A method and system for monitoring drainage volume after surgery of a drainage tube

Abstract

The present invention provides a postoperative drainage volume monitoring method and system for a drainage tube, and relates to the technical field of postoperative drainage monitoring. The method includes: collecting drainage fluid time series monitoring data of multiple postoperative patients and performing adaptive wavelet packet decomposition to obtain the sub-band coefficient matrix corresponding to multiple decomposition layers for each group of drainage fluid time series monitoring data; determining multiple target demarcation points and drainage fluid monitoring stage division results for each postoperative patient, fusing the drainage fluid monitoring stage division results of multiple postoperative patients to generate a target stage division scheme; obtaining the vital sign monitoring data of each postoperative patient, extracting the drainage fluid stage monitoring features and vital sign stage monitoring features of each postoperative patient; constructing a postoperative drainage monitoring model and monitoring the postoperative drainage volume of the target patient based on the postoperative drainage monitoring model. The present invention realizes multi-stage refined monitoring and characterization analysis of the drainage volume of postoperative patients.

Description

Technical Field

[0001] The present invention relates to the technical field of postoperative drainage monitoring, and in particular to a postoperative drainage volume monitoring method and system for a drainage tube. Background Art

[0002] After surgery, in order to prevent complications such as infection caused by postoperative fluid or blood accumulation, a drainage tube is usually inserted into the surgical site to drain excess fluid from the body to help the patient recover. Monitoring the postoperative drainage volume in this process is not only related to the patient's recovery progress, but also directly affects the timing of drainage tube removal and the prevention of postoperative complications.

[0003] At present, some postoperative drainage volume monitoring methods mostly rely on manual recording and simple numerical judgment, and fail to fully explore the deep-level information in the drainage data. For example, postoperative drainage monitoring is often done by nurses or doctors regularly recording the drainage volume and observing and judging based on experience. This method is limited by the total drainage volume and simple time trend analysis when analyzing drainage data. It is difficult to accurately identify the characteristics of different recovery stages after surgery, and it will be affected by certain subjective factors. In addition, the change in drainage volume is usually not only determined by the effusion or bleeding at the surgical site, but also closely related to the patient's overall physiological state. For example, physical sign parameters such as heart rate, blood pressure and body temperature, the drainage volume characteristics at different postoperative recovery stages are often affected by these physical sign data. For example, high heart rate or fever may be related to abnormal increase in drainage volume, while decreased drainage volume and stable physical signs may be the manifestation of the recovery period. Without combining changes in physical sign data, the drainage volume data is analyzed in isolation to deeply judge the potential causes of drainage volume fluctuations, which may cause some subtle abnormal conditions to be not discovered in time, affecting the accuracy of drainage fluid monitoring during postoperative recovery. Summary of the invention

[0004] In order to solve the above technical problems, the present invention proposes a postoperative drainage volume monitoring method and system for drainage tubes, which realizes multi-stage refined monitoring and characteristic analysis of the drainage volume, and also realizes a more comprehensive assessment of the patient's postoperative status through the fusion of vital sign data. This systematic monitoring scheme optimizes the monitoring and management of drainage tubes, improves the level of data support in the patient's postoperative recovery process, and provides a more scientific and comprehensive basis for postoperative management.

[0005] To achieve the above object, as one aspect of the present invention, a method for monitoring the postoperative drainage volume of a drainage tube is provided, comprising:

[0006] The drainage fluid time series monitoring data of multiple postoperative patients are collected, and the drainage fluid time series monitoring data of each postoperative patient are subjected to adaptive wavelet packet decomposition to obtain the sub-band coefficient matrices corresponding to multiple decomposition layers for each set of drainage fluid time series monitoring data;

[0007] Energy feature extraction and feature screening are performed on multiple sub-band coefficient matrices to determine multiple key frequency band energy features, and multiple target cut-off points for each postoperative patient are determined based on the key frequency band energy features;

[0008] Determine the drainage fluid monitoring stage division result of each postoperative patient according to multiple target cut-off points of each postoperative patient, and fuse the drainage fluid monitoring stage division results of multiple postoperative patients to generate a target stage division scheme;

[0009] Obtain the vital sign monitoring data corresponding to the drainage fluid time-series monitoring data of each postoperative patient, extract features from the drainage fluid time-series monitoring data and vital sign monitoring data of each postoperative patient based on the target stage division scheme, and generate the drainage fluid stage monitoring features and vital sign stage monitoring features of each postoperative patient;

[0010] A postoperative drainage monitoring model was constructed according to the drainage fluid stage monitoring characteristics and physical sign stage monitoring characteristics of multiple patients, and the postoperative drainage volume of target patients was monitored based on the postoperative drainage monitoring model.

[0011] Preferably, energy feature extraction and feature screening are performed on multiple sub-band coefficient matrices to determine multiple key frequency band energy features, and multiple target cut-off points for each postoperative patient are determined based on the key frequency band energy features, including:

[0012] Extracting energy features of multiple sub-bands in each sub-band coefficient matrix respectively, obtaining energy features corresponding to each sub-band coefficient matrix, and constructing a first frequency band energy matrix containing multiple frequency band energy features for each postoperative patient;

[0013] Performing feature screening on each first-band energy matrix to generate a second-band energy matrix, wherein the second-band energy matrix includes a plurality of key-band energy features;

[0014] Performing sliding window analysis on multiple key frequency band energy features in each second frequency band energy matrix, constructing a characteristic trend curve of each frequency band energy feature and determining multiple first dividing points;

[0015] Extract the temporal position features and multiple frequency band features of each first demarcation point, construct a candidate feature vector for each first demarcation point, and generate a candidate feature matrix corresponding to each postoperative patient;

[0016] The multiple first demarcation points of each postoperative patient are clustered respectively through multiple candidate feature matrices to obtain multiple demarcation point clusters, and the multiple second demarcation points of each postoperative patient are determined as the multiple target demarcation points of each postoperative patient according to the multiple demarcation point clusters corresponding to each postoperative patient.

