Multi-parameter integrated intravenous infusion extravasation early warning method and device

Abstract

The invention relates to a multi-parameter integrated intravenous infusion extravasation early warning method and device. The method comprises the steps that firstly, a physiological parameter set of a patient is collected through medical equipment, and a structured feature data set is constructed through standardization processing; secondly, dynamic modeling is conducted on the feature data set through a deep learning time sequence model, and a self-adaptive datum line and a potential exosmosis risk sequence are generated; and then, carrying out signal processing on the risk sequence, extracting multi-device associated characteristics, calculating a collaborative risk coefficient, and iteratively updating the adaptive reference line to obtain a risk judgment result. And finally, determining a final extravasation risk value according to a judgment result, comparing the final extravasation risk value with a historical threshold value, and then triggering a dynamic early warning signal of a corresponding level. By means of the method, exosmosis risks can be accurately evaluated, the problem that the adaptability of fixed threshold monitoring is insufficient is effectively solved, an automatic and intelligent solution is provided for clinical infusion exosmosis prevention, the nursing burden is reduced, and the safety of a patient is improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of medical information technology, and particularly relates to a multi-parameter integrated venous infusion exosmosis early warning method and device. BACKGROUND

[0002] As a common clinical treatment method, venous infusion is widely used. However, infusion exosmosis frequently occurs. As described in patent CN202110615158, patients with severe conditions are more likely to have infusion drug exosmosis due to long-term infusion of specific drugs, self-activity, consciousness state and other factors, which not only causes local tissue damage, phlebitis, and even tissue ulceration and necrosis, but also affects the treatment effect of drugs and increases the nursing burden of severe conditions. At present, the existing technology has obvious deficiencies in infusion exosmosis monitoring. Some methods for judging exosmosis by monitoring the infusion speed do not directly contact the exosmosis site, and only rely on the infusion flow rate, which cannot accurately monitor the condition of the puncture site, and often causes false positives and false negatives. The method of placing an alarm device on the puncture site to monitor the blood flow rate, although it improves the accuracy of the alarm to some extent, is high in cost and cannot be recycled. Moreover, the existing technology mostly stays at the alarm stage and cannot block the liquid exosmosis at the puncture site in the first time. At the same time, the existing infusion monitoring and early warning mostly rely on fixed threshold judgment and cannot adaptively monitor the risk under different infusion conditions. As pointed out in patent CN202410637265, the dynamic change adaptability of infusion monitoring and early warning is obviously insufficient and cannot meet the complex and variable clinical needs. SUMMARY

[0003] Therefore, it is necessary to provide a multi-parameter integrated venous infusion exosmosis early warning method and device capable of dynamically monitoring and accurately evaluating the risk of infusion exosmosis in view of the above technical problems.

[0004] In a first aspect, the application provides a multi-parameter integrated venous infusion exosmosis early warning method, comprising:

[0005] Obtaining a set of physiological parameters of a patient, standardizing the set of physiological parameters in combination with individualized characteristics to obtain a structured feature data set.

[0006] According to the feature data set, a deep learning time series model is used for dynamic modeling to generate an adaptive baseline and a potential exosmosis risk sequence.

[0007] The potential exosmosis risk sequence is subjected to signal processing, device-related features are extracted and a coordination coefficient is calculated, and the adaptive baseline is updated to obtain a risk determination result.

[0008] According to the risk determination result, a final exosmosis risk value is determined and compared with a historical threshold value, and a dynamic early warning signal is triggered according to the result.

[0009] In one of the embodiments, a deep learning time series model is employed to generate an adaptive baseline and a potential extravasation risk sequence based on the feature dataset, including:

[0010] Time series data extraction is performed on the feature dataset, and a deep learning algorithm is used to construct a dynamic adaptive baseline.

[0011] Based on the dynamic adaptive baseline, the deviation value at each time point is calculated to obtain the initial value of the potential extravasation risk.

[0012] The correlation features of the initial extravasation risk value and historical extravasation event data are extracted, and the corresponding risk weighting coefficient is determined through a machine learning algorithm.

[0013] The initial risk value is weighted and adjusted using the risk weighting coefficient to generate a potential extravasation risk sequence.

[0014] In one of the embodiments, the initial value of the potential extravasation risk is obtained by the following calculation formula:

[0015]

[0016] wherein, represents the initial value of the potential extravasation risk, X t represents the feature vector at time t, represents the predicted value of the dynamic adaptive baseline at time t, Var(X t-k:t-1 ) represents the variance of the feature vector at the previous k time points, λ represents the decay coefficient, α i represents the importance weight of the i-th historical extravasation event, DTW(X t-k:t-1 , C i ) represents the dynamic time warping distance between the current sliding window and the i-th historical extravasation event pattern, and σ represents the Sigmoid function.

