Step S41: extracting clinician features from historical pediatric clinical desensitization data to obtain clinician data;

[0068] Step S42: extracting clinical diagnosis and treatment features from the clinical doctor data to obtain the doctor's clinical diagnosis and treatment data, and extracting diagnosis and treatment symptom descriptions from the doctor's clinical diagnosis and treatment data to obtain the doctor's diagnosis and treatment symptom description data;

[0069] Step S43: clustering the doctor's diagnosis and treatment symptom description data to obtain diagnosis and treatment symptom clustering data;

[0070] Step S44: performing diagnosis and treatment time distribution analysis based on the doctor's clinical diagnosis and treatment data to obtain diagnosis and treatment time distribution data, and performing symptom diagnosis and treatment time statistical integration based on the diagnosis and treatment symptom clustering data and the diagnosis and treatment time distribution data to obtain the pediatrician's clinical experience data;

[0071] Step S45: pediatrician recommendation priorities are divided according to health recovery influencing factors and pediatrician clinical experience data, thereby obtaining pediatrician recommendation priority data.

[0072] The present invention extracts clinical physician characteristics to help understand the physician's background and expertise in the field of pediatrics, laying the foundation for further data analysis. By extracting diagnosis and treatment characteristics and symptom description data, we can better understand the physician's diagnosis and treatment patterns and the way he or she handles specific symptoms, thereby obtaining detailed information about the physician's diagnosis and treatment. Cluster analysis of diagnosis and treatment symptoms can identify groups of similar symptoms, thereby helping to reveal patterns and trends in common symptoms and optimize symptom treatment strategies. By analyzing the distribution of diagnosis and treatment time and combining symptom clustering data, it is possible to generate information about the physician's time allocation for the treatment of different symptoms, providing data support for optimizing clinical processes. Prioritization based on health recovery influencing factors and the physician's clinical experience data can help recommend the most appropriate physician to the patient, thereby improving treatment outcomes and patient satisfaction.

[0073] Optionally, step S45 is specifically:

[0074] Step S451: performing principal component analysis on health recovery influencing factors to obtain health recovery principal component data;

[0075] Step S452: performing health recovery principal component proportion statistics on the doctor's diagnosis and treatment symptom description data according to the health recovery principal component data, thereby obtaining health recovery principal component proportion data;

[0076] Step S453: classifying the doctor diagnosis and treatment symptom description data by the principal component proportion according to the health recovery principal component proportion data, thereby obtaining the diagnosis and treatment doctor data with a high proportion and the diagnosis and treatment doctor data with a low proportion;

[0077] Step S454: dividing clinical experience doctors according to the clinical experience data of pediatricians, thereby obtaining data of highly experienced clinical doctors and data of low experienced clinical doctors;

[0078] Step S455: performing a clinician intersection operation on the data of high-proportion treating doctors and the data of highly experienced clinicians, so as to obtain data of pediatricians with high recommendation priority; performing a clinician intersection operation on the data of low-proportion treating doctors and the data of low-experience clinicians, so as to obtain data of pediatricians with low recommendation priority;

[0079] Step S456: Merge the data of pediatricians with high recommendation priority and the data of pediatricians with low recommendation priority to obtain pediatrician recommendation priority data.

[0080] The present invention performs principal component analysis (PCA) on health recovery influencing factors to reduce the dimension of the data and extract the main components. This makes it possible to identify the most important factors in complex health recovery influencing factor data, thereby simplifying the analysis process and improving the interpretability and predictive power of the model. According to the health recovery principal component data, the principal component proportion statistics of the doctor's diagnosis and treatment symptom description data can quantify the proportion of each principal component in the diagnosis and treatment symptom description. This helps to evaluate which principal components are more prominent in the doctor's diagnosis and treatment practice, thereby identifying which doctors may have a higher effect on health recovery. Doctors are classified by the principal component proportion, and the doctors are divided into high-proportion and low-proportion groups. This classification can highlight which doctors perform better on specific principal components, and then help screen out those doctors who may perform better in related health recovery fields. Doctors are divided according to clinical experience data, and the doctors are divided into high-experience and low-experience groups. This helps to distinguish experienced doctors from less experienced doctors, and use clinical experience as an important indicator to evaluate the overall diagnosis and treatment ability of doctors. By performing an intersection operation on the data of high-proportion treating doctors and high-experience clinicians, we can screen out doctors who perform well on key health recovery principal components and have rich experience. These doctors will be identified as high-recommendation priority doctors. Similarly, performing an intersection operation on the data of low-proportion and low-experience doctors can screen out doctors with low recommendation priorities. This step helps to formulate more accurate recommendation strategies based on comprehensive indicators. The data of high-recommendation priority doctors is combined with the data of low-recommendation priority doctors to form complete pediatrician recommendation priority data. This provides a comprehensive recommendation system that can systematically recommend the most suitable doctors based on different doctor characteristics and health recovery influencing factors, thereby improving patient treatment outcomes and satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments thereof made with reference to the following drawings:

[0082] Figure 1 It is a schematic diagram of the steps of the training method of the intelligent medical decision-making model based on pediatric clinical practice of the present invention;

[0083] Figure 2 Detailed step flow diagram of step S1 in the present invention;

[0084] Figure 3 Detailed step flow diagram of step S2 in the present invention;

[0085] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0086] The technical method of the present invention is described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are 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 technicians in this field without creative work are within the scope of protection of the present invention.

[0087] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0088] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0089] To achieve this, please refer to Figures 1 to 3 The present invention provides a training method for an intelligent medical decision-making model based on pediatric clinical practice, the method comprising the following steps:

[0090] Step S1: Acquire historical pediatric clinical data, and desensitize the sensitive clinical patient information of the historical pediatric clinical data, thereby obtaining historical pediatric clinical desensitized data; fuse the pediatric disease classification clinical data according to the historical pediatric clinical desensitized data, thereby obtaining historical pediatric clinical fusion data;

[0091] In this embodiment, pediatric medical record data from the past five years are extracted from the hospital electronic health record (EHR) system. These data include basic patient information (such as age, gender), medical records, laboratory test results, and doctor's diagnosis and treatment records. Use data desensitization tools to process sensitive information. For example, encrypt or remove personal identification information such as the patient's name, address, and ID number, and anonymize other information that may reveal the patient's identity (such as medical record number, contact information, etc.). The desensitized pediatric clinical data is integrated according to the disease classification. For example, all case data on "asthma" are merged to form a comprehensive data set, including diagnostic information, treatment plans, and outcome data for various types of asthma patients.