[0017] Preferably, clustering the multiple first cut-off points of each postoperative patient respectively through multiple candidate feature matrices to obtain multiple cut-off point clusters, and determining the multiple second cut-off points of each postoperative patient as the multiple target cut-off points of each postoperative patient according to the multiple cut-off point clusters corresponding to each postoperative patient, including:

[0018] The DBSCAN algorithm was used to process the candidate feature matrix of each postoperative patient, and multiple first cut-off points of each postoperative patient were clustered based on the candidate feature vector corresponding to each first cut-off point to generate multiple candidate clusters;

[0019] Sort the multiple first demarcation points in each candidate cluster according to the temporal position features of the first demarcation points, calculate the temporal position distance between each two adjacent first demarcation points in each candidate cluster, and determine the demarcation point combination in each candidate cluster whose temporal position distance is greater than a preset time threshold;

[0020] Based on the cutoff point combination, multiple candidate clusters are segmented to generate multiple cutoff point clusters. According to the energy feature corresponding to the first cutoff point, the multiple first cutoff points in each cutoff point cluster are weighted based on the time position of the energy feature, and the second cutoff point of each cutoff point cluster is determined. The multiple second cutoff points are summarized to obtain multiple target cutoff points for each postoperative patient.

[0021] Preferably, the drainage fluid monitoring stage division result of each postoperative patient is determined according to multiple target cut-off points of each postoperative patient, and the drainage fluid monitoring stage division results of multiple postoperative patients are fused to generate a target stage division scheme, including:

[0022] Determine multiple candidate drainage fluid monitoring stages for each postoperative patient according to multiple target cutoff points of each postoperative patient, extract the time domain features of each candidate drainage fluid monitoring stage based on the drainage fluid time series monitoring data of each postoperative patient, and extract the frequency domain features of each candidate drainage fluid monitoring stage based on the sub-frequency band coefficient matrices corresponding to multiple decomposition layers of the drainage fluid time series monitoring data of each postoperative patient;

[0023] The time domain features and frequency domain features of each candidate drainage fluid monitoring stage are concatenated to generate a monitoring stage feature vector for each candidate drainage fluid monitoring stage. Multiple monitoring stage feature vectors of each postoperative patient are summarized to construct a multi-source feature fusion matrix. The multi-source feature fusion matrix is ​​clustered to generate multiple monitoring stage clusters. The target stage division scheme is determined based on multiple stage clusters.

[0024] Preferably, adaptive wavelet packet decomposition is performed on the drainage fluid time series monitoring data of each postoperative patient to obtain sub-band coefficient matrices corresponding to multiple decomposition layers for each set of drainage fluid time series monitoring data, including:

[0025] A sample patient set was determined by random sampling of multiple postoperative patients. Multiple wavelet basis functions in the wavelet basis candidate set were used to perform wavelet packet decomposition based on the target decomposition layer number on each group of drainage fluid time series monitoring data in the sample patient set. The entropy value of each wavelet basis function in different sub-frequency bands was calculated, and the wavelet basis function with the smallest total entropy value was selected as the target decomposition function.

[0026] Based on the target decomposition function, each set of drainage fluid time series monitoring data in the sample patient set is subjected to multi-layer iterative wavelet packet decomposition. In the process of multi-layer iterative decomposition, the entropy value is calculated for each sub-frequency band of the decomposition layer. When the difference between the entropy value of the current layer and the entropy value of the previous layer is greater than the preset difference, the current layer is recorded as the target layer.

[0027] Based on the target layer and the target decomposition function, wavelet packet decomposition is performed on multiple groups of drainage fluid time series monitoring data except the sample patient set to obtain the sub-band coefficient matrices corresponding to multiple decomposition layers for each group of drainage fluid time series monitoring data.

[0028] Preferably, the process of determining a plurality of first demarcation points further includes:

[0029] Performing a multi-scale analysis on each first-band energy matrix, including extracting a plurality of statistical features of each first-band energy matrix in each frequency band, and constructing a multi-scale frequency band energy matrix corresponding to each first-band energy matrix;

[0030] Based on the recursive feature elimination algorithm, feature screening is performed on each multi-scale frequency band energy matrix to generate a second frequency band energy matrix corresponding to each first frequency band energy matrix;

[0031] A sliding window analysis is performed on multiple key frequency band energy features in each second frequency band energy matrix to generate a characteristic trend curve of each frequency band energy feature, and each characteristic trend curve is analyzed based on a significance test algorithm to determine multiple first dividing points corresponding to each characteristic trend curve.

[0032] As another aspect of the present invention, a postoperative drainage volume monitoring system for a drainage tube is provided, which is used to implement the above-mentioned postoperative drainage volume monitoring method for a drainage tube, comprising:

[0033] The sample data collection module is used to collect drainage fluid time-series monitoring data of multiple postoperative patients;

[0034] The frequency domain analysis module is used to perform adaptive wavelet packet decomposition on the drainage fluid time series monitoring data of each postoperative patient to obtain the sub-frequency band coefficient matrix corresponding to each set of drainage fluid time series monitoring data with respect to multiple decomposition layers;

[0035] A cut-off point generation module is used to extract and screen energy features of multiple sub-band coefficient matrices, determine multiple key frequency band energy features, and determine multiple target cut-off points for each postoperative patient based on the key frequency band energy features;

[0036] A stage division scheme generation module is used to determine the drainage fluid monitoring stage division result of each postoperative patient according to multiple target cut-off points of each postoperative patient, and to fuse the drainage fluid monitoring stage division results of multiple postoperative patients to generate a target stage division scheme;

[0037] A feature extraction module is used to obtain the vital sign monitoring data corresponding to the drainage fluid time-series monitoring data of each postoperative patient, perform feature extraction on the drainage fluid time-series monitoring data and vital sign monitoring data of each postoperative patient based on the target stage division scheme, and generate the drainage fluid stage monitoring features and vital sign stage monitoring features of each postoperative patient;

[0038] The drainage volume monitoring module is used to build a postoperative drainage monitoring model according to the drainage fluid stage monitoring characteristics and physical sign stage monitoring characteristics of multiple patients, and to monitor the postoperative drainage volume of target patients based on the postoperative drainage monitoring model.

