A health degree evaluation system and method for endocrine care

Abstract

The present invention discloses a health degree evaluation system and method for endocrine care, which relates to the technical field of health evaluation. It includes a state vector construction module for collecting the patient's basic data and constructing an individual state vector S, and collecting the patient's physiological data in real time and using a CNN model to extract deep features; an anomaly detection module for using PCA to reduce the dimension of the deep features and then inputting a mutation point detection algorithm to detect periodic mutation points to obtain a set of mutation points, and concurrently using an SNN to identify spike activation responses to obtain a set of spike anomalies, and integrating the obtained set of mutation points and the set of spike anomalies into an anomaly feature vector A. The present invention realizes the dynamic and accurate evaluation of the health status of endocrine patients. By real-time monitoring and automatic identification of hormone fluctuations, behavioral abnormalities and changes in health status, it can not only realize accurate multi-symptom risk prediction, but also generate personalized nursing intervention plans according to the individual health status, thereby improving the nursing efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of health assessment, and particularly to a health degree assessment system and method for endocrine care. Background Art

[0002] With the continuous development of medical health information technology, personalized medicine and intelligent care have been widely applied in chronic disease management. Endocrine-related diseases (such as thyroid dysfunction, polycystic ovary syndrome, perimenopausal syndrome, etc.) have the characteristics of periodicity, strong individual heterogeneity, and fuzzy intervention windows due to their involvement of multi-dimensional factors such as physiological hormone fluctuations, emotional states, and life behaviors. Therefore, how to achieve dynamic perception, risk prediction, and intervention feedback of the health status of endocrine patients during the nursing stage has become one of the important directions for the intelligent upgrade of the current medical nursing system. At present, some studies have tried to remotely monitor single-channel physiological parameters such as blood glucose, body temperature, and HRV, or predict the trend of a single pathological index through machine learning methods, but most are limited to linear modeling or static feature extraction, and it is difficult to depict complex cross-time-scale non-linear changes. Especially in the aspect of real-time health degree quantitative assessment for the nursing stage, there are still obvious technical gaps. Summary of the Invention

[0003] In view of the above existing problems, the present invention is proposed.

[0004] Therefore, the present invention provides a health degree assessment system and method for endocrine care, which solves the problem that most of the existing technologies are limited to linear modeling or static feature extraction, and it is difficult to depict complex cross-time-scale non-linear changes. Especially in the aspect of real-time health degree quantitative assessment for the nursing stage, there are still obvious technical gaps.

[0005] To solve the above technical problems, the present invention provides the following technical solutions:

[0006] In the first aspect, the present invention provides a health degree assessment system for endocrine care, which includes,

[0007] A state vector construction module, configured to collect the patient's basic data and construct an individual state vector S, and collect the patient's physiological data in real time and use a CNN model to extract deep features;

[0008] An anomaly detection module, configured to use PCA to reduce the dimension of the deep features and then input them into a mutation point detection algorithm to detect periodic mutation points, obtain a set of mutation points, and concurrently use an SNN to identify spike activation responses to obtain a set of spike anomalies, and integrate the obtained set of mutation points and the set of spike anomalies into an anomaly feature vector A;

[0009] A health status classification module, which is used to collect the historical abnormal feature vector C of a patient, construct a new sample set D with vector A and vector C, and calculate the local density and the distance to the nearest high-density point of all samples based on the density peak clustering algorithm to obtain the corresponding health status label;

[0010] A prediction module, which is used to input the abnormal feature set A and the health status label into the XGBoost model to perform multi-symptom probability prediction and calculate the total health risk score;

[0011] A health assessment module, which is used to divide the health level according to the health risk score and generate the corresponding nursing intervention plan.

[0012] As a preferred solution of the health degree assessment system for endocrine nursing described in the present invention, wherein: after using PCA to reduce the dimension of the deep features and inputting them into the mutation point detection algorithm to detect periodic mutation points, the obtained mutation point set includes:

[0013] Collect the feature vectors of the past n moments to construct a sliding window sample set, and use the current feature vector as the last sample. Calculate the mean and standard deviation for each column of the sample set, standardize each item to generate a new sample set, calculate the covariance matrix for the standardized sample set, perform eigenvalue decomposition on the covariance matrix, arrange the eigenvalues in descending order, calculate the cumulative contribution rate of the first m principal components in turn, intercept the first m feature vectors to form the principal component projection matrix and construct the principal component loading matrix with the eigenvalue diagonal matrix, and perform a linear transformation on the current standardized feature vector to obtain a reduced-dimensional deep feature vector with a dimension of m;

[0014] Traverse the principal component dimension and calculate the total loading value on the original hormone features ;

[0015] Find the principal component dimension with the largest loading value and define this dimension as the hormone dominant component;

[0016] Trace back the last T moments, extract the principal component sequence, extract the values of the principal component dimension at each time point to form a hormone trend sequence;

[0017] Perform the first-order discrete wavelet transform on the hormone trend sequence using Daubechies 6 wavelet;

[0018] Traverse the adjacent data points in the detail coefficients and extract the set M of time indices that satisfy the modulus maximum condition;

[0019] Calculate the amplitude of the detail coefficients of all points in the set M for point selection to obtain a candidate set of mutation points;

[0020] For each candidate point, intercept the local hormone trend subsequence with its center as ;

[0021] For each subsequence, a state space model is established;

[0022] Perform the prediction-update iteration process of Kalman filtering for each subsequence;

[0023] After the filtering process is completed, the smoothed trend sequence H of the candidate point is obtained;