[0092] Step S2: dividing the historical pediatric clinical fusion data into pediatric health care stages, thereby obtaining pediatric clinical health care stage data; performing abnormal health care detection based on the pediatric clinical health care stage data, thereby obtaining abnormal pediatric health care data;

[0093] In this embodiment, based on the desensitization data, the data is divided into multiple stages according to the treatment process of pediatric patients, such as: initial diagnosis stage, acute stage, recovery stage, and follow-up stage. Machine learning algorithms (such as K-means clustering) are used to automatically group patient data. At each stage, anomaly detection algorithms (such as isolation forests or local outlier factors) are applied to identify abnormal data in the nursing process. For example, it is found that the recovery rate of some patients is significantly lower than the expected level, and they are marked as abnormal.

[0094] Step S3: correcting the abnormal pediatric health care data for the impact of epidemic diseases, thereby obtaining abnormal pediatric health care corrected data, and calculating the health recovery impact factor based on the abnormal pediatric health care corrected data, thereby obtaining the health recovery impact factor;

[0095] In this embodiment, abnormal health care data is analyzed to identify whether it is affected by epidemic diseases (such as influenza, RSV). These effects are corrected from the data using epidemiological models (such as regression analysis). Based on the corrected data, the factors that affect the patient's health recovery are calculated. For example, the effectiveness of the treatment plan, the patient's age, the severity of the disease, and other factors are analyzed to form a scoring system for health recovery factors.

[0096] Step S4: extracting clinician features from historical pediatric clinical desensitization data to obtain clinician data, and integrating pediatric clinical experience from the clinician data to obtain pediatrician clinical experience data; dividing pediatrician recommendation priorities based on health recovery influencing factors and pediatrician clinical experience data to obtain pediatrician recommendation priority data;

[0097] In this embodiment, characteristic information of doctors is extracted from historical data, such as: diagnosis and treatment records of each doctor, number of cases handled, treatment success rate, etc. Natural language processing (NLP) technology is used to extract keywords and frequency information from doctors' text records. The extracted features are analyzed and integrated, for example: each doctor's treatment plan and results are statistically summarized to form an experience database. Statistical analysis (such as principal component analysis) is used to determine which features are significantly associated with treatment success rate. Doctors are prioritized based on health recovery influencing factors and doctors' clinical experience data. For example, the recommendation priority of certain doctors is determined based on the number of successful cases they have in treating specific diseases.

[0098] Step S5: Construct a pediatric clinician recommendation model based on the pediatrician recommendation priority data and health recovery influencing factors, and iterate and adjust the pediatric clinician recommendation model based on historical pediatric clinical desensitization data to obtain an intelligent pediatric medical decision-making model.

[0099] In this embodiment, a pediatric clinical physician recommendation model is constructed using machine learning algorithms (such as random forests and support vector machines). The model input includes physician recommendation priority data, health recovery influencing factors, etc., and the output is the physician's recommendation order. The model is trained and validated using historical pediatric clinical desensitization data, and the model parameters are adjusted to improve accuracy. For example, the hyperparameters of the model are optimized through cross-validation to ensure the reliability and accuracy of the recommendation results. Based on the optimized model, an intelligent pediatric medical decision model is generated. The model can recommend the most suitable doctor to the patient in real time, and automatically adjust the recommendation strategy according to the patient's specific situation.

[0100] Optionally, step S1 specifically includes:

[0101] Step S11: Acquire historical pediatric clinical data, and desensitize the historical pediatric clinical data to sensitive clinical patient information, thereby obtaining historical pediatric clinical desensitized data;

[0102] In this embodiment, when obtaining historical pediatric clinical data, it is necessary to extract all relevant pediatric medical records from the electronic health record system, including the patient's age, gender, diagnosis results and treatment plan, etc. When desensitizing these data, data anonymization technology is required to encrypt or desensitize the patient's personal identity information (such as name, contact information, ID number, etc.). For example, the name can be replaced with "Patient A" and the phone number can be replaced with "-*" to generate historical pediatric clinical desensitization data.

[0103] Step S12: classifying the historical pediatric clinical desensitization data into pediatric clinical disease categories, thereby obtaining historical pediatric clinical surgical data and historical pediatric clinical internal medicine data;

[0104] In this embodiment, when classifying the disease categories of historical pediatric clinical desensitization data, it is necessary to apply machine learning classification algorithms (such as decision trees and random forests) to label the diseases in the desensitization data. In specific implementation, the data will be divided into two categories: "surgical diseases" and "internal medicine diseases". For example, cases with "appendicitis" and "cerebral palsy" are classified into historical pediatric clinical surgical data, while cases with "asthma" and "diabetes" are classified into historical pediatric clinical internal medicine data.

[0105] Step S13: performing pediatric clinical surgery data fusion according to historical pediatric clinical surgery data, thereby obtaining pediatric clinical surgery fusion data;

[0106] In this embodiment, when data fusion is performed based on historical pediatric clinical surgical data, data integration technology can be used to merge surgical data from multiple sources into a unified data set. For example, surgical case data from different hospitals are aggregated by patient ID, and the integrated data includes diagnostic information, surgical records, and rehabilitation status, thereby obtaining pediatric clinical surgical fusion data.

[0107] Step S14: performing pediatric clinical internal medicine data fusion according to historical pediatric clinical internal medicine data, thereby obtaining pediatric clinical internal medicine fusion data;

[0108] In this embodiment, when data fusion is performed based on pediatric clinical internal medicine, the internal medicine case data needs to be summarized and integrated. Data cleaning and standardization technology can be used to organize the internal medicine data of different hospitals in a unified standard format, for example, all symptoms of internal medicine diseases, examination results and other information are uniformly coded to generate pediatric clinical internal medicine fusion data.

[0109] Step S15: Perform clinical data integration on the pediatric clinical surgery fusion data and the pediatric clinical internal medicine fusion data to obtain historical pediatric clinical fusion data.

[0110] In this embodiment, when integrating pediatric clinical surgical fusion data and pediatric clinical internal medicine fusion data, it is necessary to use data fusion technology to merge the two types of fusion data. For example, by establishing a comprehensive data warehouse, surgical and internal medicine data are integrated according to patient ID to ensure that all information is consistent, and finally generate historical pediatric clinical fusion data for further clinical analysis and research.

[0111] Optionally, step S13 is specifically:

[0112] Step S131: extracting operation records from historical pediatric clinical surgical data to obtain pediatric clinical surgical operation data, and classifying the pediatric clinical surgical operation data by operation stage to obtain pre-operative surgical operation data, intra-operative surgical operation data, and post-operative surgical operation data;

[0113] In this embodiment, surgical record data is extracted from the electronic medical record system to screen out relevant pediatric clinical surgical records. These records are classified into preoperative, intraoperative and postoperative data. For example, preoperative data includes preoperative examination reports, patient medical history, etc.; intraoperative data includes physiological data and images recorded during the operation; and postoperative data includes postoperative recovery status and follow-up records.