[0017] In one of the embodiments, after generating the potential extravasation risk sequence, it further includes:

[0018] The anomaly fluctuation features in the time series are extracted from the potential extravasation risk sequence using a signal processing algorithm to generate an anomaly value sequence.

[0019] The anomaly value sequence is compared with a preset threshold, and high-risk points are selected through a threshold discrimination model.

[0020] Contextual environmental data is obtained from the high-risk points, and multi-medical device correlation features are extracted using feature engineering methods; the medical devices include but are not limited to infusion pumps, pressure sensors, and temperature sensors.

[0021] Based on the device correlation features, a collaborative filtering algorithm is used to calculate the collaborative risk coefficient between devices.

[0022] The risk determination result is generated by iteratively updating the dynamic adaptive baseline using the synergistic risk coefficient.

[0023] In one embodiment, the synergistic risk coefficient between devices is calculated by the following steps:

[0024]

[0025] wherein C ij represents the synergistic risk coefficient between device D i and D j , S ij represents the eigenvector cosine similarity between device D i and D j , δ(t) represents the time decay factor, ω ij represents the association weight matrix element between device D i and D j , n represents the total number of devices, and ∈ represents the minimum value.

[0026] In one embodiment, the final extravasation risk value is determined according to the risk determination result and compared with the historical threshold value, and a dynamic early warning signal is triggered according to the result, including:

[0027] Based on the risk determination result, the statistical characteristics of the patient's historical data and current physiological parameters are determined to determine the final extravasation risk value.

[0028] The dynamic time warping algorithm is used to time-align the final extravasation risk value sequence and the historical threshold value sequence, calculate the Euclidean distance and the morphological similarity, and obtain the comparison result.

[0029] The comparison result is quantitatively analyzed, and the sliding window algorithm is used to extract the characteristic parameters of the risk value deviating from the historical threshold value.

[0030] When the characteristic parameters trigger the preset conditions, the K-nearest neighbor algorithm is used to screen the matched early warning rule template based on the dynamic early warning rule library.

[0031] The parameters in the early warning rule template are dynamically adjusted and instantiated to obtain the early warning rule, and the dynamic early warning signal of the corresponding level is generated.

[0032] In one embodiment, the method further comprises:

[0033] Obtaining a set of physiological parameters of a patient; the set of physiological parameters contains multiple individualized index data.

[0034] Extracting individualized feature information from the set of physiological parameters, and processing the individualized feature information using a standardization algorithm to obtain a standardized feature data set.

[0035] The feature correlation analysis is performed on the standardized feature dataset to determine a correlation weight matrix between features.

[0036] The feature combination with high correlation is screened from the correlation weight matrix to obtain a structured feature subset.

[0037] If the dimension of the feature subset exceeds a preset threshold, a principal component analysis algorithm is used for dimension reduction processing to obtain an optimized feature dataset.

[0038] In a second aspect, the application further provides a multi-parameter integrated intravenous infusion exosmosis early warning device, which comprises:

[0039] A data processing modeling module is configured to obtain a physiological parameter set of a patient, perform standardized processing in combination with individualized features, and obtain a structured feature dataset.

[0040] A risk judgment and evaluation module is configured to perform dynamic modeling on the feature dataset by using a deep learning time series model to generate an adaptive baseline and a potential exosmosis risk sequence, and is further configured to perform signal processing on the potential exosmosis risk sequence, extract device-related features, calculate a coordination coefficient, and update the adaptive baseline to obtain a risk judgment result.

[0041] An early warning rule generation module is configured to determine a final exosmosis risk value according to the risk judgment result, compare the final exosmosis risk value with a historical threshold value, and trigger a dynamic early warning signal according to a result.

[0042] In a third aspect, the application further provides a computer device, which comprises a memory and a processor, the memory stores a computer program, and the processor implements the foregoing method when executing the computer program.

[0043] In a fourth aspect, the application further provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the foregoing method.

[0044] The multi-parameter integrated venous infusion exosmosis early warning method, device, computer equipment and storage medium, first, the physiological parameter set of the patient is collected through the medical equipment, the individualized characteristics such as age and medical history are integrated, and the structured feature data set is constructed after standardization processing. Secondly, the feature data set is dynamically modeled by using a deep learning time sequence model, and an adaptive baseline and a potential exosmosis risk sequence are generated. Then, the risk sequence is processed, the multi-device correlation features are extracted, and the cooperative risk coefficient is calculated, so as to iteratively update the adaptive baseline, and the risk determination result is obtained. Finally, the final exosmosis risk value is determined according to the determination result, and the corresponding level of dynamic early warning signal is triggered after comparing with the historical threshold value. The method realizes the accurate evaluation of the exosmosis risk through multi-link cooperation. The standardization processing eliminates the dimensional difference of the data, and guarantees the comparability of the features; the deep learning model dynamically adapts to the individual differences, and improves the accuracy of the baseline generation; the device correlation feature extraction and the cooperative coefficient calculation strengthen the fusion analysis ability of the multi-source data; the dynamic early warning mechanism combines the historical threshold value and the real-time risk value, and realizes the adaptive adjustment of the early warning rule. The overall process forms a closed loop from data acquisition to early warning triggering, effectively solves the problem of insufficient adaptability of the traditional fixed threshold monitoring, and provides automatic and intelligent technical support for clinical infusion exosmosis risk prevention and control. BRIEF DESCRIPTION OF DRAWINGS

[0045] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related art, the drawings needed to be used in the embodiments or the related art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creating any creative labor.