[0039] Preferably, for the demarcation point generation module, energy feature extraction and feature screening are performed on multiple sub-band coefficient matrices to determine multiple key frequency band energy features, and multiple target demarcation points for each postoperative patient are determined according to the key frequency band energy features, including:

[0040] Extracting energy features of multiple sub-bands in each sub-band coefficient matrix respectively, obtaining energy features corresponding to each sub-band coefficient matrix, and constructing a first frequency band energy matrix containing multiple frequency band energy features for each postoperative patient;

[0041] Performing feature screening on each first-band energy matrix to generate a second-band energy matrix, wherein the second-band energy matrix includes a plurality of key-band energy features;

[0042] Performing sliding window analysis on multiple key frequency band energy features in each second frequency band energy matrix, constructing a characteristic trend curve of each frequency band energy feature and determining multiple first dividing points;

[0043] Extract the temporal position features and multiple frequency band features of each first demarcation point, construct a candidate feature vector for each first demarcation point, and generate a candidate feature matrix corresponding to each postoperative patient;

[0044] The multiple first demarcation points of each postoperative patient are clustered respectively through multiple candidate feature matrices to obtain multiple demarcation point clusters, and the multiple second demarcation points of each postoperative patient are determined as the multiple target demarcation points of each postoperative patient according to the multiple demarcation point clusters corresponding to each postoperative patient.

[0045] The present invention has the following beneficial effects:

[0046] The postoperative drainage volume monitoring method for drainage tubes provided by the present invention relies on the real data of multiple patients for stage division based on data analysis, automatically identifies the change pattern in the recovery process through statistical analysis, and avoids the bias of subjective experience. This method can reflect the actual recovery process of different patients in a more objective way and reduce the inconsistency caused by human factors; through adaptive wavelet decomposition and feature extraction, it can capture the complex characteristics of drainage data and vital sign data, so as to identify more subtle stage changes in the patient's recovery process. Different from the extensiveness and ambiguity of empirical division, data-driven stage division can more accurately identify specific stages in the recovery process, making monitoring more detailed. Through large-scale data analysis, common stage divisions applicable to most patients can be extracted, while retaining individual differences between different patients, realizing data-driven postoperative recovery monitoring, realizing multi-stage refined monitoring and characterization analysis of drainage volume, and more comprehensive evaluation of the patient's postoperative state. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 A schematic flow chart of a method for monitoring postoperative drainage volume of a drainage tube provided in an embodiment of the present invention.

[0048] Figure 2 A schematic structural diagram of a postoperative drainage volume monitoring system for a drainage tube provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0049] In order to make those skilled in the art better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0050] See also Figure 1 The embodiment of the present invention provides a method for monitoring the postoperative drainage volume of a drainage tube, which specifically includes the following steps:

[0051] Step S1: collecting drainage fluid time series monitoring data of multiple postoperative patients, performing adaptive wavelet packet decomposition on the drainage fluid time series monitoring data of each postoperative patient, and obtaining sub-band coefficient matrices corresponding to multiple decomposition layers for each set of drainage fluid time series monitoring data.

[0052] In this embodiment, after some surgical operations such as thoracic surgery or lung surgery, it is necessary to drain excess fluid from the patient's body, such as secretions such as blood and pus, to prevent infection, reduce fluid accumulation and promote recovery. Depending on the type and location of the operation, the type of fluid drained also varies. The collected drainage fluid time series monitoring data includes but is not limited to monitoring data such as changes in drainage volume over time, flow rate, and liquid composition. During a specific monitoring period after surgery, data is generally collected from the patient regularly to obtain the patient's drainage fluid time series monitoring data to record the changes in each patient's drainage fluid at different time points. It reflects the dynamic process of the patient's postoperative recovery.

[0053] For each patient's drainage fluid time-series monitoring data, the adaptive wavelet packet decomposition method is used to decompose the data into multiple decomposition layers to extract the drainage fluid change characteristics in different frequency ranges. Specifically, the appropriate wavelet basis and decomposition layer number are selected according to the actual situation of the data, and the data is decomposed at multiple scales to obtain multiple sub-band coefficient matrices corresponding to different decomposition layers. Each sub-band coefficient matrix contains drainage characteristic information of different frequency components, which helps to identify different change patterns of postoperative drainage volume data.

[0054] Step S2: extracting and screening energy features of multiple sub-band coefficient matrices, determining multiple key frequency band energy features, and determining multiple target demarcation points for each postoperative patient based on the key frequency band energy features.

[0055] In this embodiment, the sub-band coefficient matrix obtained by wavelet packet decomposition is used to extract energy features of each sub-band to reveal the energy distribution of drainage fluid in different frequency bands. The energy feature can reflect the fluctuation intensity and frequency component of the data, which is of great significance for identifying the change characteristics of different postoperative stages. The energy features of the sub-bands of each decomposition layer are calculated separately to form a set of energy features of multiple frequency bands.

[0056] After obtaining the energy characteristics of all frequency bands, feature screening is performed to determine the energy characteristics of multiple key frequency bands and retain the characteristic dimensions that are representative and important. Through these key frequency band energy characteristics, significant change points in the drainage volume data can be identified, and multiple target cutoff points in the patient's postoperative recovery process can be further determined. Each target cutoff point marks the critical change moment in the drainage fluid time series data in time, corresponding to different recovery stages.

[0057] Step S3: Determine the drainage fluid monitoring stage division result of each postoperative patient according to multiple target demarcation points of each postoperative patient, and fuse the drainage fluid monitoring stage division results of multiple postoperative patients to generate a target stage division scheme.

[0058] In this embodiment, the drainage fluid time series data of each postoperative patient is divided into several recovery stages according to the multiple target demarcation points determined in step S2. Each stage represents a different state of the patient's postoperative recovery process, such as the acute stage, the recovery stage, the stable stage, etc. The stage division result can provide an intuitive judgment of the recovery progress of each patient and help medical staff better understand the patient's recovery status.