[0114] Step S132: integrating the preoperative description features of the surgical target site on the preoperative data of the surgical operation, thereby obtaining the preoperative description data of the surgical target site, and integrating the preoperative target site features according to the preoperative description data of the surgical target site, thereby obtaining the preoperative target site data;

[0115] In this embodiment, the description of the surgical target part in the preoperative data is integrated. The target part information is extracted from the surgical plan and preoperative images using natural language processing technology. For example, by analyzing the CT scan report, the position, size and shape data of the target part are extracted to generate a detailed preoperative target part data record.

[0116] Step S133: extracting intraoperative monitoring data according to the intraoperative data of the surgical operation, thereby obtaining intraoperative monitoring data, and classifying the intraoperative monitoring data, thereby obtaining intraoperative monitoring images and intraoperative monitoring physiological coefficient data;

[0117] In this embodiment, monitoring data is extracted from intraoperative data, for example, physiological parameters such as heart rate and blood pressure obtained by real-time monitoring instruments. These data are classified, and monitoring images (such as surgical endoscope images) and physiological coefficient data (such as blood oxygen saturation) are stored separately.

[0118] Step S134: dynamic structural modeling of the target part is performed according to the intraoperative monitoring image, thereby obtaining the dynamic structural model of the target part during the operation, and the dynamic structural model of the target part during the operation is filled with physiological coefficient data in time series according to the intraoperative monitoring physiological coefficient data, thereby updating the dynamic structural model of the target part during the operation;

[0119] In this embodiment, a dynamic structural model of the target part is established based on the intraoperative monitoring images. For example, a dynamic structural model is formed by using the real-time intraoperative images for three-dimensional reconstruction. The model is filled in time series based on real-time physiological coefficient data (such as heart rate changes), and the dynamic structural model is updated to reflect the changes in the target part during the operation.

[0120] Step S135: integrating the intraoperative target part features according to the intraoperative target part dynamic structure model, thereby obtaining intraoperative target part data;

[0121] In this embodiment, the dynamic structure model of the intraoperative target part is compared with the real-time image data to integrate the characteristics of the intraoperative target part. For example, the model is matched with the real-time intraoperative image to extract specific intraoperative target part data and generate a complete intraoperative data record.

[0122] Step S136: extracting a postoperative target part image from the postoperative surgical data to obtain a postoperative target part image, and integrating postoperative target part features based on the postoperative target part image to obtain postoperative target part data;

[0123] In this embodiment, images in the postoperative data are extracted, for example, CT or MRI images from postoperative review, and these images are analyzed using image processing technology to extract postoperative target site features, such as the healing status of the surgical site, and generate postoperative target site data.

[0124] Step S137: Perform time series feature fusion on the preoperative target site data, the intraoperative target site data, and the postoperative target site data, so as to obtain pediatric clinical surgical fusion data.

[0125] In this embodiment, the preoperative target site data, the intraoperative target site data, and the postoperative target site data are subjected to time series feature fusion. For example, through the time series analysis method, the data at different stages are integrated to form a comprehensive pediatric clinical surgical fusion data set for evaluating the overall effect of the surgery.

[0126] Optionally, step S14 is specifically:

[0127] Step S141: extracting medical symptom summaries and symptom examination records from historical pediatric clinical medical data, thereby obtaining medical symptom summary data and symptom examination data;

[0128] In this embodiment, when processing historical pediatric clinical internal medicine data, it is first necessary to extract the medical symptom summary data and symptom examination records of each patient. This can be achieved through natural language processing technology, in which medical terminology libraries and disease classification systems (such as ICD-10) are used to automatically classify and summarize symptoms. For example, extract patient symptom descriptions such as "abdominal pain" and "fever" and associate their examination records, including laboratory results and physical examination data.

[0129] Step S142: integrating the internal medicine symptom description features according to the internal medicine symptom summary data, thereby obtaining the internal medicine symptom description data;

[0130] In this example, feature engineering technology is used to integrate symptom descriptions based on the medical symptom summary data. For example, "abdominal pain" and "vomiting" are converted into vectors through the word embedding model, and feature description data for each symptom is generated in combination with context information. These data help understand the severity of the symptoms and the possible causes described by the doctor.

[0131] Step S143: extracting inspection images according to the symptom inspection data to obtain symptom inspection images, and dividing the symptom inspection images into image frames and image audio to obtain symptom inspection image frames and symptom inspection audio;

[0132] In this embodiment, images, including X-rays or CT scans, are extracted from the symptom check data. The images are subjected to frame segmentation (using CT or MRI scanning technology to decompose the image sequence into individual frames) and audio segmentation (extracting sound data from the images) through image processing techniques. For example, CT scans are frame segmented to generate each layer of images, and relevant diagnostic audio records are extracted for subsequent analysis.

[0133] Step S144: extracting physiological coefficient data features from the symptom examination data, thereby obtaining physiological coefficient data of internal medicine symptoms;

[0134] In this embodiment, feature extraction is performed on physiological coefficients (such as heart rate, blood pressure, etc.) in the symptom examination data. This involves using statistical analysis tools and machine learning algorithms to calculate the mean, standard deviation and other features of each physiological parameter to generate detailed physiological coefficient data of internal medicine symptoms. For example, the heart rate change feature is extracted by analyzing electrocardiogram data.

[0135] Step S145: selecting a target inspection part for the symptom inspection image frame according to the medical symptom description data, thereby obtaining a target inspection part image frame;

[0136] In this embodiment, image frames of target parts are selected based on the medical symptom description data. By using the correspondence between symptoms and image data, such as abdominal pain may require attention to the abdominal area, frames of relevant parts are selected from the image sequence to focus on analyzing the detailed information of these key frames.

[0137] Step S146: constructing a three-dimensional structural model of the inspection part according to the inspection image frame of the target part, and fusing the time series structural features of the three-dimensional structural model of the inspection part according to the symptom inspection audio and the physiological coefficient data of the internal medicine symptom, so as to obtain the structural model of the internal medicine inspection part;

[0138] In this embodiment, a three-dimensional structural model of the inspection site is constructed, and the two-dimensional data of the image frame of the target site is converted into a three-dimensional model through image reconstruction technology (such as slice reconstruction). In addition, the symptom inspection audio and the physiological coefficient of the internal medical symptoms are fused with the three-dimensional structural model of the inspection site to generate a temporal structural feature model. For example, the audio of the symptom inspection is combined with the physiological coefficient data, and the audio features are extracted through signal processing technology (such as Fourier transform and time-frequency analysis), and then integrated with the three-dimensional structural model constructed based on the image data.

[0139] Step S147: Integrate the characteristics of the internal medicine examination part into the internal medicine examination part structure model, so as to obtain pediatric clinical internal medicine fusion data.

