Interventional therapy patient perioperative period pain data analysis method and system

Abstract

The invention relates to the field of medical data analysis, in particular to an interventional therapy patient perioperative period pain data analysis method and system. An interventional therapy patient perioperative period pain data analysis system comprises a data acquisition module, a weight acquisition module, a pain prediction module and a pain monitoring module. According to the method, a standardized multi-source data coding system is established, pre-operative static characteristics and intra-operative dynamic time sequence signals are subjected to nonlinear weight calculation based on mutual information coefficients, complex threshold correlation and interaction effects among parameters are captured, and then a parallel architecture of an LSTM network and a static weighting layer is adopted; deep correlation information of a time sequence dependence mode generated by intraoperative operation and basic background parameters is extracted, the recognition ability of the model for high-risk patients is improved, and high-precision prediction of the pain intensity and type in the perioperative period is achieved.

Description

Technical Field

[0001] The present invention relates to the field of medical data analysis, and particularly to a method and system for analyzing periprocedural pain data of interventional therapy patients. Background Art

[0002] In the field of interventional therapy, periprocedural pain management is a core challenge directly affecting the quality of patients' postoperative recovery and the risk of complications; although minimally invasive techniques have significantly reduced surgical trauma, there are still some patients who will experience moderate to severe postoperative pain, and in some cases, secondary surgery or chronic pain syndrome may be caused by insufficient analgesia; in current clinical practice, pain assessment mostly relies on patients' subjective self-report and medical staff's experience judgment, lacking objective quantitative analysis based on multi-source data, resulting in limited timeliness and accuracy of intervention measures.

[0003] Traditional pain prediction models usually use single-dimensional preoperative indicators (such as age, BMI) or intraoperative physiological parameters (such as average heart rate) for linear regression modeling, and it is difficult to capture complex non-linear relationships such as gene polymorphisms and the co-occurrence effect of underlying diseases; for example, the Met / Met type of the COMT gene (rs4680) will significantly reduce the pain tolerance threshold, but its interaction with the surgical type (such as radiofrequency ablation and cryoablation) is often ignored; in addition, the dynamic correlation between the temporal pattern of energy released by intraoperative equipment and the autonomic response has not been fully explored, resulting in insufficient sensitivity of the prediction model to acute pain events.

[0004] Another defect of the prior art lies in the static nature of the abnormal scoring mechanism; most systems use the fixed threshold method (such as VAS score ≥ 7 points to trigger an alarm), without integrating the dynamic shift of pain types (such as the clinical significance of the transformation from "dull pain" to "radiating pain"); at the same time, the weight allocation of the prediction model depends on expert experience or univariate statistical tests, and it is impossible to accurately quantify the combined influence of multiple preoperative and intraoperative factors on pain type and intensity.

[0005] In view of the above problems, the present invention proposes a method and system for analyzing periprocedural pain data of interventional therapy patients, and establishes a full-process management closed-loop from data collection, dynamic prediction to hierarchical intervention. Summary of the Invention

[0006] The present invention establishes a standardized multi-source data coding system, combines preoperative static features with intraoperative dynamic time-series signals, and captures complex threshold associations and interaction effects between parameters based on a non-linear weight calculation mechanism of mutual information coefficients. Furthermore, a parallel architecture of an LSTM network and a static weighting layer is adopted to separately extract time-series dependence patterns generated by intraoperative operations and deep association information of basic background parameters. The pain intensity and pain type are synchronously output through a dual-task prediction branch, improving the model's ability to identify high-risk patients, avoiding overfitting or omission of key information caused by single-dimensional features, and finally achieving high-precision prediction of pain intensity and type during the perioperative period.