[0024] Initialize the smoothed trend mapping structure, create a trend mapping dictionary, register the smoothed sequence and the original sequence of each candidate point, and encapsulate them into a set of structural tuples;

[0025] Read the smoothed trend sequence corresponding to each candidate point in turn, and build a structural mutation fluctuation model and a control benchmark model based on the sequence;

[0026] Use maximum likelihood estimation to fit the parameters of the two models, and obtain the parameters and as well as the log-likelihood and , and construct the Bayes factor based on the obtained and ;

[0027] Summarize all time points that meet the mutation significance conditions, and calculate the maximum mutation amplitude of each retained mutation point in the smoothed sequence;

[0028] Calculate the average mutation amplitude, maximum mutation amplitude, average mutation position, mutation time variance of the mutation points, and construct them into a mutation feature sub-vector.

[0029] As a preferred solution of the health degree evaluation system for endocrine care described in the present invention, wherein: the parallel use of SNN to identify the spike activation response to obtain the spike anomaly set, and integrate the obtained mutation point set and spike anomaly set into the anomaly feature set A, including:

[0030] Read the original high-frequency time series data, including the heart rate variability sequence and the basal body temperature sequence, align the two-channel data in time, and splice them into a two-dimensional input matrix , standardize each column of data, calculate the data standard deviation, and set the dynamic judgment threshold as Q;

[0031] Input row by row into the pre-trained SNN model for event activation judgment;

[0032] If in any continuous 3-time-point window, any channel continuously exceeds the threshold Q, it is determined as a spike response, mark the current response window and record its time position;

[0033] Integrate the spike response mark and the spike response intensity into a spike anomaly feature vector;

[0034] Integrate the mutant feature sub-vector and the spike anomaly feature vector to generate the anomaly feature vector A.

[0035] As a preferred solution of the health degree evaluation system for endocrine care described in the present invention, wherein: calculating the local density and the distance to the nearest high-density point of all samples based on the density peak clustering algorithm to obtain the corresponding health status label means using the truncated kernel function density estimation method to calculate the number of points in the neighborhood of the sample point to obtain the density degree of the sample point;

[0036] Evaluate the minimum distance of each point to the points with higher density than it;

[0037] Calculate the score of each sample point as the clustering center , including the local density and the minimum high-density distance;

[0038] According to the actual needs of endocrine care, the health status is divided into three typical manifestations, including stable type, fluctuating type and unstable type, the fixed number of clusters is 3. After calculating the clustering center scores of all samples, select the K = 3 points with the highest scores as the center points of clustering. For each sample point, calculate its distance to the 3 center points and assign it to the nearest cluster;

[0039] Finally, the current patient will be assigned to the corresponding health classification group to obtain the corresponding health status label.

[0040] As a preferred solution of the health degree evaluation system for endocrine care described in the present invention, wherein: inputting the anomaly feature vector A and the health status label into the XGBoost model to perform multi-symptom probability prediction and calculate the total health risk score means integrating the anomaly feature vector A and the health status label into the feature vector U, inputting the vector into the pre-trained XGBoost model, performing multi-symptom probability prediction for each patient, and according to the obtained multi-symptom probability output, perform weighted summation to obtain the total health risk score.

[0041] As a preferred solution of the health degree evaluation system for endocrine care described in the present invention, wherein: dividing the health level according to the health risk score and generating the corresponding nursing intervention plan means setting thresholds and , and , if the health risk score is greater than or equal to the threshold , then the patient is in the high-risk period, push a reminder for a follow-up visit and alarm the probability of typical symptoms. If the health risk score is less than the threshold and greater than or equal to the threshold , then the patient is in the fluctuating period, push diet advice to the patient and strengthen the work and rest advice. If the health risk score is less than the threshold If so, the patient is in a stable period, continue the current care plan and push low-intensity exercise tasks to the patient.

[0042] As a preferred solution of the health degree evaluation system for endocrine care according to the present invention, wherein: collecting the patient's basic data and constructing an individual state vector S, and collecting the patient's physiological data in real time and using a CNN model to extract deep features includes:

[0043] The patient's basic data includes the type of diagnosed disease and treatment stage of the patient, three hormone indexes of the patient, average menstrual cycle length, emotional index score and sleep quality score. After preprocessing the collected basic data, it is constructed into an individual state vector S;

[0044] Medical staff fill in the type of diagnosed disease and treatment stage of the patient, number the disease type as and number the treatment stage as ;

[0045] Read the patient's recent three hormone test data from the hospital LIS system, calculate the average value of the three hormone indexes as the current basic level, extract the user's menstrual start time series in the recent n months, and calculate the average cycle length and the interval between the current date and the last menstruation , use the PHQ-9 scale to collect the patient's emotional index score , use the PSQI scale to collect the patient's sleep quality score ;

[0046] Based on the individual state vector S, adjust the sampling period of the patient's physiological data, continuously collect the patient's physiological signals, behavior indicators, and sleep and emotions for 24 hours, arrange all the data collected within 24 hours in chronological order, form a time series data matrix and perform zero-mean normalization processing on the matrix;

[0047] Input the normalized matrix into a pre-trained one-dimensional convolutional neural network to extract deep fusion feature representations and obtain a deep feature vector F.