[0059] In order to extract the common laws of multiple patients, the stage division results of different patients are fused to generate a universal target stage division scheme. The fusion process analyzes the common characteristics of different patients at each stage to form a universally applicable recovery stage division scheme, which is used to guide the recovery stage judgment of subsequent new patients and realize the standardized division of the postoperative recovery process.

[0060] Step S4: Obtain the vital sign monitoring data corresponding to the drainage fluid time-series monitoring data of each postoperative patient, perform feature extraction on the drainage fluid time-series monitoring data and vital sign monitoring data of each postoperative patient based on the target stage division scheme, and generate the drainage fluid stage monitoring features and vital sign stage monitoring features of each postoperative patient.

[0061] In this embodiment, after obtaining a general target stage division scheme, feature extraction is performed on the drainage fluid time series monitoring data and vital sign monitoring data of each patient. Vital sign monitoring data usually include heart rate, blood pressure, body temperature, etc., which can provide information on the patient's overall physiological state and help determine the potential causes of changes in drainage volume. By extracting drainage fluid features and physical sign features in each recovery stage, drainage fluid stage monitoring features and physical sign stage monitoring features are generated for each patient. The drainage fluid stage monitoring features include information such as the drainage volume status and change trend within the stage, while the physical sign stage monitoring features include the physical sign mean and fluctuation amplitude within the stage. These stage monitoring features constitute a comprehensive feature description of each patient at different recovery stages.

[0062] Step S5: construct a postoperative drainage monitoring model according to the drainage fluid stage monitoring characteristics and the physical sign stage monitoring characteristics of multiple patients, and monitor the postoperative drainage volume of the target patients based on the postoperative drainage monitoring model.

[0063] In this embodiment, a postoperative drainage monitoring model based on rule analysis is established based on the drainage fluid stage monitoring characteristics and physical sign stage monitoring characteristics of multiple postoperative patients for reference. Specifically, statistical information of key features such as mean, change rate and standard deviation are extracted for each recovery stage, and reasonable feature ranges and abnormal thresholds are set for each stage. When monitoring the postoperative drainage volume of the target patient, the drainage fluid time series data and physical sign data collected in real time can be compared with the rule ranges of each stage. According to the patient's current recovery stage, the difference between the data and the historical statistical characteristics is judged. If it deviates from the preset range, it indicates that there may be an abnormality, assisting medical staff to identify abnormal situations in time and make corresponding adjustments. For example, in the acute postoperative phase of a certain operation, the rate of change of drainage volume should be kept between -0.2 and -0.5. If it exceeds this range, it is marked as abnormal.

[0064] It is worth noting that, compared with the traditional method of dividing the recovery stage based on clinical experience, the postoperative drainage volume monitoring method for drainage tubes provided by the present invention relies on the real data of multiple patients for stage division based on data analysis, and automatically identifies the change pattern in the recovery process through statistical analysis, thus avoiding the bias of subjective experience. This method can reflect the actual recovery process of different patients in a more objective way and reduce the inconsistency caused by human factors; through adaptive wavelet decomposition and feature extraction, it can capture the complex characteristics of drainage data and vital sign data, thereby identifying more subtle stage changes in the patient's recovery process. Unlike the extensiveness and ambiguity of empirical division, data-driven stage division can more accurately identify specific stages in the recovery process, making monitoring more detailed. Through large-scale data analysis, common stage divisions applicable to most patients can be extracted while retaining individual differences between different patients.

[0065] For example, there may be slight differences in the characteristics of the acute phase, recovery phase, and other phases of different patients. Data analysis can automatically adapt to these differences and generate more universal stage division standards. With the introduction of new patient data, the stage division standards can also be iteratively updated to form a dynamic improvement process, thereby realizing data-driven postoperative recovery monitoring, making the program more scientific, accurate, and reliable. Stage division based on data analysis can generate specific feature statistical ranges and thresholds for each stage, which is suitable for the model to make refined abnormal judgments at each stage. For example, if the patient data deviates significantly from the historical statistical feature range during the recovery period, an early warning can be issued more accurately, thereby improving the accuracy and sensitivity of abnormal monitoring.

[0066] Furthermore, in step S2, energy feature extraction and feature screening are performed on multiple sub-band coefficient matrices to determine multiple key frequency band energy features, and multiple target cut-off points for each postoperative patient are determined based on the key frequency band energy features. The implementation process includes:

[0067] Step S21: extracting energy features of multiple sub-bands in each sub-band coefficient matrix respectively, obtaining energy features corresponding to each sub-band coefficient matrix, and constructing a first frequency band energy matrix containing multiple frequency band energy features for each postoperative patient.

[0068] In this embodiment, energy feature extraction is performed for each of the multiple sub-bands in the sub-band coefficient matrix. Specifically, by calculating the sum of the squares of the coefficients of each sub-band, the energy value of the band can be obtained, which represents the fluctuation intensity of the drainage fluid in the band. The energy distribution of different frequency bands can reveal the characteristics of the drainage fluid in different frequency components. The energy features of each sub-band are combined to construct a first frequency band energy matrix, which contains the energy features of multiple frequency bands. The first frequency band energy matrix of each postoperative patient reflects the energy feature distribution of the patient's drainage data in each frequency band.

[0069] Step S22: Perform feature screening on each first-band energy matrix to generate a second-band energy matrix, perform sliding window analysis on multiple key-band energy features in each second-band energy matrix, construct a feature trend curve for each band energy feature and determine multiple first dividing points.

[0070] In this embodiment, in order to reduce redundant information and improve analysis efficiency, the multidimensional energy features in the first frequency band energy matrix are screened, and the frequency band features with high importance in distinguishing the features of different stages are retained to form a simplified feature set. After feature screening, the retained key frequency band energy features are constructed into a second frequency band energy matrix. The second frequency band energy matrix contains multiple key frequency band energy features, which are more representative frequency band features in the drainage fluid time series data and can effectively characterize the characteristic patterns of different recovery stages.

