CN-122004781-A - Dehydration early warning method and system based on urine biochemical time sequence characteristic data processing

CN122004781ACN 122004781 ACN122004781 ACN 122004781ACN-122004781-A

Abstract

The invention discloses a dehydration early warning method and a system based on urine biochemical time sequence characteristic data processing, which belong to the technical field of medical treatment and nursing intelligent monitoring and comprise the steps of biochemical characteristic fingerprint extraction, time sequence construction, trend reasoning, personalized baseline construction based on time window segmentation and risk classification output, wherein biochemical fingerprints are extracted from electrochemical waveforms formed by urine for each time through a one-dimensional convolutional neural network 1D-CNN, a pre-trained long-short-term memory network LSTM model is utilized for analysis, and the risk judgment is carried out by combining personalized historical baseline information after dynamic calibration of a user, so that a deep learning framework of 'biochemical characteristic fingerprint extraction and time sequence trend prediction' is formed, and finally early warning signals of corresponding grades can be output so as to be convenient for taking proper treatment or nursing measures.

Inventors

ZHAO GUODONG
WEI FEIFEI
Xia Fapeng
XU QINGHUA

Assignees

山东杰美医疗科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260326

Claims (10)

1. The dehydration early warning method based on urine biochemical time sequence characteristic data processing is characterized by comprising the following steps, SP 1 and biochemical feature fingerprint extraction The electrochemical impedance spectroscopy EIS or cyclic voltammetry CV is utilized to obtain the original waveform data corresponding to the urination event of the past times, and the one-dimensional convolutional neural network 1D-CNN is utilized to extract the information of the original waveform data corresponding to the urination event of each time, so as to obtain the corresponding multidimensional biochemical feature vector; SP 2 and time sequence construction Acquiring a multidimensional biochemical feature vector corresponding to the current single urination event, and recording the multidimensional biochemical feature vector as the current multidimensional biochemical feature vector ; Setting a period of time before the current single urination event as a preset time window, acquiring multidimensional biochemical feature vectors corresponding to the urination events in the preset time window, and respectively recording the multidimensional biochemical feature vectors as historical multidimensional biochemical feature vectors; Acquiring the current single urination event and timestamp information corresponding to each urination event in the preset time window, and calculating to acquire time interval characteristics between adjacent urination events according to each timestamp information to acquire a plurality of time interval characteristics; the current multidimensional biochemical feature vector is processed Performing feature stitching on each historical multidimensional biochemical feature vector and the corresponding time interval feature to obtain a plurality of mixed feature vectors containing time domain dimensions; Constructing each mixed eigenvector according to time sequence to form a multi-variable time sequence matrix ; SP 3 , trend reasoning By means of a pre-trained long-short-term memory network LSTM model; Matrix the multi-variable time series Inputting the LSTM model of the long-term and short-term memory network; the LSTM model of the long-short-term memory network infers and outputs an AI probability judging result, namely a dehydration risk probability value, based on the change slope and time characteristics of the electrolyte concentration implicit in the input sequence ; SP 4 personalized baseline construction based on time window slicing SP 4-1 , partitioning of time window, partitioning 24 hours into A preset time window; Locking mechanism when the user is at the present moment When urination is performed, the current time is firstly determined A time window to which the time window belongs; maintenance of SP 4-2 , historical baseline queues For each time window, respectively maintaining a corresponding historical data queue; the historical data queue comprises a historical biochemical index data queue and an average urination interval data queue; SP 4-3 , historical volatility Calculation of (2) Calculating the standard deviation of a sample of one biochemical index data queue in the historical biochemical index data queue to obtain the historical volatility ; SP 4-4 degree of deviation in urination interval Calculation of (2) Based on the average urination interval data queue, the adoption and history fluctuation The standard deviation algorithm of the same sample is used for calculating the standard deviation of the historical urination interval queue Mean value of ; Calculating the current urination interval The Z-Score standard Score of (c) is calculated as: ; SP 5 , risk classification output And combining AI probability judgment, statistical characteristic dynamic parameter adjustment and medical absolute safety fusing to form a three-dimensional cross validation mechanism and output risk level early warning.
2. The method for dehydration early warning based on urine biochemical time series characteristic data processing as set forth in claim 1, wherein said statistical characteristic dynamic parameter adjustment in step SP 5 comprises historical volatility The self-adaptive dynamic dehydration risk probability threshold value is obtained through calculation, and the calculation formula is as follows: Wherein, the Is a self-adaptive dynamic dehydration risk probability threshold; A preset reference probability threshold value; Historical volatility calculated by step SP 4-3 ; Based on historical volatility Is provided.