[0046] Figure 1 A flowchart of a multi-parameter integrated venous infusion exosmosis early warning method provided by the embodiment of the present application;

[0047] Figure 2 A structural block diagram of a multi-parameter integrated venous infusion exosmosis early warning device provided by the embodiment of the present application. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical solutions and advantages of the present application more clear, the present application will be further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

[0049] In one embodiment, as shown in Figure 1 A multi-parameter integrated venous infusion exosmosis early warning method provided by the present application can include the following steps:

[0050] In step S101, a set of physiological parameters of a patient is acquired, and individualized features are standardized to obtain a structured feature dataset.

[0051] Specifically, various professional medical sensors such as infusion flow rate sensors, pressure sensors, skin temperature sensors, etc. are used to accurately and continuously collect a set of physiological parameters of the patient during intravenous infusion, covering infusion flow rate, puncture site pressure, local skin temperature, and vital sign data such as patient heart rate, blood pressure, etc. At the same time, the individualized features of the patient are collected, including age, gender, medical history, allergy history, current disease, infusion drug type and concentration, etc. Subsequently, data cleaning techniques are used to remove noise, outliers and duplicate data in the collected data, ensuring the accuracy and consistency of the data. Then, normalization algorithms are used to map physiological parameters and individualized feature data of different dimensions to a unified standard range, eliminating the interference of dimension differences on subsequent analysis. Finally, valuable features are extracted from the cleaned and normalized data through feature engineering methods, and a structured feature dataset is constructed.

[0052] In step S102, a deep learning time series model is used to dynamically model the feature dataset to generate an adaptive baseline and a potential extravasation risk sequence.

[0053] Specifically, the feature dataset is input into a deep learning time series model such as a long short-term memory network (LSTM) or a gated recurrent unit (GRU), which learns the variation and dependency of the patient's physiological parameters in the time dimension through the recurrent structure and gating mechanism of the model. Based on historical data and current input, the model dynamically generates an adaptive baseline to represent the trend of physiological parameters and device operating indicators under normal infusion conditions. At the same time, the model outputs potential extravasation risk values at each time point, which are concatenated to form a potential extravasation risk sequence.

[0054] In step S103, the potential extravasation risk sequence is processed, device-related features are extracted, and a collaborative coefficient is calculated to update the adaptive baseline and obtain a risk determination result.

[0055] Further, wavelet transform and other signal processing algorithms are used on the potential extravasation risk sequence to extract abnormal fluctuation features and generate an abnormal value sequence, which is then compared with a preset threshold to filter high-risk points. Context data of multiple devices such as infusion pumps and pressure sensors is obtained from the high-risk points, and device operating parameter-related features are extracted through feature engineering. A collaborative filtering algorithm is used to calculate the collaborative risk coefficient between devices. This coefficient integrates feature cosine similarity, time decay factor and device correlation weight, and is used to iteratively update the dynamic adaptive baseline to recalculate the risk sequence and output the risk determination result. The risk determination result includes the potential extravasation risk sequence recalculated based on the updated adaptive baseline and the quantified risk level.

[0056] Step S104, determine the final extravasation risk value according to the risk determination result and compare it with the historical threshold value, and trigger the dynamic warning signal according to the result.

[0057] Specifically, based on the risk determination result, the statistical characteristics of the patient's historical data and the current physiological parameters are integrated, and the final extravasation risk value is calculated by weighting through a risk weight model. The dynamic time warping algorithm is used to align the risk value sequence and the historical threshold value sequence in time, calculate the Euclidean distance and the morphological similarity, and extract the deviation characteristic parameters using a sliding window. When the parameters trigger the preset conditions, the K-Nearest Neighbor algorithm is used to match the warning rule templates based on the dynamic warning rule library, and the rules are instantiated after dynamic adjustment of the parameters, triggering the dynamic warning signal of the corresponding level.