[0071] It is worth noting that in the process of performing feature screening on each first-band energy matrix to generate a second-band energy matrix, each first-band energy matrix can be first subjected to multi-scale analysis to construct a multi-scale band energy matrix corresponding to each first-band energy matrix. Exemplarily, for each frequency band of each first-band energy matrix, statistical features such as mean, variance, skewness, etc. are calculated to form a multi-scale band energy matrix corresponding to each first-band energy matrix. Then, a recursive feature elimination algorithm is used to perform feature screening on each multi-scale band energy matrix to retain the most representative features, and finally, the second-band energy matrix corresponding to each first-band energy matrix can be obtained through recursive feature analysis.

[0072] Then, a sliding window analysis is performed on each key frequency band energy feature in the second frequency band energy matrix. In the sliding window analysis, by presetting the size of the sliding window, the mean or change rate of each frequency band energy feature in the window is calculated to construct a characteristic trend curve of each frequency band energy feature. In the characteristic trend curve, the location points of significant changes are identified, and they can be identified based on the significance test algorithm of the change rate, such as the Z test algorithm, to identify multiple key location points, and then mark these location points as the first dividing point. The first dividing point usually represents the location of significant changes in the drainage fluid characteristics within this stage, and may correspond to the characteristic turning points of different stages such as the acute phase, recovery phase and stable phase. Each first dividing point is a candidate stage boundary, which provides a preliminary basis for the subsequent screening of dividing points.

[0073] Step S23: extract the temporal position features and multiple frequency band features of each first demarcation point, construct a candidate feature vector for each first demarcation point, and generate a candidate feature matrix corresponding to each postoperative patient.

[0074] In this embodiment, for each first demarcation point, its temporal position feature in the time series is extracted, that is, in terms of time distribution, the specific time position in the monitoring period, such as the number of hours and days in the monitoring period, and the corresponding key frequency band energy feature is recorded at the same time, which can be the performance of multiple frequency band features at this moment in the time window near the demarcation point, such as the energy value of the frequency band feature at the demarcation point, the frequency band change rate in the time window, etc. The temporal position feature reflects the temporal position of the demarcation point in the recovery process, while the frequency band feature provides the multi-frequency component characteristics of the drainage fluid at this position.

[0075] The temporal position feature of each first demarcation point is combined with the key frequency band feature to generate a candidate feature vector for the demarcation point. The candidate feature vector set generated for all first demarcation points of each patient constitutes the candidate feature matrix of the patient, where the candidate feature matrix contains the key information of all candidate demarcation points.

[0076] Step S24: Clustering the multiple first demarcation points of each postoperative patient through multiple candidate feature matrices to obtain multiple demarcation point clusters, and determining multiple second demarcation points for each postoperative patient according to the multiple demarcation point clusters corresponding to each postoperative patient to serve as multiple target demarcation points for each postoperative patient.

[0077] In this embodiment, the candidate feature matrix of each patient is used as input, and cluster analysis is performed on all candidate demarcation points of the patient. The clustering process is based on the temporal position characteristics and key frequency band characteristics of the demarcation points, and the demarcation points with similar characteristics and adjacent time are classified into the same cluster. The purpose of clustering is to merge demarcation points with similar characteristics to avoid repeated demarcations caused by data fluctuations. For multiple demarcation point clusters formed by clustering, each cluster represents a set of demarcation points with similar characteristics. For each cluster, a representative demarcation point is determined as the center point of the cluster to obtain multiple streamlined demarcation point sets. The representative core point of each demarcation point cluster is used as the second demarcation point, that is, the final target demarcation point. These target demarcation points are used to divide the drainage fluid monitoring data into stages, identify the feature change positions at different stages of the patient's postoperative recovery process, and provide accurate stage boundaries for subsequent recovery process monitoring.

[0078] Furthermore, in step S24, multiple first cut-off points of each postoperative patient are clustered by multiple candidate feature matrices to obtain multiple cut-off point clusters, and multiple second cut-off points of each postoperative patient are determined according to the multiple cut-off point clusters corresponding to each postoperative patient to serve as multiple target cut-off points for each postoperative patient. The implementation process includes:

[0079] The DBSCAN algorithm was used to process the candidate feature matrix of each postoperative patient. The DBSCAN algorithm can identify the demarcation points that are adjacent and have similar features in the temporal position by setting the distance threshold and the minimum number of points. It can identify clusters of different densities and automatically filter out isolated noise points, making the clustering results more accurate. Finally, multiple first demarcation points of each postoperative patient were clustered based on the candidate feature vector corresponding to each first demarcation point to generate multiple candidate clusters.

[0080] According to the temporal position characteristics of the first dividing points, multiple first dividing points in each candidate cluster are sorted. Through temporal sorting, the temporal order of the dividing points can be clarified, which is convenient for subsequent judgment of whether the dividing points belong to the same stage. Then, the temporal position distance between each two adjacent first dividing points in each candidate cluster is calculated. The temporal position distance can reflect the proximity of the dividing points in the time dimension. The smaller the distance between adjacent dividing points, the more likely it is that they belong to the same stage. If the distance is large, the cluster may need to be further segmented. By presetting the time threshold, it is determined whether the dividing points belong to the same stage. When the temporal position distance between two adjacent dividing points is greater than the threshold, the pair of dividing points is marked as a candidate segmentation point. In this way, multiple dividing point combinations whose temporal position distance is greater than the preset time threshold can be determined.

[0081] Then, based on the combination of demarcation points, multiple candidate clusters are segmented to generate multiple demarcation point clusters. Specifically, the candidate cluster where the demarcation point whose temporal position distance is greater than the threshold is located is further split into multiple sub-cluster to generate the final demarcation point cluster. Exemplarily, if a candidate cluster includes seven consecutive first demarcation points, and the temporal position distance between the third and fourth first demarcation points is greater than the preset time threshold, it is segmented into two sub-clusters, that is, the first three and the last four first demarcation points constitute a new cluster respectively. This segmentation process ensures that the demarcation points within each demarcation point cluster are coherent in the time dimension, thereby improving the stage representativeness of the demarcation points.