[0048] In a second aspect, the present invention provides a method for evaluating the health degree of endocrine care, including,

[0049] Collecting the patient's basic data and constructing an individual state vector S, collecting the patient's physiological data in real time and using a CNN model to extract deep features;

[0050] Use PCA to reduce the dimension of the deep features and then input them into a mutation point detection algorithm to detect periodic mutation points, obtain a set of mutation points, and concurrently use SNN to identify spike activation responses to obtain a set of spike anomalies. Integrate the obtained set of mutation points and the set of spike anomalies into an abnormal feature vector A;

[0051] Collect the historical abnormal feature vector C of the patient, construct vectors A and C into a new sample set D, calculate the local density and the distance to the nearest high-density point of all samples based on the density peak clustering algorithm, and obtain the corresponding health status labels.

[0052] Input the abnormal feature set A and the health status labels into the XGBoost model for multi-symptom probability prediction and calculate the total health risk score.

[0053] Construct a logical rule tree to divide the health levels and generate corresponding nursing intervention plans.

[0054] In a third aspect, the present invention provides a computer device, including a memory and a processor, where the memory stores a computer program, and: when the computer program is executed by the processor, any step of the health degree evaluation system for endocrine nursing as described in the first aspect of the present invention is implemented.

[0055] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and: when the computer program is executed by the processor, any step of the health degree evaluation system for endocrine nursing as described in the first aspect of the present invention is implemented.

[0056] The beneficial effects of the present invention are as follows: The present invention realizes the dynamic and accurate evaluation of the health status of endocrine patients, can effectively solve the lag and limitation problems of traditional methods in dealing with the health assessment of endocrine patients, especially has significant advantages in capturing small and critical health fluctuations. By real-time monitoring and automatic identification of hormone fluctuations, behavioral abnormalities and health status changes, the present invention can not only achieve accurate multi-symptom risk prediction, but also generate personalized nursing intervention plans according to individual health conditions, thereby improving the nursing efficiency and patient compliance. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0058] Figure 1 It is a structural diagram of the health degree evaluation system for endocrine nursing in Embodiment 1.

[0059] Figure 2 It is a flowchart of the health degree evaluation method for endocrine nursing in Embodiment 1. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0060] To make the above objects, features, and advantages of the present invention more apparent and understandable, the following detailed description of the specific embodiments of the present invention will be provided in conjunction with the accompanying drawings of the specification.

[0061] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention. However, the present invention may be practiced in other ways different from those described herein. Those skilled in the art can make similar generalizations without departing from the spirit of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0062] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation manner of the present invention. The appearances of "in one embodiment" in different places in this specification do not all refer to the same embodiment, nor are they separate or alternative embodiments that are mutually exclusive of other embodiments.

[0063] Example 1, referring to Figure 1 and Figure 2 , which is the first embodiment of the present invention. This embodiment provides a health degree evaluation system for endocrine care, including the following steps:

[0064] S1. A state vector construction module for collecting basic patient data and constructing an individual state vector S, and collecting patient physiological data in real time and using a CNN model to extract deep features;

[0065] Specifically, collecting basic patient data and constructing an individual state vector S, and collecting patient physiological data in real time and using a CNN model to extract deep features include:

[0066] The basic patient data includes: the type of disease diagnosed and the treatment stage of the patient, three hormone indicators of the patient, the average menstrual cycle length, the emotional index score, and the sleep quality score. The collected basic data is preprocessed and then constructed into an individual state vector S;

[0067] Medical staff fill in the type of disease diagnosed and the treatment stage of the patient, and number the disease type as , for example:

[0068] : Hypothyroidism;

[0069] : Polycystic ovary syndrome;

[0070] : Perimenopausal syndrome;

[0071] (If it is a complex disease, it is determined according to the main diagnosis code)

[0072] Number the treatment stage as , for example:

[0073] : Initial treatment stage (starting medication < 30 days);

[0074] : Maintenance treatment stage (medication > 30 days and stable condition);

[0075] : Dose reduction observation stage (drug reduction or stage of intended discharge);

[0076] Read the patient's recent three hormone test data (E2 (estradiol), P4 (progesterone), TSH (thyroid stimulating hormone)) from the hospital LIS system, and calculate the average value of the three hormone indicators , , as the current baseline level (to avoid single fluctuations), where is the estradiol baseline value, is the progesterone baseline value, is the thyroid stimulating hormone baseline value, extract the user's recent n - month menstrual start time series, and calculate the average cycle length and the interval between the current date and the last menstrual period , collect the patient's emotional index score using the PHQ - 9 scale , collect the patient's sleep quality score using the PSQI scale ;

[0077] Adjust the sampling period of the patient's physiological data based on the individual state vector S, including determining the core monitoring channels according to the disease type (such as BT, HRV related to hormones), determining whether to enter the high - frequency transition period according to the treatment stage, judging whether high - frequency detection is required for thyroid risk based on the current baseline level, judging whether there are moderate to severe emotional fluctuations according to the emotional index score, identifying whether the cycle is disordered according to the average menstrual cycle length, and extracting the sampling frequency adjustment logic from the built - in rule library. The example is as follows (not an exhaustive list, the actual system is based on this rule):

[0078]

[0079] Continuously collect the patient's physiological signals, behavioral indicators, and sleep and emotions for 24 hours. Arrange all the data collected within 24 hours in chronological order to form a time - series data matrix and perform zero - mean normalization on the matrix to eliminate the dimensional differences of numerical values between different dimensions;

[0080] Input the normalized matrix into a pre - trained one - dimensional convolutional neural network to extract deep - fused feature representations and obtain the deep feature vector F.