[0058] The above-mentioned multi-parameter integrated intravenous infusion extravasation warning method first collects the physiological parameter set of the patient through the medical device, integrates the individualized characteristics such as age and medical history, and constructs a structured feature data set after standardization processing. Secondly, a deep learning time series model is used to dynamically model the feature data set, generate an adaptive baseline and a potential extravasation risk sequence. Then, the risk sequence is processed, the multi-device correlation features are extracted, and the collaborative risk coefficient is calculated, so as to iteratively update the adaptive baseline and obtain the risk determination result. Finally, according to the determination result, the final extravasation risk value is determined, and after comparison with the historical threshold value, the dynamic warning signal of the corresponding level is triggered. This method realizes the accurate evaluation of extravasation risk through multi-link cooperation. Standardization processing eliminates the dimensional difference of data, ensuring the comparability of features; the deep learning model dynamically adapts to individual differences, improving the accuracy of baseline generation; the extraction of device correlation features and the calculation of collaborative coefficients strengthen the fusion analysis capability of multi-source data; the dynamic warning mechanism combines historical threshold value and real-time risk value, realizing the adaptive adjustment of warning rules. The overall process forms a closed loop from data acquisition to warning triggering, effectively solving the problem of insufficient adaptability of traditional fixed threshold monitoring, and providing automatic and intelligent technical support for clinical infusion extravasation risk prevention and control.

[0059] In one embodiment, the deep learning time series model is used to dynamically model the feature data set to generate an adaptive baseline and a potential extravasation risk sequence, which can include the following steps:

[0060] Step S201, time series data extraction is performed on the feature data set, and a deep learning algorithm is used to construct a dynamic adaptive baseline.

[0061] Preferably, the structured feature dataset is first subjected to time series data extraction, arranging time-dependent physiological parameters (such as infusion flow rate, puncture site pressure, skin temperature, etc.) in chronological order into sequence data. Subsequently, the processed time series data is input into a long short-term memory (LSTM) model. The LSTM processes sequence data through a gating mechanism (forget gate, input gate, output gate), can effectively learn and remember the time-dependent relationship in long sequences, and avoid the problem of gradient vanishing or gradient explosion. After multiple rounds of training, the LSTM model can learn the variation pattern of the patient's physiological parameters under normal infusion conditions, and output a dynamic adaptive baseline. The adaptive baseline is not a fixed threshold, but a parameter variation range that is dynamically adjusted according to the individual characteristics of the patient and the infusion process.

[0062] Step S202, calculate the deviation value of each time point based on the dynamic adaptive baseline, and obtain the initial value of the potential extravasation risk.

[0063] Step S203, extract the correlation features of the initial extravasation risk value and the historical extravasation event data, and determine the corresponding risk weighting coefficient through a machine learning algorithm.

[0064] Preferably, the correlation features of the initial risk value and the historical extravasation event data are extracted to construct a feature vector set F = {f1, f2, f3,..., f n}, which includes but is not limited to: time series fluctuation features (such as root mean square error, autocorrelation coefficient), parameter change trend features (such as first derivative, second derivative), statistical distribution features (such as skewness, kurtosis). The feature vector and the historical extravasation event label L = {l1, l2, l3,..., l n} (where l i ∈{0,1} represents whether extravasation occurs) are input into a machine learning algorithm for training.

[0065] A random forest (Random Forest) algorithm is used to construct a classification model, and the importance weight of each feature is determined through ensemble learning of decision trees. During training, the model generalization ability is evaluated by out-of-bag error (Out-of-Bag Error), and the optimization parameters include the number of trees T, the maximum depth d, etc. Finally, the risk weighting coefficient vector W = {w1, w2, w3,..., w n} is output, where each coefficient w i represents the contribution of feature f i to the prediction of extravasation risk, and the calculation formula is: where represents the contribution of feature f iGini Importance. The coefficient vector is used for risk value weighting adjustment in the subsequent steps, improving the sensitivity and adaptability of the prediction model to different features.

[0066] Step S203, the initial risk value is weighted and adjusted by using the risk weighting coefficient to generate a potential extravasation risk sequence.

[0067] First, time series data is extracted from the feature data set and input into the long short-term memory network (LSTM) for training to construct a dynamic adaptive baseline. The adaptive baseline can reflect the normal trend of changes in various parameters during the infusion process of the patient. Then, the actual data at each time point is compared with the dynamic adaptive baseline to calculate the deviation value, and by analyzing the size and change rule of the deviation value, the initial value of the potential extravasation risk is obtained. Next, the initial risk value and the associated features in the historical extravasation event data are extracted, and machine learning algorithms such as random forest and gradient boosting tree are used for training to determine the risk weighting coefficient corresponding to each feature. Finally, the risk weighting coefficient is applied to the initial risk value for weighted adjustment to obtain a more accurate potential extravasation risk sequence.

[0068] This embodiment constructs a dynamic adaptive baseline through LSTM, overcoming the limitations of traditional fixed thresholds that cannot adapt to individual differences, and can more accurately capture parameter abnormalities. Combined with the training of risk weighting coefficients based on historical data, the past experience is fully utilized, making the risk assessment more in line with the actual situation. Compared with the single real-time data or simple threshold judgment method, through multi-step and multi-dimensional data processing and analysis, the accuracy and reliability of the extravasation risk assessment are significantly improved, providing strong support for timely discovery and handling of infusion extravasation problems in clinical practice.