[0007] A method for analyzing perioperative pain data of interventional therapy patients includes: Obtain the patient's preoperative data and intraoperative data. The preoperative data includes age, gender, BMI, underlying diseases, and gene detection results; the intraoperative data includes operation duration, operation type, heart rate variability time-series data, skin electrical activity time-series data, and device released energy time-series data. Calculate the influence degrees of the preoperative data and intraoperative data on the patient's postoperative pain intensity and pain type respectively, and then obtain the pain peak weights and pain type weights of age, gender, BMI, underlying diseases, gene detection results, operation duration, and operation type. Use the patient's preoperative data, intraoperative data, pain peak weights, and pain type weights as inputs to a pain prediction model, and output the predicted pain intensity peak and predicted pain type of the patient after surgery. During the patient's postoperative recovery process, record the patient's pain data at any monitoring time point, including actual pain intensity and actual pain type. If the patient's actual pain intensity exceeds the predicted pain intensity peak or the actual pain type deviates from the predicted pain type, it indicates that the patient's condition is abnormal. Then, calculate the patient's pain abnormality score based on the patient's pain data. Based on the patient's pain abnormality score, implement differential pain management measures for the patient.

[0008] Preferably, the age is directly represented by an integer value; The BMI is represented by a continuous value; The gender is encoded by a binary value, where `0` represents male and `1` represents female; The underlying diseases are represented by a multi-label binary vector; The gene detection result is a classification code for key loci, and each gene locus is independently encoded according to the phenotype type; the GG, AG, and AA genotypes of the OPRM1 gene are encoded as `0`, `1`, and `2` respectively, and the Val / Val, Val / Met, and Met / Met genotypes of the COMT gene are encoded as `0`, `1`, and `2`. The operation duration is an integer-valued parameter; The operation type is represented by serial number coding. Radiofrequency ablation, cryoablation, balloon dilation, and other types are `0`, `1`, `2`, and `3` in sequence; The pain intensity is represented by continuous numerical values, and the acquisition of the pain intensity is based on the visual analogue scale; The pain type is represented by multi-label binary coding.

[0009] Preferably, calculate the influence degrees of the preoperative data and intraoperative data on the pain intensity and pain type of the patient after the operation respectively. The specific operations are as follows: Obtain a number of historical normal patient samples. Each historical normal patient sample contains the age, gender, BMI, underlying diseases, gene test results, operation duration, and operation type of a normal patient, as well as the peak pain intensity and pain type after the operation; numericalize the underlying diseases, gene test results, and pain type in any historical normal patient sample; Based on the independent variable data of all historical normal patient samples and the dependent variable data , i = 1, 2, …, ; n represents the number of historical normal patient samples; j = 1, 2, …, 7; the independent variable data to represent age, gender, BMI, underlying diseases, gene test results, operation duration, and operation type in sequence; k = 1, 2; the dependent variable data represents the peak pain intensity after the operation; the dependent variable data represents the pain type; For each independent variable data and each dependent variable data form an independent analysis group ; For each analysis group perform the following operations: Step 1: Set multiple grid combinations , , ; Traverse all grid combinations. For each historical normal patient sample, determine the position of the value combination of this historical normal patient sample in the current grid combination ; Use the formula to calculate the joint probability distribution , where represents the number of value combinations of n historical normal patient samples that fall into the grid ; Subsequently, use the formula respectively and calculate the marginal probability distribution and , where and respectively represent the value and value of the current grid combination; finally, use the formula to calculate the mutual information coefficient of the independent variable data and the dependent variable data when applying the current grid combination ; Step 2: Based on the mutual information coefficients corresponding to all the obtained grid combinations, select the maximum mutual information coefficient as the influence degree of the independent variable data on the dependent variable data .

[0010] Preferably, obtain the pain peak weight and pain type weight of age, gender, BMI, underlying diseases, gene test results, operation duration, and operation type. The specific operations are as follows: Take the influence degree of any independent variable data on the pain intensity peak as the pain peak weight of the independent variable data ; take the influence degree of any independent variable data on the pain type as the pain type weight of the independent variable data .

[0011] Preferably, the pain prediction model is established based on the LSTM model, including an input layer, a static feature extraction layer, a static feature weighting layer, a temporal feature extraction layer, a feature splicing layer, a fully connected layer, and an output layer.

[0012] Preferably, calculate the pain abnormality score of the patient through the patient's pain data. The specific operations are as follows: Based on the predicted pain intensity peak and the actual pain intensity of the patient, use the formula to calculate and obtain the pain intensity abnormality value ; Obtain the predicted pain type and the actual pain type , which respectively represent the predicted presence and actual presence of the pain types; respectively obtain the intersection size and union size of the actual pain type and the predicted pain type , and calculate the ratio of the intersection to the union to obtain the coincidence degree ; Subsequently, use the formula to calculate and obtain the pain type outlier ; Finally, use the formula to calculate and obtain the pain abnormality score , and are both weight coefficients, , and The values of are optimized by the genetic algorithm.