[0081] By collecting the patient's basic data (including disease type, treatment stage, hormone levels, menstrual cycle, mood, and sleep quality, etc.) and constructing it into an individual state vector S, the system can comprehensively and dynamically reflect the patient's health status. Compared with traditional single physiological parameter monitoring methods, the individual state vector can integrate multi-dimensional data together, provide a more comprehensive and personalized health assessment, and effectively improve the system's perception ability of the patient's health condition.

[0082] S2. Anomaly detection module, which is used to input the deep features after dimensionality reduction using PCA into a mutation point detection algorithm to detect periodic mutation points, obtain a set of mutation points, concurrently use SNN to identify spike activation responses to obtain a set of spike anomalies, and integrate the obtained set of mutation points and the set of spike anomalies into an anomaly feature vector A;

[0083] Specifically, after using PCA to perform dimensionality reduction on the deep features and inputting them into a mutation point detection algorithm to detect periodic mutation points, the set of mutation points obtained includes:

[0084] Collect the feature vectors at the past n moments to construct a sliding window sample set, and use the current feature vector as the last sample (such as every hour for the past 72 hours). Calculate the mean and standard deviation for each column of the sample set, standardize each item to generate a new sample set, calculate the covariance matrix for the standardized sample set, perform eigenvalue decomposition on the covariance matrix to obtain the feature vectors and the corresponding set of eigenvalues, arrange the eigenvalues in descending order, calculate the cumulative contribution rate of the first m principal components in turn, select the minimum number of principal components m that makes the cumulative variance interpretation rate greater than or equal to the set threshold G (set empirically, such as 95%), intercept the first m feature vectors to form a principal component projection matrix and construct a principal component loading matrix with the eigenvalue diagonal matrix, and perform a linear transformation on the current standardized feature vector to obtain a dimensionality-reduced deep feature vector of dimension m;

[0085] Traverse the principal component dimensions and calculate the total loading value on the original hormone features (such as E2, P4, TSH) :

[0086] ;

[0087] Where is the linear weight of the i-th original feature on the j-th principal component, is the index number containing the columns where E2, P4, and TSH are located;

[0088] Find the principal component dimension with the largest loading value and define this dimension as the hormone dominant component;

[0089] Trace back the most recent T moments (such as T = 72 hours), extract the principal component sequence, extract the values of the principal component dimensions at each time point to form a hormone trend sequence;

[0090] Perform a first-order discrete wavelet transform on the hormone trend sequence using Daubechies 6 (db6) wavelets:

[0091] ;

[0092] where, is the hormone dominant principal component time series, is the wavelet detail coefficient sequence, is the approximation coefficient;

[0093] Traverse adjacent data points in the detail coefficients and extract the set M of time indices that satisfy the modulus maximum condition;

[0094] Calculate the amplitudes of the detail coefficients at all points in set M, calculate the median and absolute median difference and sum them as the threshold Y, and screen out the points with amplitudes greater than the threshold Y to obtain the candidate set of mutation points;

[0095] For each candidate point, intercept the local hormone trend subsequence with its center at ;

[0096] For each subsequence, establish a state space model: including a state transition model (representing the trend of hormone levels over time) and an observation model (representing the collected hormone values);

[0097] ;

[0098] ;

[0099] where, is the estimated smoothed hormone value, is the actual observed value, is the state transition coefficient, is the system noise term, is the observation noise term;

[0100] Perform the prediction-update iteration process of the Kalman filter for each subsequence:

[0101] Prediction step:

[0102] ;

[0103] where, is the state estimate value at the previous moment, is the state transition coefficient, is the covariance estimate value, Q is the state process noise covariance, is the predicted state at the current moment, is the state estimate covariance at the previous moment, is the current predicted covariance;

[0104] Update step:

[0105] ;

[0106] In the formula, is the actual observed value, R is the observation noise covariance, is the Kalman gain coefficient, is the current state estimate value;

[0107] After the filtering process is completed, the smoothed trend sequence H of the candidate point is obtained;

[0108] Initialize the smoothed trend mapping structure, create a trend mapping dictionary, register the smoothed sequence and the original sequence of each candidate point and encapsulate them into a set of structure tuples;

[0109] Read the smoothed trend sequence corresponding to each candidate point in turn and build a structural mutation volatility model (β-ARCH model) based on the sequence:

[0110] ;

[0111] In the formula, is the conditional variance, is the smoothed hormone trend value, is the benchmark term of the conditional variance, obtained by maximum likelihood estimation, is the weight of the historical observation term, m is the volatility scaling factor, is the non-linear exponent, is the sequence residual term, obtained by system simulation, is the observed value corresponding to the j-th lag time (i.e., the j-th past time step) at the current time point t, q is the historical lag order considered in the model), and the control benchmark model (ordinary ARCH model: );

[0112] Use maximum likelihood estimation to fit the parameters of the two models, and obtain the parameters and as well as the log-likelihood and ;

[0113] Based on the obtained and Construct the Bayes factor :

[0114] ;

[0115] Calculate the Bayes factor B for each candidate point, and compare B with a preset threshold U. The threshold U is set through experimental tuning. If B is greater than the threshold U, the mutant structure is considered significant and the point is retained; otherwise, it is regarded as non-structural fluctuation and eliminated.

[0116] Summarize all time points that meet the mutant significance condition, and calculate the maximum mutant amplitude of each retained mutant point in the smoothed sequence:

[0117] ;

[0118] In the formula, is the smoothed value of the candidate mutant point, is the smoothed value at the next moment, is the amplitude change of this mutant point;

[0119] Calculate the average mutant amplitude, maximum mutant amplitude, average mutant position, mutant time variance of the mutant points and construct them into a mutant feature sub-vector.

