[0038] Example 2
[0039] This embodiment provides a data processing method for an intelligent health monitoring system, including:
[0040] The data processing method of the intelligent health monitoring system includes:
[0041] The information collection component collects the evaluation index information of human health;
[0042] The data processing center preprocesses the evaluation index information to obtain the user's health score, and compares the health score with the preset value. When the health score is less than the preset value, the DNN deep neural network algorithm is used to evaluate the corresponding health score. Analyze the information to get the disease prediction table;
[0043] The display terminal displays the health score and disease prediction table.
[0044] As a preferred implementation of this embodiment, the evaluation index information collected by the information collection component includes heart rate information collected by the heart rate module, blood pressure information collected by the blood pressure module, and blood oxygen information collected by the blood oxygen module.
[0045] As a preferred implementation of this embodiment, the data processing center performs the following steps when preprocessing the evaluation index information:
[0046] Step S51: Take one day as a period of information collection, and divide the period into at least three time periods according to the difference of the human body's heart rate in different time periods of the day.
[0047] It should be noted that under normal health conditions, the human body’s health index information during a day is not exactly the same, so processing and analyzing the information in a day segment can make the analysis results more accurate. The health index information includes heart rate information, blood pressure information, and blood oxygen information. Since the heart rate information changes significantly throughout the day, the human body’s health index information at different time periods is matched according to the difference in heart rate information, and 4 consecutive samples are calculated through experiments The average value of the heart rate at the time is compared with the set threshold to obtain the user's time of falling asleep and waking up, and then classify the data. Preferably, in this embodiment, the sliding window method is used to divide the time period described as follows:
[0048] The moment of falling asleep is:
[0049] (x 1 (t 1 -1)+x 1 (t 1 -2)+...+x 1 (t 1 -s+1)+x 1 (t 1 -s))/(s*x 1 (t 1 ))≥C%;
[0050] Where x 1 Indicates the heart rate, t 2 Represents the moment of falling asleep, C% represents the set segmentation threshold, in the formula, s represents the length of the sliding window, n represents the sampling time before falling asleep, where n = 1, 2, ..., s, in this embodiment specifically They are the first to fourth sampling moments, that is, the value of s is 4 in this embodiment.
[0051] Wake up time:
[0052] (x 1 (t 2 +1)+x 1 (t 2 +2)+...+x 1 (t 2 +s-1)+x 1 (t 2 +s))/(s*x 1 (t 2 ))≥C%;
[0053] Where t 2 Indicates the moment of falling asleep.
[0054] Further, after obtaining the time of falling asleep and the time of waking up corresponding to the lunch break and night sleep respectively, the data is divided by the time of falling asleep and the time of waking up as the dividing point. Preferably, the data is divided according to the difference in human behavior during the day There are three stages: daytime activity, lunch break, and night sleep. And perform dimensionality reduction processing on the data in different time periods. Among them, the specific process of principal component analysis (PCA) for data dimensionality reduction processing is expressed as follows:
[0055] First, the data matrix obtained after standardizing the collected data samples is:
[0056]
[0057] In the formula, it represents n rows of p-dimensional data (the number of physiological indicators is p, n can be N according to different time periods 1 , N 2 Or N 3 ),
[0058] Then, calculate the correlation coefficient matrix of each indicator:
[0059]
[0060] Where r ij =cov(x i ,x j ).
[0061] Let E denote the identity matrix and get:
[0062] (λE-R)a=0;
[0063] Calculate the eigenvalues and eigenvectors of the correlation coefficient matrix R as:
[0064] Characteristic value: λ 1 ,λ 2....λ p , Feature vector: a i =(a i1 ,a i2 ,...,a ip ),i=1,2...,p
[0065] Then the data after PCA dimensionality reduction is expressed as:
[0066] P=aX;
[0067] It is worth pointing out that the new component matrix after dimensionality reduction corresponding to different time periods is expressed as: P 1 ,P 2 ,P 3.