[0013] A data analysis system for the pain of patients during the perioperative period of interventional therapy, comprising: A data acquisition module for acquiring the preoperative data and intraoperative data of the patient; A weight acquisition module for acquiring the pain peak weights and pain type weights of age, gender, BMI, underlying diseases, gene test results, operation duration, and operation type; A pain prediction module for using the preoperative data, intraoperative data, pain peak weights, and pain type weights of the patient as the input of a pain prediction model, and outputting the predicted pain intensity peak and predicted pain type of the patient after the operation; A pain monitoring module for recording the pain data of the patient at any monitoring time point during the postoperative recovery process of the patient, including the actual pain intensity and actual pain type. If the actual pain intensity of the patient exceeds the predicted pain intensity peak or the actual pain type deviates from the predicted pain type, it indicates that the patient's condition is abnormal. Furthermore, calculate the pain abnormality score of the patient through the pain data of the patient; based on the pain abnormality score of the patient, implement differential pain management measures for the patient.

[0014] The present invention has the following advantages: 1. By establishing a standardized multi-source data coding system, the present invention combines preoperative static features with intraoperative dynamic time series signals, and based on a non-linear weight calculation mechanism of mutual information coefficients, captures complex threshold associations and interaction effects between parameters. Furthermore, a parallel architecture of an LSTM network and a static weighting layer is adopted to respectively extract the time series dependence patterns generated by intraoperative operations and the deep association information of basic background parameters; through a dual-task prediction branch, the pain intensity and pain type are synchronously output, improving the model's ability to identify high-risk patients, avoiding overfitting or omission of key information caused by single-dimensional features, and finally realizing high-precision prediction of the pain intensity and type during the perioperative period.

[0015] 2. The present invention replaces the traditional fixed threshold alarm rule through a dynamic abnormality scoring mechanism, combines the genetic algorithm to optimize the priority of the response strategy, forms a "prediction-monitoring-intervention" closed-loop management process, greatly shortens the clinical decision-making delay, and reduces the risk of secondary complications caused by pain out of control. Description of the Drawings

[0016] Figure 1 This is a schematic structural diagram of a data analysis system for perioperative pain of patients undergoing interventional therapy according to an embodiment of the present invention. Specific embodiments

[0017] In order to enable those skilled in the art to 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 with reference to the accompanying drawings in the embodiments of the present invention.