3. The dewatering early warning method based on urine biochemical time sequence characteristic data processing according to claim 2, wherein the adaptive dynamic dewatering risk probability threshold is corrected by combining with environmental parameters, namely, the environment is corrected, the specific process is that, Presetting an environment correction temperature limit value; Detecting an actual temperature value of the environment and comparing the actual temperature value with the environment correction temperature limit value; introducing a correction factor when the ambient actual temperature value is higher than the ambient correction temperature limit Correcting the self-adaptive dynamic dehydration risk probability threshold value, and <1.0, I.e By using And dehydration risk probability value A comparison is made.
4. The method for dehydration early warning based on urine biochemical time series characteristic data processing as set forth in claim 3, wherein said medical absolute safety fusing in step SP 5 introduces a physical numerical absolute safety upper limit of target biochemical index data As the highest priority fusing decision criterion; Based on multidimensional biochemical feature vectors corresponding to single urination events, current biochemical index data corresponding to the target biochemical index data type is obtained, and the current biochemical index data and the absolute upper safety limit of the physical value are obtained before the classification early warning signals are generated Comparing; Absolute safety upper limit of the physical value is exceeded by the current biochemical index data In the state of (2), ignoring the dehydration risk probability value And the comparison result with the self-adaptive dynamic dehydration risk probability threshold value forcedly triggers a high-level dehydration early warning signal to realize the execution of the highest priority fusing mechanism.
5. The dehydration early warning method based on urine biochemical time sequence characteristic data processing according to claim 4, wherein the risk level early warning output by the step SP 5 is a secondary early warning and a primary early warning in sequence from high to low; The triggering condition of the secondary early warning is a sub-condition ① or a sub-condition ②; sub-condition ① that the current value of the target biochemical marker data > the absolute upper safety limit of the physical value ; Sub-condition ② the dehydration risk probability value > And the degree of deviation of urination interval > , And (3) with Are set values; The triggering condition of the primary early warning is a sub-condition ① , or a sub-condition ② , ; Sub-condition ① , the dehydration risk probability value under the state of combining the self-adaptive dynamic dehydration risk probability threshold value and environmental parameter correction ; Sub-condition ② , the dehydration risk probability value in the state where the adaptive dynamic dehydration risk probability threshold is not modified by the environmental parameter 。
6. The method of claim 1 wherein the multi-dimensional biochemical characteristic vector comprises at least a sodium ion response current value, a potassium ion response current value, a urine conductivity and a urea nitrogen estimate.
7. The method for dehydration early warning based on urine biochemical time sequence characteristic data processing according to claim 1, wherein a forgetting gate mechanism is arranged on the long-short-term memory network LSTM model, so that the weight of long-term historical data on a current prediction result can be reduced in a self-adaptive manner, and biochemical index mutation in a preset time window is focused.
8. The dewatering early warning method based on urine biochemical time sequence characteristic data processing according to claim 1, wherein when historical data accumulation for forming the personalized historical base line is less than X days, preset clinical standard general data are automatically called to supplement until days required for forming the personalized base line construction are met, and smooth transition from a clinical standard general mode to a user personalized mode is achieved.
9. The dehydration early warning system based on urine biochemical time series characteristic data processing is used for implementing the dehydration early warning method based on urine biochemical time series characteristic data processing according to any one of claims 1 to 8, and is characterized by comprising, The data acquisition terminal is matched with the electrochemical impedance spectrum EIS or the cyclic voltammetry CV and is used for receiving raw waveform data formed by the electrochemical impedance spectrum EIS or the cyclic voltammetry CV based on urine electrochemical signals; The edge calculation module is in signal connection with the data acquisition terminal, is internally provided with a quantized compressed 1D-CNN feature extractor, and is used for receiving the original waveform data acquired by the data acquisition terminal, and performing biochemical feature fingerprint extraction on the original waveform data to form the multidimensional biochemical feature vector; and the server is in signal connection with the edge calculation module, runs with the long-short-term memory network LSTM model, and sequentially completes the tasks of trend reasoning, dynamic baseline calibration and risk classification output by utilizing the multidimensional biochemical feature vector.
10. The dewatering early warning system based on urine biochemical time sequence characteristic data processing according to claim 9, wherein the server drives an audible and visual alarm and a voice alarm, and the server is also connected with an intelligent terminal through signals.