[0068] Step S52: Calculate the reference points of the health information in different time periods:
[0069]
[0070] Where Indicates the x coordinate of the reference point during T period, Y coordinate of reference point in T period, N T Indicates the number of data points whose status is healthy in the corresponding T period, Represents the principal component 1 of the i-th data after PCA dimensionality reduction of the corresponding health data point in the T period, Represents the principal component 2 of the i-th piece of data after PCA dimensionality reduction of the corresponding health data point in the T period.
[0071] Step S53: Calculate the Euclidean distance between the evaluation index information collected in different time periods and the health information reference point in the corresponding time period:
[0072]
[0073] Where Represents the principal component 1 of the i-th data of the evaluation index corresponding to the T period, Corresponding to the principal component 1 of the i-th data of the evaluation index in the T period, the user's health score is obtained according to the Euclidean distance.
[0074] Preferably, the principal component analysis method is used to reduce the dimensionality of the collected data to facilitate flat display, wherein the part of the health index information after time division is shown in Table 1 below, where systolic blood pressure and diastolic blood pressure both represent blood pressure:
[0075] Table 1 Partial Health Index Information Table
[0076] Serial number
[0077] It should be noted that there is a certain correlation between different health indicators. After experiments, the correlation between the health indicators is shown in Table 2 below (take the daytime activity period as an example):
[0078] Table 2 Correlation between various health indicators
[0079]
[0080] Furthermore, the contribution rates of the first four characteristic roots of the above coefficient matrix are shown in Table 3 below. Among them, the contribution rate of the characteristic roots reflects the degree of interpretation of the health index data information by the characteristic roots, that is, the relationship between the health index data and the disease. Strong and weak relevance:
[0081] Table 3 The contribution rate of characteristic roots
[0082] Characteristic root
[0083] As mentioned in Table 3 above, the cumulative contribution rate of the first two feature roots is above 85%, that is, 53.9338+31.4628=85.3966. According to the principal component extraction cumulative standard that reaches 80, it is considered that the first two feature roots contain most of the data Information, so choose two principal components to reduce the dimensionality of the data. It should be noted that the dimensionality reduction process is convenient for the plane representation of the data, and the distribution of the sample health status data points can be observed in the plane coordinates, which realizes the visualization in the data analysis process. In specific experiments, the feature vectors corresponding to the first two feature roots are shown in Table 4 below:
[0084] Table 4 The eigenvector situation table corresponding to the characteristic root
[0085]
[0086] The health status is scored based on Euclidean distance. Different time periods are based on a large amount of normal health data in the corresponding time period in the database to calculate the average value of dimensionality reduction to obtain the health reference points of different periods as shown in Table 5 below:
[0087] Table 5 Health reference points in different time periods in a cycle
[0088]
[0089] According to the physiological data of different known health states in the database, the Euclidean distance value of each data sample from the health standard point at different time periods is calculated separately, and the Euclidean distance corresponding to the misjudgment rate of illness is less than 1% as the health threshold. The health threshold is R=35, that is, when the Euclidean distance between the sample and the corresponding standard point in a certain period of time is greater than the threshold R=35, the user may be diagnosed as unhealthy.
[0090] For specific questions, the relationship between the Euclidean distance and health status determined in this embodiment is shown in Table 6 below:
[0091] Table 6 Relationship between Euclidean distance and health status
[0092]
[0093] Taking an old man A as an example, calculate the average Euclidean distances of old man A from the standard point of the corresponding time period during the night sleep period, daily activity period, and lunch break respectively. The average values are 41.6, 47.5, and 46.8. 10,12,2, calculated by the health status score: y i =-1.143D i +100; (i is 1, 2, and 3, representing the three time periods of morning, middle and evening respectively, y i That is the individual health score of the corresponding period, a 1,2,3 Respectively represent different time periods, D 1,2,3 Respectively represent the Euclidean distance between the physiological data and the standard point in different time periods, y is the final score of the individual's health, and 0 is taken below 0)
[0094] The relationship between health score and health status is shown in Table 7 below:
[0095] Table 7 Relationship between health score and health status
[0096]
[0097] The final health status score of old man A is y=49, which is in a poor state. When the health status score is lower than 60 points, the disease prediction of the individual is initiated.