[0140] In this embodiment, the constructed three-dimensional structural model is feature integrated, for example, by analyzing the abnormal features displayed in the model through a machine learning algorithm. These features are integrated to generate pediatric clinical internal medicine fusion data to provide comprehensive diagnostic information and treatment recommendations. For example, the key features in the three-dimensional model are combined with clinical symptom data to form a complete pathology analysis report.

[0141] Optionally, step S2 specifically includes:

[0142] Step S21: extracting clinical treatment time series features and clinical symptom description features from historical pediatric clinical fusion data, thereby obtaining pediatric clinical treatment time series data and clinical symptom description data;

[0143] In this embodiment, clinical treatment time series features and clinical symptom description features are extracted from historical pediatric clinical fusion data. This can be achieved by analyzing the diagnosis and treatment records, examination reports, and medical record annotations of pediatric patients in the electronic health record (EHR). Natural language processing (NLP) technology is used to convert the doctor's description into standardized clinical symptom description data, and time series analysis is used to extract key time series features in the treatment process. These data include the time of onset of symptoms, treatment methods, drug dosages, and their corresponding effects.

[0144] Step S22: sorting and dividing the description time according to the clinical symptom description data, thereby obtaining the initial symptom description data and the final symptom description data;

[0145] In this embodiment, the extracted clinical symptom description data is time-sorted and divided. For example, each symptom description is sorted using a timestamp, the first symptom description is marked as the first symptom description, and the last symptom description is marked as the last symptom description. The sorting process can be implemented using the Pandas library in Python, sorting the symptom descriptions by timestamp, and using a threshold segmentation to distinguish between the first and last symptom descriptions.

[0146] Step S23: Describe and time-series associate the pediatric clinical treatment time series data according to the initial symptom description data, so as to obtain the initial diagnosis and treatment time data; describe and time-series associate the pediatric clinical treatment time series data according to the final symptom description data, so as to obtain the review treatment time data;

[0147] In this embodiment, the initial symptom description data and the pediatric clinical treatment time series data are used for time series correlation analysis to determine the time of initial diagnosis and treatment. This can be achieved through a time-based matching algorithm, such as dynamic time warping (DTW), to compare the time point of the symptom description with the time point of the treatment record. Similarly, the final symptom description data is also used to determine the review treatment time, which involves matching the review record with the final symptom description.

[0148] Step S24: dividing the pediatric clinical treatment time series data into time windows according to the initial diagnosis and treatment time data and the reexamination and treatment time data, thereby obtaining pediatric treatment stage data and pediatric treatment observation stage data;

[0149] In this embodiment, the pediatric clinical treatment time series data is divided into time windows according to the initial diagnosis and treatment time data and the reexamination treatment time data. For example, a time window is defined, such as the time period between the initial diagnosis and the reexamination as the treatment stage, and the time period after the reexamination as the observation stage. The data is divided into these stages using a time window division algorithm (such as a sliding window technique) for subsequent analysis.

[0150] Step S25: integrating the historical pediatric clinical fusion data in time windows according to the pediatric treatment stage data and the pediatric treatment observation stage data, thereby obtaining pediatric clinical health care stage data;

[0151] In this embodiment, the pediatric treatment stage data and the pediatric treatment observation stage data are integrated to obtain the pediatric clinical health care stage data. The integration process includes merging the various indicators, examination results and treatment records of the treatment stage and the observation stage to form a comprehensive nursing data set. Data fusion technology (such as data integration tools) is used to unify the data of different stages.

[0152] Step S26: performing abnormal health care detection according to the pediatric clinical health care stage data, thereby obtaining abnormal pediatric health care data.

[0153] In this embodiment, abnormal health care detection is performed using pediatric clinical health care stage data. For example, data points that deviate significantly from normal patterns during the care process are identified by setting a threshold or using an anomaly detection algorithm (such as an isolation forest algorithm). This may include abnormal test results, treatment effect deviations, or unexpected side effects, and an abnormal pediatric health care data report is generated.

[0154] Optionally, step S26 is specifically:

[0155] Step S261: extracting nursing stage physiological coefficient data from pediatric clinical health nursing stage data, thereby obtaining nursing stage physiological coefficient data;

[0156] In this embodiment, when extracting physiological coefficient data from the pediatric clinical health care stage data, sensors can be used to monitor the patient's heart rate, body temperature, respiratory rate and other indicators. Through the data acquisition system, these indicators are recorded during the care stage to form preliminary physiological coefficient data. These data need to be sorted in chronological order and ensure the integrity and accuracy of the data. During implementation, digital signal processing technology can be used to clean and standardize the raw data to reduce the impact of noise.

[0157] Step S262: performing frequency domain conversion according to the physiological coefficient data of the nursing stage, thereby obtaining a spectrum of the physiological coefficient of the nursing stage;

[0158] In this embodiment, when the frequency domain conversion is performed on the physiological coefficient data of the nursing stage, the fast Fourier transform (FFT) algorithm is used to convert the time domain signal into a frequency domain signal. The extracted physiological coefficient data can be divided into several data windows, a window function is applied to reduce the boundary effect, and then the FFT algorithm is applied to each window. Finally, the obtained spectrum data can be used for further feature analysis to identify potential abnormal patterns.

[0159] Step S263: extracting vibration amplitude features from the physiological coefficient spectrum during the nursing stage, thereby obtaining physiological coefficient vibration amplitude data, and performing physiological coefficient spectrum vibration amplitude threshold statistics based on the physiological coefficient vibration amplitude data, thereby obtaining a spectrum vibration amplitude threshold;

[0160] In this embodiment, when the vibration amplitude feature is extracted from the spectrum of the physiological coefficients in the nursing stage, the vibration amplitude of each frequency band in the spectrum is calculated. The amplitude spectrum of the Fourier analysis result can be used to represent the vibration intensity of each frequency band. Through statistical analysis, the frequency bands whose vibration amplitude exceeds the preset threshold are identified to determine the vibration amplitude threshold of the physiological coefficient spectrum. These thresholds can be used for subsequent abnormality detection.

[0161] Step S264: performing abnormal vibration amplitude stage spectrum recognition on the nursing stage physiological coefficient spectrum according to the spectrum vibration amplitude threshold, thereby obtaining the abnormal vibration amplitude stage spectrum;

[0162] In this embodiment, when the abnormal vibration amplitude stage spectrum of the nursing stage physiological coefficient spectrum is identified according to the spectrum vibration amplitude threshold, after setting the threshold, each frequency band in the spectrum data is scanned and the vibration amplitude exceeding the threshold is marked. The identified abnormal vibration amplitude stage spectrum can be screened by the set threshold to find potential abnormal fluctuations or data deviations. This helps to find physiological abnormalities or equipment failures.

[0163] Step S265: performing time domain conversion on the abnormal vibration amplitude stage spectrum to obtain abnormal pediatric health care stage data, and performing initial diagnosis abnormal correction on the abnormal pediatric health care stage data to obtain abnormal pediatric health care data.