Description

Dehydration early warning method and system based on urine biochemical time sequence characteristic data processing Technical Field The invention relates to the technical field of medical treatment and nursing intelligent monitoring, in particular to a dehydration early warning method based on urine biochemical time sequence characteristic data processing, and also relates to a system for implementing the method. Background In the nursing or treatment process of special crowds such as disabled old people, sick infants and the like, dehydration is one of important items for detection and monitoring, and according to different degrees of dehydration, the influence on human bodies and the nursing measures adopted are different, and the dehydration is mainly classified into the following grades. Slight dehydration, namely water loss accounting for about 1-3% of the weight, manifests as thirst, dry mouth, reduced urine volume, darkened urine color, fatigue, dizziness, dry skin, reduced elasticity and the like. Moderate dehydration, about 4-6% of body weight, serious thirst, increased heart rate, reduced blood pressure, muscle cramps or headaches, blurred consciousness, dysphoria, depressed eyes, and obviously dry skin. Heavy dehydration, which is characterized by water loss of about 7% or more of body weight, manifested by shock, impaired organ function, nervous system problems, and failure of thermoregulation. Therefore, when nursing or treating special people, the people need to pay attention to whether the water loss phenomenon and the degree of water loss occur or not, so that corresponding measures can be taken in time for improvement. In the prior art, biochemical index data of urine is usually detected and analyzed, a corresponding parameter threshold is set, a detected and analyzed value is compared with the set parameter threshold, whether dehydration and the degree thereof are judged according to a comparison result, for example, when the urine specific gravity is detected to be more than 1.025 or the conductivity is detected to be more than 20mS/cm, the dehydration phenomenon is indicated, and then an alarm can be implemented. The dehydration judgment and alarm mode mainly has the following defects: 1. It is difficult to eliminate the disturbance of diet. For example, the electrolyte of urine can temporarily surge for the old who just drinks thick soup or eat salty food, at this time, the conductivity of urine is liable to be larger than the set threshold value, and if only the alarm is triggered in sequence, the false positive alarm can be formed with high probability. 2. It is difficult to eliminate interference of individual differences. For some old people with renal insufficiency, the urine concentration capability is impaired, and even if the old people are severely dehydrated, the specific gravity of the urine can not be obviously increased, so that the alarm is difficult to trigger in time, namely false negative is formed. Therefore, simply relying on the simple comparison of the current detection analysis value and the set threshold value, it is difficult to accurately judge the dehydration and degree of the nursing crowd, so that the unnecessary nursing or the untimely nursing problem is caused. Disclosure of Invention The invention aims to solve the technical problem of providing the dehydration early warning method based on the urine biochemical time sequence characteristic data processing, which can effectively distinguish whether the change is physiological fluctuation caused by feeding, drinking water and the like or pathological trend caused by continuous dehydration when certain biochemical index data in urine is increased, and can remarkably improve the early warning accuracy of dehydration, electrolyte disorder and the like of special nursing people so as to take correct nursing measures. In order to solve the technical problems, the technical proposal of the invention is that the dehydration early warning method based on urine biochemical time sequence characteristic data processing comprises the following steps, SP 1 and biochemical feature fingerprint extraction The electrochemical impedance spectroscopy EIS or cyclic voltammetry CV is utilized to obtain the original waveform data corresponding to the urination event of the past times, and the one-dimensional convolutional neural network 1D-CNN is utilized to extract the information of the original waveform data corresponding to the urination event of each time, so as to obtain the corresponding multidimensional biochemical feature vector; SP 2 and time sequence construction Acquiring a multidimensional biochemical feature vector corresponding to the current single urination event, and recording the multidimensional biochemical feature vector as the current multidimensional biochemical feature vector; Setting a period of time before the current single urination event as a preset time window, acquiring multidimensional biochemical