[0120] Dimensionality reduction of deep features is performed using principal component analysis (PCA), which can reduce the computational complexity while retaining the main information of the data. By extracting the main components and removing redundant information, PCA significantly improves the efficiency and accuracy of subsequent data processing and anomaly detection. Through the method of combining wavelet transform and Kalman filtering, the present invention can efficiently extract the tiny but crucial mutation points in hormone fluctuations. Using the Daubechies 6 (db6) wavelet to perform the first-order discrete wavelet transform on the hormone trend sequence can capture the high-frequency changes in the hormone level fluctuations, and potential candidate sets of mutation points are screened out through the modulus maximum condition. The prediction-update iterative process of Kalman filtering is performed on the candidate sets of mutation points, which smooths the trend of the hormone level and eliminates the interference of system noise. By modeling each subsequence through a state space model, not only the accuracy of the smoothing result is improved, but also a β-ARCH model with structural mutation fluctuations can be further constructed based on the smoothed hormone trend data to accurately evaluate the conditional variance and non-linear changes of the endocrine level fluctuations. The present invention fits the model parameters through maximum likelihood estimation and calculates the Bayes factor, which can accurately evaluate the significance of each mutation point. This process ensures that only those mutation points with significant structural fluctuations are retained, avoiding misjudgment and noise interference, and improving the accuracy and reliability of detection. After the mutation points are screened, the present invention further calculates the maximum mutation amplitude and temporal variance of each retained mutation point in the smoothed sequence, providing rich feature data for subsequent health risk prediction. These quantitative indicators help to comprehensively evaluate the severity, onset time and fluctuation pattern of the mutation points, providing an accurate basis for the real-time assessment of the endocrine health status. Combining mutation point detection and abnormal feature generation, the present invention can identify abnormal patterns of endocrine fluctuations in a short time, featuring high sensitivity and low error rate.

[0121] Further, the SNN is used in parallel to identify spike activation responses to obtain a spike anomaly set, and the obtained set of mutation points and the spike anomaly set are integrated into an abnormal feature set A, including:

[0122] Read the original high-frequency time series data, including the heart rate variability sequence and the basal body temperature sequence, align the two-channel data in time, and splice them into a two-dimensional input matrix , standardize each column of data, calculate the data standard deviation, and set the dynamic judgment threshold as Q (±3σ);

[0123] The is input row by row into a pre-trained SNN model for event activation judgment. The model structure includes: an input layer (2 channels), a sparsely connected spiking neuron layer (using the LIF neuron model), and an output discrimination layer (spike response state output);

[0124] If in any three consecutive time - point windows, any channel continuously exceeds the threshold Q, it is determined as a spike response. Mark the current response window and record its time position. The response structure is a boolean variable: ;

[0125] wherein, is the spike - anomaly trigger flag, indicating that if a spike response is triggered;

[0126] Calculate the standardized deviation value for all the moments when spike events are triggered, and take the maximum value as the current spike - response intensity index. If no spike response is triggered, the spike - response intensity index is zero;

[0127] Integrate the spike - response mark and the spike - response intensity into a spike - anomaly feature vector;

[0128] Integrate the mutation - feature sub - vector and the spike - anomaly feature vector to generate the anomaly feature vector A.

[0129] By standardizing the high - frequency time - series data (such as heart - rate variability sequences and basal - body - temperature sequences) and setting the dynamic judgment threshold as Q (±3σ), this step can achieve unbiased comparison of data and eliminate the calculation errors caused by different dimensions of each physiological parameter. The standardized data can effectively eliminate the interference of environmental and external factors on physiological signals, thus improving the accuracy of signal recognition. In addition, the setting of using the ±3σ threshold for spike activation judgment can better ensure the recognition of mutation points and abnormal fluctuations under a high - noise background, so as to ensure that only significant health - state fluctuations are captured, rather than normal physiological fluctuations. The SNN can utilize its time - encoding characteristics to capture the instantaneous changes of physiological signals. Compared with traditional neural - network models, the SNN can better reflect the time - dependence of physiological changes when processing physiological data with time - series characteristics. By judging the spike activation response of high - frequency data, the SNN model can effectively identify sudden and non - linear health fluctuations, such as abnormal mutations or short - term fluctuations in physiological data such as heart rate and body temperature, thus providing accurate original signals for subsequent anomaly recognition and health - risk assessment. By simultaneously detecting mutation points and spike responses, potential health problems can be identified from different perspectives (time change, sudden fluctuation). This multi - dimensional anomaly detection can improve the detection sensitivity of the system and timely discover health problems with unobvious clinical manifestations. By capturing anomalies in multiple dimensions such as physiology, emotion, and behavior of an individual, a customized health - risk assessment model can be established for each patient to provide accurate health - management solutions. Real - time collection and analysis of physiological signals can reflect the immediate changes in the patient's health state, rather than a lagged assessment based on historical data. This dynamic real - time monitoring can timely discover potential health problems, especially in the intervention process of endocrine regulation and periodic diseases.