[0164] In this embodiment, when the spectrum of the abnormal vibration amplitude stage is converted to the time domain, the frequency domain data is converted back to the time domain signal using an inverse Fourier transform (IFFT). This will provide physiological data for the abnormal care stage. When performing the initial diagnosis abnormal correction, by comparing the data difference before and after the correction, an algorithm is applied to correct the abnormal conditions generated during the initial diagnosis period, for example, by adjusting the abnormal fluctuations through filtering or data smoothing techniques, and finally generating accurate abnormal pediatric health care data. These corrected data are helpful for further diagnosis and intervention.

[0165] Optionally, step S3 specifically includes:

[0166] Step S31: Obtain epidemic disease description data, and extract epidemic time features from the epidemic disease description data, thereby obtaining epidemic time data;

[0167] In this embodiment, epidemic disease description data is obtained, such as extracting historical data of influenza and measles through the hospital electronic health record system or public health database. Then, statistical analysis methods, such as seasonal time series analysis, are used to extract the time characteristics of the epidemic of the disease, such as the peak of influenza in winter and the peak of measles in spring. These characteristics will become the basis of the time data of the epidemic of the disease.

[0168] Step S32: performing time series correlation on the abnormal pediatric health care data according to the disease epidemic time data, thereby obtaining time-correlated abnormal pediatric health care data;

[0169] In this embodiment, abnormal pediatric health care data (e.g., children with high fever, persistent cough, etc.) are sorted in chronological order, and time series analysis techniques, such as sequence pattern mining, are applied to identify the temporal associations of the abnormal data. For example, it is found that high fever is strongly associated with the flu season. These temporal associations will form temporally correlated abnormal pediatric health care data.

[0170] Step S33: extracting symptom description features from the time-related abnormal pediatric health care data, thereby obtaining time-related care symptom description data;

[0171] In this embodiment, the symptom description features are extracted from the time-related abnormal pediatric health care data, for example, the symptoms (high fever, cough) and their severity are extracted. This can be achieved through text analysis technology such as natural language processing (NLP). The extracted features form the time-related nursing symptom description data.

[0172] Step S34: Calculating the similarity of the symptom description according to the time-related nursing symptom description data and the epidemic disease description data, thereby obtaining the epidemic disease description similarity data, and identifying the epidemic disease nursing data for the time-related nursing symptom description data according to the epidemic disease description similarity data, thereby obtaining the epidemic disease nursing data;

[0173] In this embodiment, the similarity between the time-related nursing symptom description data and the epidemic disease description data is calculated. For example, the cosine similarity or Jaccard index is used to quantify the similarity. This can help identify whether the symptom is related to an epidemic disease. Based on the similarity data, the abnormal pediatric health care data is classified to identify the epidemic disease care data.

[0174] Step S35: performing epidemic disease correction on the abnormal pediatric health care data according to the epidemic disease care data, thereby obtaining abnormal pediatric health care correction data;

[0175] In this embodiment, the epidemic disease care data is used to correct the abnormal pediatric health care data. For example, the high fever data is adjusted based on the epidemic disease pattern to distinguish between influenza-related high fever and other diseases. The corrected data is called abnormal pediatric health care correction data, which is the disease data without the diseases caused by influenza.

[0176] Step S36: Calculate the health recovery impact factor based on the abnormal pediatric health care correction data to obtain the health recovery impact factor.

[0177] In this embodiment, the health recovery influencing factors are calculated based on the corrected data. For example, regression analysis is used to evaluate the effectiveness of influenza treatment to generate health recovery influencing factors. These factors can reveal the specific effects of various treatments on recovery.

[0178] Optionally, step S36 is specifically:

[0179] Step S361: extracting the heart rate characteristics of children and the body temperature characteristics of children from the historical pediatric clinical desensitization data, thereby obtaining the heart rate data and the body temperature data of children;

[0180] In this embodiment, when extracting heart rate and temperature features from historical pediatric clinical desensitization data, the heart rate and temperature data of the child patient are first extracted from the electronic health record. Heart rate feature extraction includes calculating the heart rate variability (HRV) and the average value of the heart rate, while temperature feature extraction includes the average temperature, temperature fluctuation amplitude, etc. For example, if the data set includes heart rate and temperature values ​​recorded every hour, the average heart rate and temperature per hour can be calculated using a sliding window technique, and features can be extracted using statistical analysis methods (such as mean, standard deviation).

[0181] Step S362: constructing a health recovery degree evaluation system according to the heart rate data and the body temperature data of the child patient, and performing a health recovery degree evaluation on the heart rate data and the body temperature data of the child patient according to the health recovery degree evaluation system, thereby obtaining the health recovery degree data of the child patient;

[0182] In this embodiment, when constructing a health recovery evaluation system, heart rate and body temperature data can be used to define health recovery indicators. For example, the recovery status can be evaluated by comparing the patient's current data with the data difference in the normal healthy range. The specific method may include setting a health recovery threshold and using linear regression analysis to establish the relationship between heart rate and body temperature and health recovery. For example, by comparing heart rate data and body temperature data with the hallmark features of health recovery (such as the normal physiological range for each age stage), a recovery score can be calculated.

[0183] Step S363: extracting the time distribution characteristics of the physiological coefficients of the child patient according to the abnormal pediatric health care correction data, thereby obtaining the physiological coefficient distribution data of the child patient;

[0184] In this embodiment, when extracting the time distribution characteristics of physiological coefficients for abnormal pediatric health care correction data, it is necessary to analyze the time series data of physiological coefficients (such as blood pressure and respiratory rate). Time series analysis techniques, such as autocorrelation function (ACF) and partial autocorrelation function (PACF), can be used to extract time distribution characteristics. For example, daily distribution statistics are performed on the physiological coefficients of each patient, and the time change trend and periodic characteristics are extracted to obtain time distribution data.

[0185] Step S364: constructing a health recovery regression model according to the health recovery degree data of the child patient and the physiological coefficient distribution data of the child patient;

[0186] In this embodiment, when constructing a health recovery regression model based on the health recovery data and the physiological coefficient distribution data, a regression analysis technique, such as multiple linear regression or ridge regression, is used to establish the model. The input variables of the model include the time distribution characteristics of heart rate, body temperature and physiological coefficients, and the output variable is the health recovery data. Statistical software (such as scikit-learn in R or Python) can be used to train the model and evaluate the goodness of fit and predictive ability of the model.

[0187] Step S365: Calculate the health recovery impact factor for the abnormal pediatric health care correction data according to the health recovery regression model, thereby obtaining the health recovery impact factor.