[0098] Step S61: Obtain disease information in different time periods as the input of the DNN deep neural network algorithm for model training, so as to obtain the evaluation index information in different time periods and the prediction model of the corresponding disease.
[0099] Step S62: When the health score is less than the preset value, input the evaluation index information corresponding to the health score into the prediction model in the corresponding time period to obtain the disease prediction table
[0100] Further, the DNN deep neural network algorithm is used for disease analysis. It should be noted that the disease prediction method uses the collected original disease data as a training set, and uses DNN to generate a prediction algorithm model to achieve disease prediction for individual health status.
[0101] Specifically, disease prediction gives an individual the probability of suffering from various diseases based on the individual's health index data, and gives an early warning of a high probability disease. As a preferred implementation of this embodiment, firstly obtain disease information in different time periods as the input of the DNN deep neural network algorithm for model training, so as to obtain the evaluation index information in different time periods and the corresponding disease prediction model as:
[0102]
[0103] Where Represents the output prediction model, k represents the total number of layers of the DNN deep neural network in this embodiment, q represents the number of neurons in the hidden layer, Represents the connection weight between the jth neuron and the hth neuron in layer k-1, Represents the threshold of the hth neuron in layer k-1, Represents the output value of the jth neuron in the k-1 layer, where, with Obtained through training of disease information obtained in different time periods. It should be noted that the training data used in different time periods are inconsistent, so the parameters of the prediction model will also be different.
[0104] Then the health index information corresponding to the 49 points of the health status of old man A is used as the input of the corresponding model. In this embodiment, the DNN deep neural network has a total of 120 layers (4 neurons in the input layer, and 8 neurons in each layer in the middle layer. The output layer is the 4 output neurons representing the probabilities of 4 kinds of diseases). The network output value is used to estimate the probability of related diseases to realize the assessment and monitoring of the physiological condition of the human body.
[0105] Due to the different DNN model parameters corresponding to different diseases in different time periods, model switching is required. For hypertension disease prediction, the output values of the neural network output layer of the three time periods are 0.94, 0.97, and 0.96 respectively, and the weight ratio is the time ratio. , The overall probability of elderly A suffering from hypertension is 95.7%. The probability prediction methods for other diseases are the same, and the prediction results for the probability of elderly A disease are shown in Table 8 below:
[0106] Table 8 Prediction results for the probability of elderly A disease
[0107] Related diseases
[0108] Then, the health score and disease prediction table of old man A are displayed through the display terminal. Preferably, in this embodiment, the display terminal is set as a wristband, which is convenient to carry and can intuitively display results.
[0109] As mentioned above, the intelligent health monitoring system and its data processing method provided by the present invention include an information collection component for collecting human health evaluation index information, a data processing center wirelessly connected with the information collection component, and a data processing center connected The information collection component is worn on the user’s body; the data processing center is used to preprocess the evaluation index information collected by the information collection component to obtain the user’s health score. When the health score is less than the preset value, the data processing center is also used The DNN deep neural network algorithm is used to analyze the evaluation index information corresponding to the health score, and the disease prediction table is obtained. The display terminal is used to display the health score and also used to display the disease prediction table; the monitoring system of the present invention fully considers the difference of data, Multi-period processing and analysis of physiological index data reduces the interference caused by time factors to data analysis. The DNN algorithm is used to make the analysis results of health status more accurate and reliable, and the linkage between various devices is close, the implementation cost is low, and the implementation cost is smart. High degree of chemistry.