[0130] S3. A health status classification module, which is used to collect the historical abnormal feature vectors C of patients, construct a new sample set D with vectors A and vectors C, and calculate the local density and the distance to the nearest high-density point of all samples based on the density peak clustering algorithm to obtain the corresponding health status labels;

[0131] Specifically, calculating the local density and the distance to the nearest high-density point of all samples based on the density peak clustering algorithm to obtain the corresponding health status labels means using the truncated kernel function density estimation method to calculate the number of points within the neighborhood of the sample point to obtain the density degree of the sample point:

[0132] ;

[0133] ;

[0134] In the formula, is the sample point and the sample point is the Euclidean distance between them, is a preset truncated distance threshold, set based on ROC curve analysis, and is used to determine the neighborhood range of the sample point, is the step function, N is the size of the sample set, is the local density of each sample point;

[0135] Evaluate the minimum distance of each point to the points with higher density than it:

[0136] ;

[0137] In the formula, represents the local density of the sample point ;

[0138] Calculate the score of each sample point as the clustering center, including local density and minimum high-density distance:

[0139] ;

[0140] According to the actual needs of endocrine care, the health status is divided into three typical manifestations, including stable type, fluctuating type, and unstable type. The fixed number of clusters is 3. After calculating the clustering center scores of all samples, select the K = 3 points with the highest scores as the center points of the clustering. For each sample point, calculate its distance to the 3 center points and assign it to the nearest cluster;

[0141] The specific classification labels are as follows:

[0142] Group A (rhythm stable type): The number of mutation points is less than the preset threshold W, and no spike abnormality is triggered;

[0143] Group B (rhythmic fluctuation type): The number of mutation points is greater than or equal to the preset threshold W, and no spike anomaly is triggered;

[0144] Group C (mutation instability type): The number of mutation points is greater than or equal to the preset threshold W, and is accompanied by spike anomalies;

[0145] The threshold W is set empirically;

[0146] Finally, the current patient will be assigned to the corresponding healthy classification group to obtain the corresponding healthy status label.

[0147] The truncated kernel density estimation is an improved kernel density estimation method used to estimate the density distribution of sample data. The present invention performs clustering and classification on sample points through the density peak clustering algorithm (DPC), and automatically identifies changes in the healthy status according to the local density of each sample and the distance to high-density points. This method has the characteristics of strong self-adaptability and unsupervisedness, and can dynamically determine the clustering center according to the actual data of the patient, avoiding the limitation of the need to preset the number of clusters in the traditional method. In this way, the system can automatically and accurately classify the patient's healthy status into three categories: stable type, fluctuating type, and unstable type without relying on manual setting. Through the truncated kernel density estimation method, the system can calculate the local density more accurately when processing the patient's physiological data, especially during complex healthy fluctuation periods, and can effectively identify small but critical healthy changes. The healthy classification method based on the density peak clustering algorithm dynamically divides the patient's healthy status by analyzing the local density of sample points and the minimum high-density distance. This method can evaluate the patient's health risk in real time and accurately distinguish different healthy statuses, thus providing data support for subsequent nursing interventions. By dividing patients into Group A (rhythmic stable type), Group B (rhythmic fluctuation type), and Group C (mutation instability type), the system can provide personalized nursing plans according to the characteristics of each healthy status.

[0148] S4. A prediction module, configured to input the abnormal feature set A and the healthy status label into the XGBoost model to perform multi-symptom probability prediction and calculate the total health risk score;

[0149] Specifically, the abnormal feature vector A and the health status label are input into the XGBoost model for multi-symptom probability prediction and calculation of the total health risk score. That is, the abnormal feature vector A and the health status label are integrated into the feature vector U. This step effectively integrates the individual health characteristics of the patient and their health status label, providing rich input information for the subsequent machine learning model. By converting multi-dimensional data (such as physiological indicators, behavior patterns, emotional states, etc.) into feature vectors, the system can comprehensively evaluate the patient's health status from a global perspective. The vector is input into the pre-trained XGBoost model, and the XGBoost model will perform multi-symptom probability prediction for each patient based on the model weights and feature importance learned during the training process. According to the obtained multi-symptom probability outputs, including the probability of the patient having menstrual disorders, the probability of the patient having sleep disorders, the probability of the patient having mood swings, and the risk probability of the patient having endocrine re-fluctuations, the probability of each symptom is weighted and summed to obtain the total health risk score. This score provides a quantitative health indicator that can comprehensively reflect the patient's health status and potential risks. Compared with the traditional health assessment method based on a single indicator, the total health risk score not only considers the occurrence probabilities of multiple symptoms but also combines the importance of each symptom through a weighted method, making the assessment result more scientific and adaptable.

[0150] As an efficient machine learning model, XGBoost has powerful prediction capabilities. Especially when dealing with complex and multi-dimensional data, it shows high accuracy and generalization ability. Taking the abnormal feature vector A and the health status label as inputs, the XGBoost model can automatically extract feature importance and optimize the model based on historical training data, making the prediction more in line with the actual situation. This improvement in prediction ability can help clinical staff timely discover various health problems that patients may face (such as menstrual disorders, mood swings, etc.), and then adjust the nursing measures.

[0151] S5, a health assessment module, is used to divide the health level according to the health risk score and generate a corresponding nursing intervention plan;

[0152] Specifically, dividing the health level according to the health risk score and generating a corresponding nursing intervention plan means setting thresholds through statistical analysis of historical data and , and , if the health risk score is greater than or equal to the threshold , then the patient is in the high-risk period, and a reminder for a follow-up visit is pushed and the probability of typical symptoms is alarmed. If the health risk score is less than the threshold and greater than or equal to the threshold , then the patient is in the fluctuating period, and diet is pushed to the patient and advice on strengthening work and rest is given. If the health risk score is less than the threshold If so, the patient is in a stable period, continue the current care plan and push low-intensity exercise tasks to the patient.