[0018] Embodiment 1, a method for analyzing perioperative pain data of patients undergoing interventional therapy, includes: Obtain the preoperative data and intraoperative data of the patient. The preoperative data includes age, gender, BMI, underlying diseases, and gene detection results; the intraoperative data includes the operation duration, operation type, heart rate variability time series data, skin electrical activity time series data, and device energy release time series data. By collecting the patient's preoperative baseline information (such as age, gender, and BMI reflecting physiological compensatory ability, the combination of underlying diseases revealing the synergistic damage of comorbidities to the pain pathway, and the gene detection results analyzing the genetic differences in drug metabolism and pain sensitivity) and intraoperative multi-dimensional dynamic parameters (the operation duration and type quantifying the intensity of operative trauma, the heart rate variability and skin electrical activity time series signals capturing the stress fluctuations of the autonomic nervous system, and the device energy release curve mapping the real-time cumulative effect of tissue damage), multi-dimensional and high-resolution pain-causing factor inputs can be provided for the subsequent model, thereby realizing full-link pain management support from preoperative risk warning to intraoperative real-time regulation; Calculate the influence degrees of the preoperative data and intraoperative data on the postoperative pain intensity and pain type of the patient respectively, and then obtain the pain peak weights and pain type weights of age, gender, BMI, underlying diseases, gene detection results, operation duration, and operation type. The core of this step is to solve the problem that traditional analgesic regimens rely on empirical rules and ignore individual heterogeneity by quantifying the differential effects of multi-source variables in the perioperative period on pain outcomes; since different factors (such as genotypes mainly change the pain type distribution by regulating the sensitivity of opioid receptors, while the cumulative amount of intraoperative energy release is non-linearly positively correlated with pain intensity) have asymmetric threshold effects on pain intensity (continuous variable) and pain type (discrete classification, such as neuropathic pain, inflammatory pain, etc.), a dual-channel correlation matrix of "pain peak weight" and "pain type weight" needs to be established respectively; for example, the underlying disease network may increase the intensity risk in a superimposed manner (the weight value accumulatively increases), while a specific operation type (such as operation near the nerve plexus) significantly increases the weight probability of neuropathic pain; by screening out key pain-causing factors, guiding the model to perform adaptive feature weighting on the pain intensity regression task and multi-label classification task, thereby improving the ability to subdivide and identify complex pain phenotypes, and providing an interpretable dose-target optimization basis for postoperative analgesic drug selection; Taking the patient's preoperative data, intraoperative data, pain peak weight, and pain type weight as the input of the pain prediction model, the predicted postoperative pain intensity peak and predicted pain type of the patient are output; by introducing the pain peak weight and pain type weight into the model input, the system can structurally weight the heterogeneous contributions of preoperative and intraoperative multidimensional data, achieving "prior transmission of feature importance"; for example, if the pain type weight of a gene test result (such as OPRM1 mutation) is significantly higher than other features, the model will give priority to the logical association between this feature and neuropathic pain in the classification task, while suppressing irrelevant noise; and the peak weight of the intraoperative energy release curve guides the regression model to more sensitively capture key mutation points in the time series signal (such as the instantaneous pain surge corresponding to high-frequency energy pulses) by quantifying its non-linear driving effect on the pain intensity threshold (such as the dynamic amplification of tissue damage by energy accumulation); this weight-based feature adaptive screening mechanism can not only solve the "dilution effect" caused by redundant data dimensions in traditional models (key signals are covered by low-correlation features), but also enhance the modeling ability for individual-specific pain-causing patterns (such as the interaction between the genotype of the elderly and long-term surgery), thereby improving the joint prediction accuracy and clinical interpretability of pain intensity grading and type classification; During the patient's postoperative recovery process, the patient's pain data, including the actual pain intensity and actual pain type, are recorded at any monitoring time point. If the patient's actual pain intensity exceeds the predicted pain intensity peak or the actual pain type deviates from the predicted pain type, it indicates that the patient's condition is abnormal, and then the patient's pain abnormality score is calculated based on the patient's pain data; the core significance of the pain abnormality score is to convert the real-time monitoring of postoperative pain into a quantifiable risk warning indicator. Its calculation is not simply a comparison of the deviation between the predicted value and the actual value, but a dynamic risk index is constructed through multi-layer abnormal signal fusion; Differentiated pain management measures are implemented for patients based on their pain abnormality scores. For example, when a patient's pain abnormality score is in the low-risk range, the system preferentially takes dynamic observation and non-pharmacological interventions (such as adjusting body position, low-temperature analgesic patches); when it is in the medium-risk range, targeted drugs are matched according to the pain type weight (such as gabapentin for neuropathic pain and NSAIDs added for inflammatory pain), and the recommended dose is adjusted based on the patient's gene metabolism parameters; when it is in the high-risk range, multimodal collaborative treatment is triggered (such as the combination of an analgesic pump and ultrasound-guided nerve block), and at the same time, potential injury areas are predicted based on the intraoperative energy release weight for preventive intervention; in the case of critical risk, a multidisciplinary consultation is initiated, and an intensive analgesia plan (such as epidural analgesia or intrathecal drug administration) is formulated according to the abnormal score characteristics (such as persistent over-limit accompanied by type mutation), and the drug metabolism model parameters are optimized through real-time feedback; the processing logics for different grades are not only based on the score value itself, but also dynamically adjust the individualized plan through the pain type weight (such as preferential interventional treatment for patients with neuropathic pain) and the time cumulative effect, avoiding over-medical treatment and ensuring that high-risk patients receive timely and accurate analgesic support.