[0069] In one of the embodiments, the initial value of the potential extravasation risk can be obtained by the following calculation formula:

[0070]

[0071] wherein, represents the initial value of the potential extravasation risk, X t represents the feature vector at time t, represents the predicted value of the dynamic adaptive baseline at time t, Var(X t-k:t-1 ) represents the variance of the feature vector of the previous k time points, λ represents the decay coefficient, α i represents the importance weight of the i-th historical extravasation event, DTW(X t-k:t-1 , C i represents the dynamic time warping distance between the current sliding window and the i-th historical extravasation event pattern, and σ represents the Sigmoid function.

[0072] The embodiment improves the risk assessment accuracy by multi-dimensional feature fusion. The numerator calculates the deviation degree of the current feature from the baseline, and the denominator performs volatility normalization to suppress data noise interference. The index part introduces historical extravasation event pattern matching, and captures the dynamic changes of the time sequence characteristics by the DTW algorithm. The Sigmoid function maps the result to the interval [0, 1] for subsequent threshold judgment.

[0073] In one of the embodiments, after generating the potential extravasation risk sequence, the following steps can also be included:

[0074] In step S301, the abnormal fluctuation features in the time sequence are extracted from the potential extravasation risk sequence by using a signal processing algorithm to generate an abnormal value sequence.

[0075] In step S302, the abnormal value sequence is compared with a preset threshold, and a high-risk point is screened out by a threshold discrimination model.

[0076] In step S303, context environment data is obtained from the high-risk point association, and multi-medical device association features are extracted by using a feature engineering method. The medical devices include but are not limited to infusion pumps, pressure sensors, and temperature sensors.

[0077] In step S304, a collaborative risk coefficient between devices is calculated based on the device association features by using a collaborative filtering algorithm.

[0078] In step S305, the dynamic adaptive baseline is iteratively updated by using the collaborative risk coefficient, the potential extravasation risk sequence is recalculated, and a risk judgment result is generated.

[0079] Specifically, first, the potential extravasation risk sequence is subjected to signal processing, the abnormal fluctuation features in the time sequence are extracted, and an abnormal value sequence is generated. Second, the abnormal value sequence is compared with a preset threshold, and a high-risk point is screened out by a threshold discrimination model. Third, context environment data is obtained from the high-risk point association, and multi-medical device association features such as infusion pumps, pressure sensors, and temperature sensors are extracted by using a feature engineering method. Fourth, based on the device association features, a collaborative risk coefficient between devices is calculated by using a collaborative filtering algorithm. Finally, the dynamic adaptive baseline is iteratively updated by using the coefficient, the potential extravasation risk sequence is recalculated, and a risk judgment result is generated.

[0080] The embodiment improves the risk assessment precision by multi-technology fusion: signal processing and threshold discrimination realize accurate identification of abnormal fluctuations, avoiding single data point misjudgment; multi-device association feature extraction integrates multi-source data such as infusion pump pressure, sensor temperature and humidity, breaking through the limitation of single device monitoring; the collaborative filtering algorithm excavates the implicit association between devices, quantifies the influence of device collaboration on risk; the baseline iterative updating mechanism enables the model to dynamically adapt to patient individual differences and device running state changes. The sensitivity and reliability of extravasation risk identification are effectively improved.

[0081] In one embodiment, the inter-device collaborative risk coefficient can be calculated by the following steps:

[0082]

[0083] wherein C ij represents the collaborative risk coefficient between device D i and D j , S ij represents the eigenvector cosine similarity between device D i and D j , δ(t) represents the time decay factor, ω ij represents the association weight matrix element of device D i and D j , n represents the total number of devices, and ∈ represents the minimum value.

[0084] Preferably, δ(t) = e -λt , wherein λ represents the decay coefficient and t represents the time interval.

[0085] This embodiment realizes the quantitative evaluation of multi-dimensional device abnormality association by fusing feature similarity, time decay factor and device association weight. The eigenvector cosine similarity can capture the inherent association of device abnormality features, the time decay factor ensures the timeliness of abnormal association and avoids excessive interference of historical data on current risk assessment, and the device association weight matrix reflects the inherent business connection between devices in the clinical scene (such as the physical connection relationship between infusion pumps and pressure sensors). Through the normalization processing of the denominator, the collaborative risk coefficient has cross-device and cross-scene comparability. This method breaks through the limitations of traditional single device abnormality judgment and can find the collaborative effect of multi-medical device abnormality, such as the joint risk of infusion pump flow rate abnormality and puncture site pressure sensor fluctuation. By quantifying the inter-device collaborative risk, potential exosmosis risk sources can be more accurately identified, providing data support for dynamically adjusting adaptive baseline and optimizing risk assessment model, and thus improving the comprehensiveness and accuracy of intravenous infusion exosmosis early warning.