[0153] The present invention quantifies the health status of patients through health risk scores and divides them into different levels, so as to realize the dynamic assessment of the health status of patients. The division of health levels can ensure the risk stratification management of patients, enabling high-risk patients to receive timely attention and treatment at the first time, patients in the fluctuating period to be effectively controlled by adjusting their lifestyles, and patients in the stable period to continue to maintain the existing care plan to avoid over-intervention. This grading system enhances the personalization and accuracy of health management and avoids over-treatment and inefficient intervention.

[0154] This embodiment also provides a method for evaluating the health degree of endocrine care, including:

[0155] Collect the patient's basic data and construct an individual state vector S, collect the patient's physiological data in real time and use a CNN model to extract deep features;

[0156] Use PCA to reduce the dimension of the deep features and then input them into a mutation point detection algorithm to detect periodic mutation points, obtain a set of mutation points, and parallelly use an SNN to identify spike activation responses to obtain a set of spike anomalies. Integrate the obtained set of mutation points and the set of spike anomalies into an abnormal feature vector A;

[0157] Collect the patient's historical abnormal feature vector C, construct a new sample set D with vector A and vector C, calculate the local density and the distance to the nearest high-density point of all samples based on the density peak clustering algorithm, and obtain the corresponding health status label;

[0158] Input the abnormal feature set A and the health status label into the XGBoost model, perform multi-symptom probability prediction and calculate the total health risk score;

[0159] Construct a logical rule tree to divide health levels and generate corresponding care intervention plans.

[0160] This embodiment also provides a computer device applicable to the situation of the method for evaluating the health degree of endocrine care, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the method for evaluating the health degree of endocrine care proposed in the above embodiment.

[0161] The computer device may be a terminal, which includes a processor, a memory, a communication interface, a display screen, and an input device connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, a carrier network, NFC (Near Field Communication), or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, a trackball, or a touchpad provided on the housing of the computer device, or an external keyboard, a touchpad, or a mouse, etc.

[0162] This embodiment also provides a storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the method for evaluating the health level of endocrine care as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM for short), Electrically Erasable Programmable Read-Only Memory (EEPROM for short), Erasable Programmable Read-Only Memory (EPROM for short), Programmable Read-Only Memory (PROM for short), Read-Only Memory (ROM for short), magnetic memory, flash memory, a magnetic disk, or an optical disc.

[0163] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. A health degree evaluation system for endocrine care, characterized in that: including, a state vector construction module, which is used to collect the patient's basic data and construct an individual state vector S, collect the patient's physiological data in real time, and extract deep features using a CNN model; an anomaly detection module, which is used to reduce the dimension of the deep features using PCA and then input them into a mutation point detection algorithm to detect periodic mutation points, obtain a set of mutation points, parallelly use an SNN to identify spike activation responses to obtain a set of spike anomalies, and integrate the obtained set of mutation points and the set of spike anomalies into an anomaly feature vector A; a health status classification module, which is used to collect the patient's historical anomaly feature vector C, construct a new sample set D with vector A and vector C, calculate the local density and the distance to the nearest high-density point of all samples based on the density peak clustering algorithm, and obtain the corresponding health status labels; a prediction module, which is used to input the anomaly feature set A and the health status labels into an XGBoost model, perform multi-symptom probability prediction and calculate the total health risk score; a health assessment module, which is used to divide the health level according to the health risk score and generate a corresponding nursing intervention plan; The calculation of the local density and the distance to the nearest high-density point of all samples based on the density peak clustering algorithm to obtain the corresponding health status labels means using the truncated kernel function density estimation method to calculate the number of points within the neighborhood of the sample point to obtain the density degree of the sample point; evaluating the minimum distance of each point to the points with higher density than it; Calculate the score of each sample point as a clustering center , including local density and minimum high-density distance; According to the actual needs of endocrine care, the health status is divided into three typical manifestations, including stable type, fluctuating type, and unstable type. The fixed number of clusters is 3. After calculating the clustering center scores of all samples, select the K = 3 points with the highest scores as the center points of the clustering. For each sample point, calculate its distance to the 3 center points and assign it to the nearest cluster; Finally, the current patient will be assigned to the corresponding health classification group to obtain the corresponding health status label.

2. The health degree evaluation system for endocrine care according to claim 1, characterized in that: The detection of periodic mutation points by inputting the deep features whose dimension is reduced by using PCA into a mutation point detection algorithm to obtain a set of mutation points includes: Collecting the feature vectors of the past n moments to construct a sliding window sample set, and using the current feature vector as the last sample. Calculate the mean and standard deviation for each column of the sample set, standardize each item to generate a new sample set, calculate the covariance matrix for the standardized sample set, perform eigenvalue decomposition on the covariance matrix, arrange the eigenvalues in descending order, calculate the cumulative contribution rate of the first m principal components in turn, intercept the first m feature vectors to form a principal component projection matrix and construct a principal component loading matrix with the eigenvalue diagonal matrix, and perform a linear transformation on the current standardized feature vector to obtain a reduced-dimension deep feature vector with dimension m; Traverse the principal component dimensions and calculate the total loading value on the original hormone features ; Finding out the principal component dimension with the maximum loading value and defining this dimension as the hormone dominant component; Retrospecting the last T moments, extracting the principal component sequence, and extracting the values of the principal component dimensions at each time point to form a hormone trend sequence; Performing a first-order discrete wavelet transform on the hormone trend sequence using Daubechies 6 wavelet; Traversing the adjacent data points in the detail coefficients and extracting the set M of time indices that satisfy the modulus maximum condition; Calculate the magnitude of the detail coefficients of all points in set M for point position screening to obtain a candidate set of mutation points; For each candidate point, intercept the local hormone trend subsequence with its center being ; For each subsequence, establish a state space model; Perform the prediction-update iteration process of Kalman filtering for each subsequence; After the filtering process is completed, obtain the smoothed trend sequence H of the candidate point; Initialize the smoothed trend mapping structure, create a trend mapping dictionary, register the smoothed sequence and the original sequence of each candidate point and encapsulate them into a set of structural tuples; Read the smoothed trend sequences corresponding to each candidate point in turn and construct a structural mutation fluctuation model and a control benchmark model based on the sequences; The maximum likelihood estimation is used to fit two model parameters, and the parameters and as well as the log-likelihood and are obtained. Based on the obtained and a Bayes factor is constructed. Summarize all time points that meet the mutation significance conditions, and calculate the maximum mutation amplitude of each retained mutation point in the smoothed sequence; Calculate the average mutation amplitude, maximum mutation amplitude, average mutation position, and mutation time variance of the mutation points and construct them into a mutation feature sub-vector.