[0019] Age is directly represented by an integer value, based on the patient's actual age in full years (rounding down to the nearest whole year according to the date of birth if less than a full year); BMI (Body Mass Index) is represented by a floating-point value, accurate to one decimal place, and is calculated as weight (kg) divided by the square of height (m); for example, for a patient with a weight of 70 kg and a height of 1.75 m, the calculated BMI result is `22.9`; Gender is encoded by a binary value, where `0` represents male and `1` represents female; Underlying diseases are represented by a multi-label binary vector; the vector dimension order strictly follows the preset disease list (such as diabetes, hypertension, chronic obstructive pulmonary disease, etc.), each dimension corresponds to a disease, with the presence being `1` and otherwise `0`; for example, the vector `[1,0,1,0]` indicates that the patient has both diabetes and chronic obstructive pulmonary disease; the disease list needs to be consistent with the preset disease order during model training to avoid prediction deviation caused by dimension misalignment; The gene test results are categorical encodings for key loci, and each gene locus is independently encoded according to the phenotypic type; the GG, AG, and AA genotypes of the OPRM1 gene (rs1799971) are encoded as `0`, `1`, and `2` respectively, while the Val / Val, Val / Met, and Met / Met genotypes of the COMT gene (rs4680) are encoded as `0`, `1`, and `2`; for example, `[2,1]` indicates that OPRM1 is of the AA type and COMT is of the Val / Met type; The operation duration is an integer numerical parameter in minutes, defined as the actual time elapsed from the first activation of the surgical instrument to the completion of the final suture; The surgical types are represented by serial numbers. Radiofrequency ablation, cryoablation, balloon dilation, and others are `0`, `1`, `2`, and `3` in sequence; The pain intensity is represented by continuous numerical values, and the basis for obtaining the pain intensity is the Visual Analogue Scale (VAS); The pain types are represented by multi-label binary encoding. The order of the vector dimensions strictly follows the preset pain type list (such as stabbing pain, burning pain, dull pain, radiating pain, spastic pain, etc.). Each dimension corresponds to a pain type. If it exists, it is `1`, otherwise it is `0`.

[0020] Calculate the influence degrees of the preoperative data and intraoperative data on the postoperative pain intensity and pain types of the patient respectively. The specific operations are as follows: Obtain a number of historical normal patient samples. Each historical normal patient sample contains the age, gender, BMI, underlying diseases, gene test results, operation duration, and surgical type of a normal patient, as well as the peak postoperative pain intensity and pain types; Numerically process the underlying diseases, gene test results, and pain types in any historical normal patient sample; The numerical processing is carried out using the preset position ordinal weighted summation method. For each dimension marked as existing in the vector, its numerical contribution degree is determined by the order position of this dimension in the preset list (the first dimension is 1, and the subsequent ones increase sequentially. The weight of the z-th dimension is z). The final result is the sum of the values of all existing dimensions multiplied by the numerical contribution degrees of these dimensions respectively; For example, the numerical processing result of the vector [1, 0, 1, 0] is 1 + 3 = 4, the numerical processing result of the vector [2, 1] is 2 + 2 = 4, and the numerical processing result of the vector [0, 2] is 4; Based on the independent variable data of all historical normal patient samples and the dependent variable data , i = 1, 2, …, ; where \(n\) represents the number of historical normal patient samples; j = 1, 2, …, 7; The independent variable data \(x_{ij}\) to \(x_{i7}\) represent age, gender, BMI, underlying diseases, gene test results, operation duration, and surgical type in sequence; k = 1, 2; The dependent variable data \(y_{ik}\) represents the peak postoperative pain intensity; The dependent variable data \(y_{i2}\) represents the pain type; For each independent variable data \(x_{ij}\) and each dependent variable data \(y_{ik}\), form an independent analysis group ; For each analysis group perform the following operations: Step 1: Set multiple grid combinations , , ; Traverse all grid combinations. For each historical normal patient sample, determine the position of the value combination of this historical normal patient sample in the current grid combination ; Use the formula to calculate the joint probability distribution , where represents the number of value combinations of historical normal patient samples that fall into the grid ; Subsequently, use the formulas and to calculate the marginal probability distributions and , where and represent the value and value of the current grid combination respectively; Finally, use the formula to calculate the mutual information coefficient of the independent variable data and the dependent variable data when applying the current grid combination ; Step 2: Based on the mutual information coefficients corresponding to all obtained grid combinations, select the maximum mutual information coefficient as the influence degree of the independent variable data on the dependent variable data .