[0086] In one embodiment, the final exosmosis risk value is determined according to the risk judgment result and compared with the historical threshold, and a dynamic early warning signal is triggered according to the result, which can include the following steps:

[0087] Step S401, wavelet transform denoising is performed on the exosmosis risk original signal, data cleaning is performed through an adaptive filtering module, and a modified risk signal data is obtained.

[0088] Step S402, based on the risk judgment result, the statistical characteristics of patient historical data and current physiological parameters are determined to determine the final exosmosis risk value.

[0089] Preferably, based on the risk determination result, patient historical extravasation event data, previous physiological parameter fluctuation range and current real-time collected physiological parameters (such as blood pressure, infusion flow rate, puncture site pressure, etc.) are integrated, and statistical methods are used to calculate the mean, standard deviation, quantile and other characteristic indexes of the data. By constructing a risk weight model, combining the influence degree of each parameter on infusion extravasation and giving corresponding weights, the weighted operation of the quantized statistical characteristics and the weight coefficients is carried out, and finally the standardized final extravasation risk value is obtained, which provides a quantitative basis for subsequent risk early warning.

[0090] Step S403, the dynamic time warping algorithm is used to time-align the final extravasation risk value sequence and the historical threshold value sequence, calculate the Euclidean distance and the morphological similarity, and obtain the comparison result.

[0091] Step S404, the comparison result is quantitatively analyzed, and the characteristic parameters of the risk value deviating from the historical threshold value are extracted by using the sliding window algorithm.

[0092] Step S405, when the characteristic parameters trigger the preset conditions, the K-nearest neighbor algorithm is used to screen the matched early warning rule template based on the dynamic early warning rule library.

[0093] Step S406, the parameters in the early warning rule template are dynamically adjusted and instantiated to obtain the early warning rule, and the dynamic early warning signal of the corresponding level is generated.

[0094] Specifically, first, the wavelet transform denoising and adaptive filtering cleaning are performed on the extravasation risk original signal to obtain the corrected risk signal data. Secondly, based on the risk determination result, the statistical characteristics of the patient's historical data and the current physiological parameters are fused to determine the final extravasation risk value. Then, the dynamic time warping algorithm is used to time-align the final extravasation risk value sequence and the historical threshold value sequence, calculate the Euclidean distance and the morphological similarity, and obtain the comparison result. Then, the comparison result is quantitatively analyzed by using the sliding window algorithm, and the characteristic parameters of the risk value deviating from the historical threshold value are extracted. When the characteristic parameters trigger the preset conditions, the K-nearest neighbor algorithm is used to screen the matched early warning rule template based on the dynamic early warning rule library, and the template parameters are dynamically adjusted and instantiated, and finally the dynamic early warning signal of the corresponding level is generated.

[0095] The embodiment improves the accuracy and timeliness of risk early warning through multi-technology cooperation. Wavelet transform and adaptive filtering effectively remove noise from the original signal, ensuring data quality; combining historical and real-time data to determine risk values takes into account individual differences and dynamic changes; the dynamic time warping algorithm achieves accurate alignment and similarity measurement of sequences, avoiding time sequence misalignment interference; the combination of sliding window and K-nearest neighbor algorithm not only captures risk fluctuation characteristics, but also quickly matches early warning strategies based on historical rules; the parameter dynamic adjustment mechanism makes the early warning rule adapt to different scenarios.

[0096] In one of the embodiments, the method can further comprise the following steps:

[0097] Step S501, obtaining a set of physiological parameters of a patient; the set of physiological parameters comprises a plurality of individualized index data.

[0098] Step S502, extracting individualized feature information according to the set of physiological parameters, processing the individualized feature information by using a standardization algorithm, and obtaining a standardized feature dataset.

[0099] Step S503, performing feature correlation analysis on the standardized feature dataset, and determining a correlation weight matrix between features.

[0100] Step S504, screening a high-correlation feature combination from the correlation weight matrix, and obtaining a structured feature subset.

[0101] Step S505, if the dimension of the feature subset exceeds a preset threshold, performing dimension reduction processing by using a principal component analysis algorithm, and obtaining an optimized feature dataset.

[0102] Specifically, first, a set of physiological parameters of a patient is obtained through medical sensors, electronic medical record systems, and other channels, which covers a plurality of individualized index data such as infusion flow rate, puncture site pressure, skin temperature, heart rate, blood pressure, and the like. Second, feature information related to the risk of venous infusion exosmosis is extracted from the set of physiological parameters, such as patient age, medical history, infusion drug type, puncture site state, and the like, and normalization, standardization, and other algorithms are used to process the original feature data to eliminate dimensional differences, and a standardized feature dataset is obtained. Then, correlation analysis is performed on the standardized feature dataset, the Pearson correlation coefficient between each feature is calculated, and a correlation weight matrix between features is constructed. Based on the correlation weight matrix, a feature combination with high correlation and effective representation of exosmosis risk is screened, and a structured feature subset is formed. Finally, if the dimension of the structured feature subset exceeds a preset threshold (such as the number of features is greater than 15), a principal component analysis (PCA) algorithm is used for dimension reduction processing, the original high-dimensional features are mapped to a low-dimensional space through linear transformation, the data redundancy is reduced on the premise of retaining the main information, and finally an optimized feature dataset is obtained.