[0082] For the multiple cut-off point clusters finally determined, the multiple first cut-off points in each cut-off point cluster are weighted based on the time series position of the energy characteristics according to the energy characteristics corresponding to the first cut-off point. The cut-off points with higher energy values ​​have a greater weight in the weighted average, thereby obtaining a comprehensive cut-off point position representing the cluster. This time series position calculated by weighted average is used as the second cut-off point of the cluster. The second cut-off point represents the comprehensive time series characteristics and energy characteristics of multiple first cut-off points in the cluster, reflecting the key change positions in the patient's recovery process at this stage. After summarizing multiple second cut-off points, multiple target cut-off points are obtained for each postoperative patient. These target cut-off points mark the key change moments of the patient's drainage fluid monitoring data and can be used to accurately divide different recovery stages.

[0083] Furthermore, in step S3, the drainage fluid monitoring stage division result of each postoperative patient is determined according to multiple target cut-off points of each postoperative patient, and the drainage fluid monitoring stage division results of multiple postoperative patients are fused to generate a target stage division scheme, which specifically includes:

[0084] Multiple candidate drainage fluid monitoring stages for each postoperative patient are determined based on multiple target cutoff points for each postoperative patient. Specifically, using multiple target cutoff points for each postoperative patient, the drainage fluid time series data is segmented into multiple candidate drainage fluid monitoring stages, each candidate monitoring stage representing a specific period of time during the patient's postoperative recovery process, which may correspond to different physiological state changes.

[0085] Based on the drainage fluid time series monitoring data of each postoperative patient, the time domain features of each candidate drainage fluid monitoring stage are extracted. Among them, the time domain features can describe the change trend, volatility and stability of the drainage fluid in this stage, such as the mean, standard deviation, change rate and other characteristic parameters of the drainage volume. The time domain features of each candidate stage are combined to form a time domain feature vector, which represents the time domain performance of the candidate stage in terms of drainage volume.

[0086] Based on the sub-band coefficient matrices corresponding to multiple decomposition layers of each postoperative patient's drainage fluid time series monitoring data, the frequency domain features of each candidate drainage fluid monitoring stage are extracted. The frequency domain features can reveal the variation characteristics of the drainage volume at different frequencies, such as sub-band energy, frequency band ratio, frequency distribution characteristics, etc. The frequency domain features are combined to generate frequency domain feature vectors, which complement the time domain features to fully describe the feature information of each candidate stage.

[0087] The time domain features and frequency domain features of each candidate drainage fluid monitoring stage are concatenated to generate the monitoring stage feature vector of each candidate drainage fluid monitoring stage, which integrates the multi-dimensional information of this stage in time and frequency and can accurately characterize the changing characteristics of the drainage volume. The feature vectors of multiple monitoring stages of each postoperative patient are summarized to construct a multi-source feature fusion matrix, forming a comprehensive feature set to reflect the multi-level features of the drainage fluid data in this stage. Finally, the multi-source feature fusion matrix is ​​clustered to generate multiple monitoring stage clusters, and the target stage division scheme is determined based on multiple stage clusters.

[0088] It is worth noting that the number and timing information of candidate drainage fluid monitoring stages vary between different postoperative patients. For example, some patients may have 3 stages, while others may have 4 or 5 stages. If the timing is forcibly considered, it will lead to unnecessary complexity and may obscure the commonalities between patients. Therefore, combining the stage feature vectors of different patients and clustering them as independent data points can focus more on the similarities in features. This hodgepodge clustering method can help us find the commonalities in features of each stage. For example, if all patients have relatively high energy values, change rates and other features in the acute phase, the acute phase will automatically form a cluster in the clustering. The clustering results will naturally identify the common rules between patients, that is, no matter how long the patient's recovery process is or how many stages there are, they will present some specific feature combinations at different stages of recovery.

[0089] The monitoring stage clusters in the clustering results can be regarded as universal recovery stage features. Each cluster represents a stage with similar characteristic patterns, such as the acute stage, recovery stage, stable stage, etc. Even if the specific stage sequence and duration of different patients are different, this common feature can still reveal the general law. In this way, we can extract the typical characteristics of different patients at the same stage, and generate a universal recovery stage division scheme based on this. It was finally determined that the target stage division scheme has strong applicability, making the monitoring of drainage fluid in the recovery process of new patients more reliable.

[0090] See also Figure 2The embodiment of the present invention provides a postoperative drainage volume monitoring system for a drainage tube, which is used to implement the above-mentioned postoperative drainage volume monitoring method for a drainage tube, and specifically includes the following structure:

[0091] The sample data collection module is used to collect drainage fluid time-series monitoring data of multiple postoperative patients;

[0092] The frequency domain analysis module is used to perform adaptive wavelet packet decomposition on the drainage fluid time series monitoring data of each postoperative patient to obtain the sub-frequency band coefficient matrix corresponding to each set of drainage fluid time series monitoring data with respect to multiple decomposition layers;

[0093] Specifically, a sample patient set is determined by randomly sampling multiple postoperative patients, and a plurality of wavelet basis functions in the wavelet basis candidate set are used to perform wavelet packet decomposition based on the target decomposition layer number on each group of drainage fluid time series monitoring data in the sample patient set, and the entropy value of each wavelet basis function in different sub-frequency bands is calculated, and the wavelet basis function with the smallest total entropy value is selected as the target decomposition function;

[0094] Based on the target decomposition function, each set of drainage fluid time series monitoring data in the sample patient set is subjected to multi-layer iterative wavelet packet decomposition. In the process of multi-layer iterative decomposition, the entropy value is calculated for each sub-frequency band of the decomposition layer. When the difference between the entropy value of the current layer and the entropy value of the previous layer is greater than the preset difference, the current layer is recorded as the target layer.

[0095] Based on the target layer and the target decomposition function, wavelet packet decomposition is performed on multiple groups of drainage fluid time series monitoring data except the sample patient set to obtain the sub-band coefficient matrices corresponding to multiple decomposition layers for each group of drainage fluid time series monitoring data.