[0188] In this embodiment, when calculating the health recovery influencing factor according to the health recovery regression model, the abnormal pediatric health care correction data is input into the regression model to obtain the health recovery influencing factor. The model can be used to compare the changes in health recovery under different nursing strategies in the abnormal pediatric health care correction data to identify key influencing factors. For example, the model can calculate the impact of different nursing interventions on health recovery and generate an influencing factor report to optimize the nursing strategy.

[0189] Optionally, step S4 is specifically:

[0190] Step S41: extracting clinician features from historical pediatric clinical desensitization data to obtain clinician data;

[0191] In this embodiment, in the historical pediatric clinical desensitization data, clinical records containing diagnostic information, treatment plans and patient feedback are first collected. The characteristic data of doctors are extracted by analyzing their personal information, such as professional background, years of experience and work location. For example, each doctor's professional field is marked as "respiratory department", "endocrinology department", etc., and their years of practice are recorded, so that "doctor data" can be obtained for subsequent analysis.

[0192] Step S42: extracting clinical diagnosis and treatment features from the clinical doctor data to obtain the doctor's clinical diagnosis and treatment data, and extracting diagnosis and treatment symptom descriptions from the doctor's clinical diagnosis and treatment data to obtain the doctor's diagnosis and treatment symptom description data;

[0193] In this embodiment, the doctor data obtained is used to further extract the doctor's diagnosis and treatment characteristics. For example, by analyzing the doctor's diagnosis and treatment records, their commonly used treatment methods, drugs and diagnostic techniques are extracted. Then, these characteristics are integrated to extract the doctor's diagnosis and treatment description when dealing with certain symptoms. For example, the treatment description for "asthma" includes the drugs used, treatment methods, etc., to form "doctor's diagnosis and treatment symptom description data".

[0194] Step S43: clustering the doctor's diagnosis and treatment symptom description data to obtain diagnosis and treatment symptom clustering data;

[0195] In this embodiment, cluster analysis is performed on the "doctor's diagnosis and treatment symptom description data" to classify similar symptom descriptions into one category. Using a clustering algorithm, such as K-means, similar symptom descriptions (such as "mild shortness of breath" and "moderate dyspnea") are grouped to obtain "diagnosis and treatment symptom clustering data", which helps to understand the distribution and treatment methods of common symptoms.

[0196] Step S44: performing diagnosis and treatment time distribution analysis based on the doctor's clinical diagnosis and treatment data to obtain diagnosis and treatment time distribution data, and performing symptom diagnosis and treatment time statistical integration based on the diagnosis and treatment symptom clustering data and the diagnosis and treatment time distribution data to obtain the pediatrician's clinical experience data;

[0197] In this embodiment, the time distribution analysis of "doctors' clinical diagnosis and treatment data" is performed to determine the average diagnosis and treatment time of different symptoms. By statistically analyzing data, such as "the average time for asthma diagnosis is 30 minutes", "diagnosis and treatment time distribution data" is generated. Combined with "diagnosis and treatment symptom clustering data", symptom diagnosis and treatment time statistics and integration can be performed to obtain "pediatrician clinical experience data", which provides comprehensive information such as diagnosis and treatment time and frequency of different symptoms.

[0198] Step S45: pediatrician recommendation priorities are divided according to health recovery influencing factors and pediatrician clinical experience data, thereby obtaining pediatrician recommendation priority data.

[0199] In this embodiment, the recommendation priority of pediatricians is formulated based on "health recovery influencing factors" (such as drug effects, treatment time) and "pediatrician clinical experience data". By establishing a weighted model, the doctor's experience is combined with the treatment effect to obtain the recommendation priority. For example, doctors who deal with "asthma" are given a higher priority because they have better experience and effects, thus forming "pediatrician recommendation priority data".

[0200] Optionally, step S45 is specifically:

[0201] Step S451: performing principal component analysis on health recovery influencing factors to obtain health recovery principal component data;

[0202] In this embodiment, quantitative data on factors affecting the patient's health recovery include but are not limited to the patient's recovery speed, relapse rate, health-related quality indicators, etc. These data are standardized so that data of different dimensions can be analyzed under the same standard. For example, the Z-score standardization method is used. Using the principal component analysis method, a linear transformation is applied to convert the original multidimensional data set into a new set of variables, namely the principal components. The first few principal components with a cumulative value of 80% or more of the explained variance ratio are selected. These principal components will capture the main variability in the data. The load coefficient of each principal component is calculated and used to generate health recovery principal component data. These principal component data will be used for further analysis.

[0203] Step S452: performing health recovery principal component proportion statistics on the doctor's diagnosis and treatment symptom description data according to the health recovery principal component data, thereby obtaining health recovery principal component proportion data;

[0204] In this embodiment, for the symptom description data provided by each doctor, its proportion under each principal component is calculated. This can be done by performing a dot product calculation on the principal component data and the symptom description data to obtain the proportion of each principal component in the symptom description. The principal component proportion data of each doctor is counted to generate the proportion data of each doctor under different principal components. This can be achieved by writing a statistical analysis script, for example, using Python's Pandas library to process data and generate statistical results. The doctor's diagnosis and treatment symptom description data can be mapped to the same principal component space, and the principal component proportion of each doctor's diagnosis and treatment data can be calculated. Specifically, the projection value of each doctor's diagnosis and treatment data on each principal component is calculated to obtain the proportion of each principal component.

[0205] Step S453: classifying the doctor diagnosis and treatment symptom description data by the principal component proportion according to the health recovery principal component proportion data, thereby obtaining the diagnosis and treatment doctor data with a high proportion and the diagnosis and treatment doctor data with a low proportion;

[0206] In this embodiment, a threshold is set according to the data in the statistical results of the principal component proportions to divide doctors into two categories: high proportion and low proportion. For example, the doctors with the highest proportion of the top 20% principal components are selected as high-proportion doctors, and the remaining doctors are selected as low-proportion doctors. Create two data sets, one containing the data of high-proportion doctors and the other containing the data of low-proportion doctors. This can be done by writing a classification script, such as using the classification function in Python's Scikit-learn library.

[0207] Step S454: dividing clinical experience doctors according to the clinical experience data of pediatricians, thereby obtaining data of highly experienced clinical doctors and data of low experienced clinical doctors;

[0208] In this embodiment, a threshold is set based on the experience data to classify doctors into two categories: high experience and low experience. For example, doctors with more than 10 years of clinical experience are selected as high experience doctors, and the rest are low experience doctors. Two data sets are created, one containing data of high experience clinicians and the other containing data of low experience clinicians. This can be achieved by defining a grading standard for years of experience.