[0103] The embodiment improves data quality and feature effectiveness through multi-stage operations. Standardization eliminates the interference of dimension on data analysis, ensuring the comparability of different types of indicators. Feature correlation analysis and screening mechanism can eliminate redundant features, avoid model overfitting caused by excessive features, and focus on highly correlated features to enhance risk representation ability. Principal component analysis reduces data dimension, reduces computational complexity, and improves subsequent modeling efficiency while preserving key information. The overall process provides a clear structure, low redundancy, and strong representation data set for intravenous infusion extravasation risk assessment, enabling deep learning models or risk assessment algorithms based on the data set to operate more efficiently and accurately, improving the reliability and practicality of the early warning device.

[0104] In one embodiment, as shown in Figure 2 The application also provides a multi-parameter integrated intravenous infusion extravasation early warning device. The device can include:

[0105] The data processing and modeling module 601 is configured to obtain a set of physiological parameters of a patient, and to perform standardization processing in combination with individualized features to obtain a structured feature data set.

[0106] The risk judgment and evaluation module 602 is configured to use a deep learning time series model to perform dynamic modeling based on the feature data set, to generate an adaptive baseline and a potential extravasation risk sequence, and to perform signal processing on the potential extravasation risk sequence, to extract device-related features and calculate a coordination coefficient, and to update the adaptive baseline to obtain a risk judgment result.

[0107] The early warning rule generation module 603 is configured to determine a final extravasation risk value based on the risk judgment result and to compare the final extravasation risk value with a historical threshold value, and to trigger a dynamic early warning signal based on the comparison result.

[0108] The multi-parameter integrated venous infusion exosmosis early warning device is characterized in that the data processing modeling module is responsible for collecting physiological parameters of a patient, fusing individualized characteristics (such as age, medical history) for standardized processing, and constructing a structured feature data set. The risk judgment and evaluation module generates an adaptive baseline and a potential exosmosis risk sequence based on the feature data set by using a deep learning time sequence model; at the same time, the risk sequence is subjected to signal processing, multi-device correlation characteristics are extracted, a coordination coefficient is calculated, the adaptive baseline is iteratively updated, and a risk judgment result is output. The early warning rule generation module determines a final exosmosis risk value according to the risk judgment result, compares the exosmosis risk value with a historical threshold value, and triggers a dynamic early warning signal of a corresponding level. The device realizes intelligent monitoring of exosmosis risk. The data processing module unifies data formats and standards, and eliminates differences and noise interference of original data; the risk judgment and evaluation module combines deep learning and multi-device collaborative analysis, dynamically adapts to individual differences of a patient and influences of device interaction, and improves risk evaluation accuracy; and the early warning rule generation module compares a historical threshold value with a real-time risk value, and ensures timeliness and reliability of early warning triggering.

[0109] It should be understood that, although each step in the flowchart involved in each embodiment as described above is displayed in sequence according to the indication of the arrow, these steps are not necessarily executed in sequence according to the indication of the arrow. Unless explicitly stated herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other sequences. Moreover, at least part of the steps in the flowchart involved in each embodiment as described above can include multiple steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution sequence of these steps or stages is not necessarily sequential, but can be executed in rotation or alternation with at least part of other steps or steps or stages in other steps.

[0110] In one embodiment, a computer device is provided, comprising a memory and a processor, the memory stores a computer program, and the processor implements the steps of the multi-parameter integrated venous infusion exosmosis early warning method and device as described above when executing the computer program.

[0111] In one embodiment, a computer readable storage medium is provided, which stores a computer program, and the computer program is executed by a processor to implement the steps in each method embodiment described above.

[0112] For the device embodiment, since it basically corresponds to the method embodiment, the relevant part can be seen from the part of the method embodiment. The device embodiment described above is only schematic, wherein the components shown as separate components can or can not be physically separate, and the components shown as a unit can or can not be a physical unit, i.e., can be located in one place, or can be distributed to multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the present disclosure. Those skilled in the art can understand and implement it without creative labor.

[0113] The above-described embodiments only express several implementation manners of the present application, which are described in detail, but cannot be understood as a limitation on the patent scope of the application. It should be pointed out that, for those skilled in the art, without departing from the concept of the present application, several modifications and improvements can be made, which are all within the protection scope of the present application.