[0096] A cut-off point generation module is used to extract and screen energy features of multiple sub-band coefficient matrices, determine multiple key frequency band energy features, and determine multiple target cut-off points for each postoperative patient based on the key frequency band energy features;

[0097] Specifically, energy features are extracted for multiple sub-bands in each sub-band coefficient matrix to obtain energy features corresponding to each sub-band coefficient matrix, and a first frequency band energy matrix containing multiple frequency band energy features is constructed for each postoperative patient;

[0098] Performing feature screening on each first-band energy matrix to generate a second-band energy matrix, wherein the second-band energy matrix includes a plurality of key-band energy features;

[0099] Performing sliding window analysis on multiple key frequency band energy features in each second frequency band energy matrix, constructing a characteristic trend curve of each frequency band energy feature and determining multiple first dividing points;

[0100] Extract the temporal position features and multiple frequency band features of each first demarcation point, construct a candidate feature vector for each first demarcation point, and generate a candidate feature matrix corresponding to each postoperative patient;

[0101] The multiple first demarcation points of each postoperative patient are clustered respectively through multiple candidate feature matrices to obtain multiple demarcation point clusters, and the multiple second demarcation points of each postoperative patient are determined as the multiple target demarcation points of each postoperative patient according to the multiple demarcation point clusters corresponding to each postoperative patient.

[0102] A stage division scheme generation module is used to determine the drainage fluid monitoring stage division result of each postoperative patient according to multiple target cut-off points of each postoperative patient, and to fuse the drainage fluid monitoring stage division results of multiple postoperative patients to generate a target stage division scheme;

[0103] A feature extraction module is used to obtain the vital sign monitoring data corresponding to the drainage fluid time-series monitoring data of each postoperative patient, perform feature extraction on the drainage fluid time-series monitoring data and vital sign monitoring data of each postoperative patient based on the target stage division scheme, and generate the drainage fluid stage monitoring features and vital sign stage monitoring features of each postoperative patient;

[0104] The drainage volume monitoring module is used to build a postoperative drainage monitoring model according to the drainage fluid stage monitoring characteristics and physical sign stage monitoring characteristics of multiple patients, and to monitor the postoperative drainage volume of target patients based on the postoperative drainage monitoring model.

[0105] The above is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. Parts not described in detail in this specification belong to the prior art known to those skilled in the art.

Claims

1. A method for monitoring drainage volume after surgery of a drainage tube, characterized in that: include: The drainage fluid time series monitoring data of multiple postoperative patients were collected, and the adaptive wavelet packet decomposition was performed on the drainage fluid time series monitoring data of each postoperative patient to obtain the sub-band coefficient matrix corresponding to multiple decomposition layers for each set of drainage fluid time series monitoring data; Energy feature extraction and feature screening are performed on multiple sub-band coefficient matrices to determine multiple key frequency band energy features. Based on the key frequency band energy features, multiple target cut-off points are determined for each postoperative patient. Determine the drainage fluid monitoring stage division results for each postoperative patient based on multiple target cutoff points for each postoperative patient, and fuse the drainage fluid monitoring stage division results for multiple postoperative patients to generate a target stage division scheme; Obtain the vital sign monitoring data corresponding to the drainage fluid time-series monitoring data of each postoperative patient, perform feature extraction on the drainage fluid time-series monitoring data and vital sign monitoring data of each postoperative patient based on the target stage division scheme, and generate the drainage fluid stage monitoring features and vital sign stage monitoring features of each postoperative patient; A postoperative drainage monitoring model was constructed based on the drainage fluid and physical sign monitoring characteristics of multiple patients, and the postoperative drainage volume of target patients was monitored based on the postoperative drainage monitoring model. Energy feature extraction and feature screening are performed on multiple sub-band coefficient matrices to determine multiple key frequency band energy features. Based on the key frequency band energy features, multiple target cutoff points are determined for each postoperative patient, including: Energy features are extracted for each of the multiple sub-bands in each sub-band coefficient matrix to obtain energy features corresponding to each sub-band coefficient matrix, and a first frequency band energy matrix containing multiple frequency band energy features is constructed for each postoperative patient; Performing feature screening on each first-band energy matrix to generate a second-band energy matrix, where the second-band energy matrix includes a plurality of key-band energy features; Performing a sliding window analysis on multiple key frequency band energy features in each second frequency band energy matrix, constructing a characteristic trend curve of each frequency band energy feature and determining multiple first dividing points; Extract the temporal position features and multiple frequency band features of each first demarcation point, construct a candidate feature vector for each first demarcation point, and generate a candidate feature matrix corresponding to each postoperative patient; The multiple first cut-off points of each postoperative patient are clustered using multiple candidate feature matrices to obtain multiple cut-off point clusters, and the multiple second cut-off points of each postoperative patient are determined as the multiple target cut-off points of each postoperative patient based on the multiple cut-off point clusters corresponding to each postoperative patient.

2. A method for monitoring drainage volume after surgery for a drainage tube according to claim 1, characterized in that: Clustering the multiple first cutoff points of each postoperative patient using multiple candidate feature matrices to obtain multiple cutoff point clusters, and determining the multiple second cutoff points of each postoperative patient as the multiple target cutoff points of each postoperative patient based on the multiple cutoff point clusters corresponding to each postoperative patient, including: The DBSCAN algorithm was used to process the candidate feature matrix of each postoperative patient, and multiple first cutoff points of each postoperative patient were clustered based on the candidate feature vector corresponding to each first cutoff point to generate multiple candidate clusters. Sort the multiple first dividing points in each candidate cluster according to the temporal position characteristics of the first dividing points, calculate the temporal position distance between each two adjacent first dividing points in each candidate cluster, and determine the combination of dividing points in each candidate cluster whose temporal position distance is greater than a preset time threshold; Based on the combination of cutoff points, multiple candidate clusters are segmented to generate multiple cutoff point clusters. According to the energy characteristics corresponding to the first cutoff point, the multiple first cutoff points in each cutoff point cluster are weighted based on the time position of the energy characteristics, and the second cutoff point of each cutoff point cluster is determined. The multiple second cutoff points are summarized to obtain multiple target cutoff points for each postoperative patient.