[0209] Step S455: performing a clinician intersection operation on the data of high-proportion treating doctors and the data of highly experienced clinicians, so as to obtain data of pediatricians with high recommendation priority; performing a clinician intersection operation on the data of low-proportion treating doctors and the data of low-experience clinicians, so as to obtain data of pediatricians with low recommendation priority;

[0210] In this embodiment, the data of high-proportion doctors and the data of highly experienced clinicians are intersected. This operation can be implemented through set operations, such as using the set data structure in Python to perform intersection calculations to obtain a data set of doctors that belong to both high-proportion and high-experience categories, that is, high-recommendation priority pediatrician data. The data of low-proportion doctors and the data of low-experience clinicians are intersected to obtain low-recommendation priority pediatrician data, and the low-recommendation priority data is obtained by the same set operation calculation.

[0211] Step S456: Merge the data of pediatricians with high recommendation priority and the data of pediatricians with low recommendation priority to obtain pediatrician recommendation priority data.

[0212] In this embodiment, the data of pediatricians with high recommendation priority are merged with the data of pediatricians with low recommendation priority to form a final pediatrician recommendation priority data set. The merging operation can be completed by a data merging function, such as using the concat function in Python. On the merged data set, the doctors' recommendation priorities are sorted, and the doctors with high recommendation priorities are placed in front. The sorting can be achieved by a standard sorting algorithm, such as using the sort method in Python.

[0213] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is therefore intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0214] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A training method for an intelligent medical decision-making model based on pediatric clinical practice, characterized in that: The following steps are involved: Step S1: Acquire historical pediatric clinical data, and desensitize the sensitive clinical patient information of the historical pediatric clinical data, thereby obtaining historical pediatric clinical desensitized data; fuse the pediatric disease classification clinical data according to the historical pediatric clinical desensitized data, thereby obtaining historical pediatric clinical fusion data; Step S2: dividing the historical pediatric clinical fusion data into pediatric health care stages, thereby obtaining pediatric clinical health care stage data; Performing abnormal health care detection based on pediatric clinical health care stage data to obtain abnormal pediatric health care data; Step S3: Correct the abnormal pediatric health care data for the impact of epidemic diseases, thereby obtaining abnormal pediatric health care correction data, and calculate the health recovery impact factor based on the abnormal pediatric health care correction data, thereby obtaining the health recovery impact factor. Step S3 is specifically as follows: Step S31: Obtain epidemic disease description data, and extract epidemic time features from the epidemic disease description data, thereby obtaining epidemic time data; Step S32: performing time series correlation on the abnormal pediatric health care data according to the disease epidemic time data, thereby obtaining time-correlated abnormal pediatric health care data; Step S33: extracting symptom description features from the time-related abnormal pediatric health care data, thereby obtaining time-related care symptom description data; Step S34: Calculating the similarity of the symptom description according to the time-related nursing symptom description data and the epidemic disease description data, thereby obtaining the epidemic disease description similarity data, and identifying the epidemic disease nursing data for the time-related nursing symptom description data according to the epidemic disease description similarity data, thereby obtaining the epidemic disease nursing data; Step S35: performing epidemic disease correction on the abnormal pediatric health care data according to the epidemic disease care data, thereby obtaining abnormal pediatric health care correction data; Step S36: Calculating the health recovery impact factor according to the abnormal pediatric health care correction data, thereby obtaining the health recovery impact factor; Step S4: extracting clinician features from historical pediatric clinical desensitization data to obtain clinician data, and integrating pediatric clinical experience from the clinician data to obtain pediatrician clinical experience data; The pediatrician recommendation priority is divided according to the health recovery influencing factors and the pediatrician's clinical experience data, so as to obtain the pediatrician recommendation priority data; Step S5: Construct a pediatric clinician recommendation model based on the pediatrician recommendation priority data and health recovery influencing factors, and iterate and adjust the pediatric clinician recommendation model based on historical pediatric clinical desensitization data to obtain an intelligent pediatric medical decision-making model.

2. The training method of the intelligent medical decision-making model based on pediatric clinical practice according to claim 1 is characterized in that: Step S1 is specifically as follows: Step S11: Acquire historical pediatric clinical data, and desensitize the historical pediatric clinical data to sensitive clinical patient information, thereby obtaining historical pediatric clinical desensitized data; Step S12: classifying the historical pediatric clinical desensitization data into pediatric clinical disease categories, thereby obtaining historical pediatric clinical surgical data and historical pediatric clinical internal medicine data; Step S13: performing pediatric clinical surgery data fusion according to historical pediatric clinical surgery data, thereby obtaining pediatric clinical surgery fusion data; Step S14: performing pediatric clinical internal medicine data fusion according to historical pediatric clinical internal medicine data, thereby obtaining pediatric clinical internal medicine fusion data; Step S15: Perform clinical data integration on the pediatric clinical surgery fusion data and the pediatric clinical internal medicine fusion data to obtain historical pediatric clinical fusion data.

3. The training method of the intelligent medical decision-making model based on pediatric clinical practice according to claim 2 is characterized in that: Step S13 is specifically as follows: Step S131: extracting operation records from historical pediatric clinical surgical data to obtain pediatric clinical surgical operation data, and classifying the pediatric clinical surgical operation data by operation stage to obtain pre-operative surgical operation data, intra-operative surgical operation data, and post-operative surgical operation data; Step S132: integrating the preoperative description features of the surgical target site on the preoperative data of the surgical operation, thereby obtaining the preoperative description data of the surgical target site, and integrating the preoperative target site features according to the preoperative description data of the surgical target site, thereby obtaining the preoperative target site data; Step S133: extracting intraoperative monitoring data according to the intraoperative data of the surgical operation, thereby obtaining intraoperative monitoring data, and classifying the intraoperative monitoring data, thereby obtaining intraoperative monitoring images and intraoperative monitoring physiological coefficient data; Step S134: performing dynamic structural modeling of the target part according to the intraoperative monitoring image, thereby obtaining a dynamic structural model of the target part during surgery, and performing time series filling of physiological coefficient data of the dynamic structural model of the target part during surgery according to the intraoperative monitoring physiological coefficient data, thereby obtaining a dynamic structural model of the target part during surgery; Step S135: integrating the intraoperative target part features according to the intraoperative target part dynamic structure model, thereby obtaining intraoperative target part data; Step S136: extracting a postoperative target part image from the postoperative surgical data to obtain a postoperative target part image, and integrating postoperative target part features based on the postoperative target part image to obtain postoperative target part data; Step S137: Perform time series feature fusion on the preoperative target site data, the intraoperative target site data, and the postoperative target site data, so as to obtain pediatric clinical surgical fusion data.