Claims

1. A multi-parameter integrated intravenous infusion extravasation early warning method, characterized in that: The method comprises: Obtain the patient's physiological parameter set, combine it with individual characteristics for standardization, and obtain a structured feature data set; Perform dynamic modeling using a deep learning time series model based on the feature data set to generate an adaptive baseline and a potential extravasation risk sequence; Performing signal processing on the potential extravasation risk sequence, extracting device association features and calculating a synergy coefficient, and updating the adaptive baseline to obtain a risk determination result; A final extravasation risk value is determined based on the risk determination result and compared with a historical threshold value, and a dynamic early warning signal is triggered based on the result.

2. The method according to claim 1, characterized in that The method of performing dynamic modeling based on the feature data set using a deep learning time series model to generate an adaptive baseline and a potential extravasation risk sequence includes: Extracting time series data from the feature data set and constructing a dynamic adaptive baseline using a deep learning algorithm; Calculating the deviation value at each time point based on the dynamic adaptive baseline to obtain an initial value of potential extravasation risk; Extracting correlation features between the initial extravasation risk value and historical extravasation event data, and determining a corresponding risk weighting coefficient through a machine learning algorithm; The initial risk value is weighted and adjusted using the risk weighting coefficient to generate a potential extravasation risk sequence.

3. The method according to claim 2, characterized in that The initial value of the potential extravasation risk is obtained by the following calculation formula: in, represents the initial value of potential extravasation risk, X t represents the eigenvector at time t, Represents the predicted value of the dynamic adaptive baseline at time t, Var(X t-k:t-1 ) represents the variance of the feature vector at the first k time points, λ represents the attenuation coefficient, α i represents the importance weight of the i-th historical extravasation event, DTW(X t-k:t-1 ,C i ) represents the dynamic time warping distance between the current sliding window and the i-th type of historical extravasation event pattern, and σ represents the Sigmoid function.

4. The method according to claim 2, characterized in that After generating the potential extravasation risk sequence, the method further includes: Extracting abnormal fluctuation features in the time series of the potential extravasation risk sequence using a signal processing algorithm to generate an outlier sequence; Compare the outlier sequence with a preset threshold, and screen out high-risk points using a threshold discrimination model; Obtain contextual environment data from the high-risk points, and use feature engineering methods to extract multi-medical device association features; the medical devices include but are not limited to infusion pumps, pressure sensors, and temperature sensors; Calculate the inter-device collaborative risk coefficient using a collaborative filtering algorithm based on the device association characteristics; The collaborative risk coefficient is used to iteratively update the dynamic adaptive baseline to generate a risk determination result.

5. The method according to claim 4, characterized in that The inter-device collaboration risk coefficient is calculated by the following steps: Among them, C ij Indicates device D i With D j The synergistic risk coefficient between ij Indicates device D i With D j The cosine similarity of the eigenvector, δ(t) represents the time decay factor, ω ij Indicates device D i With D j The associated weight matrix element of , n represents the total number of devices, and ∈ represents the minimum value.

6. The method according to claim 1, characterized in that Determining a final extravasation risk value based on the risk determination result and comparing it with a historical threshold value, and triggering a dynamic early warning signal based on the result, includes: Determining a final extravasation risk value based on the risk determination result in combination with the patient's historical data and statistical characteristics of current physiological parameters; A dynamic time warping algorithm is used to align the final extravasation risk value sequence with the historical threshold sequence, and the Euclidean distance and morphological similarity are calculated to obtain a comparison result; Quantitatively analyze the comparison results and use a sliding window algorithm to extract characteristic parameters of risk values ​​that deviate from historical thresholds; When the characteristic parameters trigger the preset conditions, the K-nearest neighbor algorithm is used to screen the matching warning rule templates based on the dynamic warning rule library; The parameters in the warning rule template are dynamically adjusted and instantiated to obtain warning rules, and generate dynamic warning signals of corresponding levels.

7. The method according to claim 1, characterized in that The method further comprises: Acquiring a set of physiological parameters of a patient; the set of physiological parameters includes multiple individualized indicator data; Extracting individualized feature information based on the physiological parameter set, and processing the individualized feature information using a standardized algorithm to obtain a standardized feature data set; Performing feature correlation analysis on the standardized feature data set to determine a correlation weight matrix between features; Screening the association weight matrix for highly correlated feature combinations to obtain a structured feature subset; If the dimension of the feature subset exceeds a preset threshold, a principal component analysis algorithm is used to reduce the dimension to obtain an optimized feature data set.

8. A multi-parameter integrated intravenous infusion extravasation warning device, characterized in that: The device comprises: The data processing and modeling module is used to obtain the patient's physiological parameter set, standardize it based on individual characteristics, and obtain a structured feature data set; The risk assessment module is used to dynamically model the feature data set using a deep learning time series model to generate an adaptive baseline and a potential extravasation risk sequence. It is also used to perform signal processing on the potential extravasation risk sequence, extract device-related features, calculate the synergy coefficient, and update the adaptive baseline to obtain the risk assessment result. The early warning rule generation module is used to determine the final extravasation risk value based on the risk judgment result and compare it with the historical threshold value, and trigger a dynamic early warning signal based on the result.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images