3. The method for monitoring drainage volume after surgery of a drainage tube according to claim 1, wherein: The drainage fluid monitoring stage division results of each postoperative patient are determined based on multiple target cutoff points of each postoperative patient. The drainage fluid monitoring stage division results of multiple postoperative patients are fused to generate a target stage division scheme, including: Determining multiple candidate drainage fluid monitoring stages for each postoperative patient based on multiple target cutoff points of each postoperative patient, extracting time domain features of each candidate drainage fluid monitoring stage based on the drainage fluid time series monitoring data of each postoperative patient, and extracting frequency domain features of each candidate drainage fluid monitoring stage based on the sub-frequency band coefficient matrices corresponding to multiple decomposition layers of the drainage fluid time series monitoring data of each postoperative patient; The time domain features and frequency domain features of each candidate drainage fluid monitoring stage are concatenated to generate a monitoring stage feature vector for each candidate drainage fluid monitoring stage. Multiple monitoring stage feature vectors of each postoperative patient are summarized to construct a multi-source feature fusion matrix. The multi-source feature fusion matrix is ​​clustered to generate multiple monitoring stage clusters. The target stage division scheme is determined based on the multiple stage clusters.

4. The method for monitoring drainage volume after surgery of a drainage tube according to claim 1, wherein: Adaptive wavelet packet decomposition is performed on the drainage fluid time series monitoring data of each postoperative patient to obtain the sub-band coefficient matrices corresponding to multiple decomposition layers of each set of drainage fluid time series monitoring data, including: A sample patient set was determined by random sampling of multiple postoperative patients. Multiple wavelet basis functions from the wavelet basis candidate set were used to perform wavelet packet decomposition based on the target decomposition level for each set of drainage fluid time series monitoring data in the sample patient set. The entropy value of each wavelet basis function in different sub-frequency bands was calculated, and the wavelet basis function with the smallest total entropy value was selected as the target decomposition function. Based on the target decomposition function, each set of drainage fluid time series monitoring data in the sample patient set is subjected to multi-layer iterative wavelet packet decomposition. During the multi-layer iterative decomposition process, the entropy value is calculated for each sub-frequency band of the decomposition layer. When the difference between the entropy value of the current layer and the entropy value of the previous layer is greater than the preset difference, the current layer is recorded as the target layer; Based on the target layer and target decomposition function, wavelet packet decomposition is performed on multiple groups of drainage fluid time series monitoring data except the sample patient set to obtain the sub-band coefficient matrices corresponding to multiple decomposition layers for each group of drainage fluid time series monitoring data.

5. The method for monitoring drainage volume after surgery of a drainage tube according to claim 1, wherein: The process of determining multiple first demarcation points also includes: Performing a multi-scale analysis on each first-band energy matrix, including extracting a plurality of statistical features of each first-band energy matrix in each frequency band, and constructing a multi-scale frequency band energy matrix corresponding to each first-band energy matrix; Based on the recursive feature elimination algorithm, each multi-scale frequency band energy matrix is ​​subjected to feature screening to generate a second frequency band energy matrix corresponding to each first frequency band energy matrix; A sliding window analysis is performed on multiple key frequency band energy features in each second frequency band energy matrix to generate a characteristic trend curve of each frequency band energy feature. Each characteristic trend curve is analyzed based on a significance test algorithm to determine multiple first dividing points corresponding to each characteristic trend curve.

6. A postoperative drainage volume monitoring system for a drainage tube, characterized in that: The system is used to implement the postoperative drainage volume monitoring method for a drainage tube according to any one of claims 1 to 5, comprising: Sample data acquisition module, used to collect drainage fluid time series monitoring data of multiple postoperative patients; The frequency domain analysis module is used to perform adaptive wavelet packet decomposition on the drainage fluid time series monitoring data of each postoperative patient to obtain the sub-band coefficient matrix corresponding to multiple decomposition layers for each set of drainage fluid time series monitoring data; The cut-off point generation module is used to extract and screen the energy features of multiple sub-band coefficient matrices, determine the energy features of multiple key frequency bands, and determine multiple target cut-off points for each postoperative patient based on the energy features of the key frequency bands; A stage division scheme generation module is used to determine the drainage fluid monitoring stage division results for each postoperative patient based on multiple target cutoff points of each postoperative patient, and to fuse the drainage fluid monitoring stage division results of multiple postoperative patients to generate a target stage division scheme; A feature extraction module is used to obtain the vital sign monitoring data corresponding to the drainage fluid time-series monitoring data of each postoperative patient, perform feature extraction on the drainage fluid time-series monitoring data and vital sign monitoring data of each postoperative patient based on the target stage division scheme, and generate the drainage fluid stage monitoring features and vital sign stage monitoring features of each postoperative patient; The drainage volume monitoring module is used to build a postoperative drainage monitoring model based on the drainage fluid stage monitoring characteristics and physical sign stage monitoring characteristics of multiple patients, and monitor the postoperative drainage volume of target patients based on the postoperative drainage monitoring model.

7. A postoperative drainage volume monitoring system for a drainage tube according to claim 6, characterized in that: The cutoff point generation module extracts and screens energy features of multiple sub-band coefficient matrices to determine multiple key frequency band energy features. Based on these key frequency band energy features, multiple target cutoff points are determined for each postoperative patient, including: Energy features are extracted for each of the multiple sub-bands in each sub-band coefficient matrix to obtain energy features corresponding to each sub-band coefficient matrix, and a first frequency band energy matrix containing multiple frequency band energy features is constructed for each postoperative patient; Performing feature screening on each first-band energy matrix to generate a second-band energy matrix, where the second-band energy matrix includes a plurality of key-band energy features; Performing a sliding window analysis on multiple key frequency band energy features in each second frequency band energy matrix, constructing a characteristic trend curve of each frequency band energy feature and determining multiple first dividing points; Extract the temporal position features and multiple frequency band features of each first demarcation point, construct a candidate feature vector for each first demarcation point, and generate a candidate feature matrix corresponding to each postoperative patient; The multiple first cut-off points of each postoperative patient are clustered using multiple candidate feature matrices to obtain multiple cut-off point clusters, and the multiple second cut-off points of each postoperative patient are determined as the multiple target cut-off points of each postoperative patient based on the multiple cut-off point clusters corresponding to each postoperative patient.

Images