4. The training method of the intelligent medical decision-making model based on pediatric clinical practice according to claim 2 is characterized in that: Step S14 is specifically as follows: Step S141: extracting medical symptom summaries and symptom examination records from historical pediatric clinical medical data, thereby obtaining medical symptom summary data and symptom examination data; Step S142: integrating the internal medicine symptom description features according to the internal medicine symptom summary data, thereby obtaining the internal medicine symptom description data; Step S143: extracting inspection images according to the symptom inspection data to obtain symptom inspection images, and dividing the symptom inspection images into image frames and image audio to obtain symptom inspection image frames and symptom inspection audio; Step S144: extracting physiological coefficient data features from the symptom examination data, thereby obtaining physiological coefficient data of internal medicine symptoms; Step S145: selecting a target inspection part for the symptom inspection image frame according to the medical symptom description data, thereby obtaining a target inspection part image frame; Step S146: constructing a three-dimensional structural model of the inspection part according to the inspection image frame of the target part, and fusing the time series structural features of the three-dimensional structural model of the inspection part according to the symptom inspection audio and the physiological coefficient data of the internal medicine symptom, so as to obtain the structural model of the internal medicine inspection part; Step S147: Integrate the characteristics of the internal medicine examination part into the internal medicine examination part structure model, so as to obtain pediatric clinical internal medicine fusion data.

5. The training method of the intelligent medical decision-making model based on pediatric clinical practice according to claim 1 is characterized in that: Step S2 is specifically as follows: Step S21: extracting clinical treatment time series features and clinical symptom description features from historical pediatric clinical fusion data, thereby obtaining pediatric clinical treatment time series data and clinical symptom description data; Step S22: sorting and dividing the description time according to the clinical symptom description data, thereby obtaining the initial symptom description data and the final symptom description data; Step S23: Describe and time-series associate the pediatric clinical treatment time series data according to the initial symptom description data, so as to obtain the initial diagnosis and treatment time data; describe and time-series associate the pediatric clinical treatment time series data according to the final symptom description data, so as to obtain the review treatment time data; Step S24: dividing the pediatric clinical treatment time series data into time windows according to the initial diagnosis and treatment time data and the reexamination and treatment time data, thereby obtaining pediatric treatment stage data and pediatric treatment observation stage data; Step S25: integrating the historical pediatric clinical fusion data in time windows according to the pediatric treatment stage data and the pediatric treatment observation stage data, thereby obtaining pediatric clinical health care stage data; Step S26: performing abnormal health care detection according to the pediatric clinical health care stage data, thereby obtaining abnormal pediatric health care data.

6. The training method of the intelligent medical decision-making model based on pediatric clinical practice according to claim 5 is characterized in that: Step S26 is specifically as follows: Step S261: extracting nursing stage physiological coefficient data from pediatric clinical health nursing stage data, thereby obtaining nursing stage physiological coefficient data; Step S262: performing frequency domain conversion according to the physiological coefficient data of the nursing stage, thereby obtaining a spectrum of the physiological coefficient of the nursing stage; Step S263: extracting vibration amplitude features from the physiological coefficient spectrum during the nursing stage, thereby obtaining physiological coefficient vibration amplitude data, and performing physiological coefficient spectrum vibration amplitude threshold statistics based on the physiological coefficient vibration amplitude data, thereby obtaining a spectrum vibration amplitude threshold; Step S264: performing abnormal vibration amplitude stage spectrum recognition on the nursing stage physiological coefficient spectrum according to the spectrum vibration amplitude threshold, thereby obtaining the abnormal vibration amplitude stage spectrum; Step S265: performing time domain conversion on the abnormal vibration amplitude stage spectrum to obtain abnormal pediatric health care stage data, and performing initial diagnosis abnormal correction on the abnormal pediatric health care stage data to obtain abnormal pediatric health care data.

7. The training method of the intelligent medical decision-making model based on pediatric clinical practice according to claim 1 is characterized in that: Step S36 is specifically as follows: Step S361: extracting the heart rate characteristics of children and the body temperature characteristics of children from the historical pediatric clinical desensitization data, thereby obtaining the heart rate data and the body temperature data of children; Step S362: constructing a health recovery degree evaluation system according to the heart rate data and the body temperature data of the child patient, and performing a health recovery degree evaluation on the heart rate data and the body temperature data of the child patient according to the health recovery degree evaluation system, thereby obtaining the health recovery degree data of the child patient; Step S363: extracting the time distribution characteristics of the physiological coefficients of the child patient according to the abnormal pediatric health care correction data, thereby obtaining the physiological coefficient distribution data of the child patient; Step S364: constructing a health recovery regression model according to the health recovery degree data of the child patient and the physiological coefficient distribution data of the child patient; Step S365: Calculate the health recovery impact factor for the abnormal pediatric health care correction data according to the health recovery regression model, thereby obtaining the health recovery impact factor.

8. The training method of the intelligent medical decision-making model based on pediatric clinical practice according to claim 1 is characterized in that: Step S4 is specifically as follows: Step S41: extracting clinician features from historical pediatric clinical desensitization data to obtain clinician data; Step S42: extracting clinical diagnosis and treatment features from the clinical doctor data to obtain the doctor's clinical diagnosis and treatment data, and extracting diagnosis and treatment symptom descriptions from the doctor's clinical diagnosis and treatment data to obtain the doctor's diagnosis and treatment symptom description data; Step S43: clustering the doctor's diagnosis and treatment symptom description data to obtain diagnosis and treatment symptom clustering data; Step S44: performing diagnosis and treatment time distribution analysis based on the doctor's clinical diagnosis and treatment data to obtain diagnosis and treatment time distribution data, and performing symptom diagnosis and treatment time statistical integration based on the diagnosis and treatment symptom clustering data and the diagnosis and treatment time distribution data to obtain the pediatrician's clinical experience data; Step S45: pediatrician recommendation priorities are divided according to health recovery influencing factors and pediatrician clinical experience data, thereby obtaining pediatrician recommendation priority data.

9. The training method of the intelligent medical decision-making model based on pediatric clinical practice according to claim 8, characterized in that: Step S45 is specifically as follows: Step S451: performing principal component analysis on health recovery influencing factors to obtain health recovery principal component data; Step S452: performing health recovery principal component proportion statistics on the doctor's diagnosis and treatment symptom description data according to the health recovery principal component data, thereby obtaining health recovery principal component proportion data; Step S453: classifying the doctor diagnosis and treatment symptom description data by the principal component proportion according to the health recovery principal component proportion data, thereby obtaining the diagnosis and treatment doctor data with a high proportion and the diagnosis and treatment doctor data with a low proportion; Step S454: dividing clinical experience doctors according to the clinical experience data of pediatricians, thereby obtaining data of highly experienced clinical doctors and data of low experienced clinical doctors; Step S455: performing a clinician intersection operation on the data of high-proportion treating doctors and the data of highly experienced clinicians, thereby obtaining data of pediatricians with high recommendation priority; Perform clinician intersection operations on the data of low-proportion treating doctors and low-experience clinicians to obtain the data of pediatricians with low recommendation priority; Step S456: Merge the data of pediatricians with high recommendation priority and the data of pediatricians with low recommendation priority to obtain pediatrician recommendation priority data.

Images