3. The health degree evaluation system for endocrine care according to claim 2, characterized in that: Parallelly use SNN to identify the spike activation response to obtain a set of spike anomalies, and integrate the obtained set of mutation points and the set of spike anomalies into an anomaly feature set A, including: Read the original high-frequency time series data, including the heart rate variability sequence and the basal body temperature sequence, align the two-channel data in time, and splice them into a two-dimensional input matrix , standardize each column of data, calculate the standard deviation of the data, and set the dynamic judgment threshold as Q; Input line by line into the pre-trained SNN model for event activation judgment; If in any continuous 3-time-point window, any channel continuously exceeds the threshold Q, it is determined as a spike response, mark the current response window and record its time position; Integrate the spike response label and the spike response intensity into a spike anomaly feature vector; Integrate the mutation feature sub-vector and the spike anomaly feature vector to generate an anomaly feature vector A.

4. The health status assessment system for endocrine care according to claim 3, wherein: The step of inputting the abnormal feature vector A and the healthy status label into the XGBoost model for multi-symptom probability prediction and calculating the total health risk score means integrating the abnormal feature vector A and the healthy status label into the feature vector U, inputting the vector into the pre-trained XGBoost model, performing multi-symptom probability prediction for each patient, and according to the obtained multi-symptom probability output, weighting and summing the probabilities of each symptom. to obtain the total health risk score.

5. The health level assessment system for endocrine care according to claim 4, wherein: Said dividing health levels according to health risk scores and generating corresponding nursing intervention plans means setting thresholds and , and , if the health risk score is greater than or equal to the threshold , then the patient is in a high-risk period, a follow-up reminder is pushed and the probability of typical symptoms is alarmed. If the health risk score is less than the threshold and greater than or equal to the threshold , then the patient is in a fluctuating period, diet is pushed to the patient and suggestions for strengthening work and rest are given. If the health risk score is less than the threshold , then the patient is in a stable period, the current nursing plan is continued and low-intensity exercise tasks are pushed to the patient.

6. The health degree evaluation system for endocrine care according to claim 5, characterized in that: Collect the patient's basic data and construct an individual state vector S, and collect the patient's physiological data in real time and use a CNN model to extract deep features, including: The patient's basic data includes the type of diagnosed disease and treatment stage of the patient, three hormone indicators of the patient, average menstrual cycle length, emotional index score, and sleep quality score. After preprocessing the collected basic data, construct an individual state vector S; Medical staff fill in the type of disease diagnosed and the treatment stage of the patient, number the disease type as , and number the treatment stage as ; Read the patient's recent three hormone test data from the hospital LIS system, calculate the average value of the three hormone indicators as the current baseline level, extract the user's menstrual start time series in the recent n months, and calculate the average cycle length and the interval between the current date and the last menstruation , collect the patient's emotional index score using the PHQ-9 scale , collect the patient's sleep quality score using the PSQI scale ; Based on the individual state vector S, adjust the sampling period of the patient's physiological data, continuously collect the patient's physiological signals, behavioral indicators, and sleep and emotions for 24 hours, arrange all the data collected within 24 hours in chronological order, form a time series data matrix, and perform zero-mean normalization processing on the matrix; Input the normalized matrix into a pre-trained one-dimensional convolutional neural network to extract deep fusion feature representations and obtain a deep feature vector F.

7. A method for evaluating the health level of endocrine care, based on the endocrine care health level evaluation system according to any one of claims 1 to 6, characterized in that: Including, Collect the patient's basic data and construct an individual state vector S, collect the patient's physiological data in real time and use a CNN model to extract deep features; Use PCA to reduce the dimension of the deep features and then input them into a mutation point detection algorithm to detect periodic mutation points, obtain a set of mutation points, parallelly use SNN to identify the spike activation response to obtain a set of spike anomalies, and integrate the obtained set of mutation points and the set of spike anomalies into an anomaly feature vector A; Collect the patient's historical anomaly feature vector C, construct the vector A and the vector C into a new sample set D, and calculate the local density and the distance to the nearest high-density point of all samples based on the density peak clustering algorithm to obtain the corresponding health status label; Input the anomaly feature set A and the health status label into the XGBoost model for multi-symptom probability prediction and calculate the total health risk score; Construct a logical rule tree to divide the health level and generate a corresponding nursing intervention plan.

8. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the health level assessment method for endocrine care according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the health level assessment method for endocrine care according to any one of claims 1 to 6.

Images