[0021] Obtain the pain peak weight and pain type weight of age, gender, BMI, underlying diseases, gene test results, operation duration, and operation type. The specific operations are as follows: Take the influence degree of any independent variable data on the pain intensity peak as the pain peak weight of the independent variable data ; Take the influence degree of any independent variable data on the pain type as the pain type weight of the independent variable data .

[0022] The pain prediction model is established based on the LSTM model, including an input layer, a static feature extraction layer, a static feature weighting layer, a temporal feature extraction layer, a feature splicing layer, a fully connected layer, and an output layer; The input layer is used to receive the patient's age, gender, BMI, underlying diseases, gene test results, operation duration, operation type, heart rate variability time series data, skin electrical activity time series data, and device released energy time series data; The static feature extraction layer is used to perform non-linear transformation and feature enhancement on age, gender, BMI, underlying diseases, gene test results, operation duration, and operation type, and output a static feature vector; The static feature weighting layer includes a pain peak prediction branch and a pain type prediction branch. The pain peak prediction branch is used to utilize the pain peak weight to weight the static feature vector and obtain the first weighted static feature vector; the pain type prediction branch is used to utilize the pain type weight to weight the static feature vector and obtain the second weighted static feature vector; The temporal feature extraction layer is used to capture the temporal features of heart rate variability temporal data, skin electrical activity temporal data, and device released energy temporal data through an LSTM network, and output a temporal feature vector; The feature concatenation layer is used to concatenate the first weighted static feature vector with the temporal feature vector to obtain the first joint feature vector; at the same time, concatenate the second weighted static feature vector with the temporal feature vector to obtain the second joint feature vector; The fully connected layer includes a fully connected layer for the pain peak branch and a fully connected layer for the pain type branch. The fully connected layer for the pain peak branch is used to perform non-linear transformation on the first joint feature vector; the fully connected layer for the pain type branch is used to perform non-linear transformation on the second joint feature vector; The output layer includes a pain peak prediction task branch and a pain type prediction task branch. The pain peak prediction task branch is used to linearly activate the output of the fully connected layer for the pain peak branch and output the predicted pain intensity peak; the pain type prediction task branch is used to perform Softmax activation on the output of the fully connected layer for the pain type branch and output the predicted pain type.

[0023] For the training of the pain prediction model, the specific operations are as follows: Take the postoperative pain intensity peak and pain type in any historical normal patient sample as the label of the current historical normal patient sample. Divide all historical normal patient samples into a training set and a validation set. Use the training set to train the pain prediction model with initialized parameters, and verify the pain prediction model through the validation set to obtain a verification result; set training conditions, and judge whether the verification result meets the training conditions. If so, output the trained pain prediction model; if not, continue to use the training set to train the pain prediction model.

[0024] Calculate the pain abnormality score of the patient through the patient's pain data. The specific operations are as follows: Based on the predicted pain intensity peak of the patient and the actual pain intensity , use the formula to calculate and obtain the pain intensity abnormality value ; Obtain the predicted pain type and the actual pain type , which respectively represent the presence of prediction and the actual presence of various pain types; respectively obtain the actual pain type and the predicted pain type to calculate the sizes of the intersection and union, and calculate the ratio of the intersection to the union to obtain the degree of overlap ; Subsequently, use the formula to calculate and obtain the pain type outlier ; Finally, use the formula to calculate and obtain the pain anomaly score , and are both weight coefficients, , and The values of are optimized by a genetic algorithm.

[0025] Example 2, a data analysis system for perioperative pain of interventional treatment patients, as Figure 1 shown, includes: A data acquisition module for acquiring the patient's preoperative data and intraoperative data. The preoperative data includes age, gender, BMI, underlying diseases, and gene test results; the intraoperative data includes the operation duration, operation type, heart rate variability time series data, skin electrical activity time series data, and device released energy time series data; A weight acquisition module for calculating the influence degrees of the preoperative data and intraoperative data on the patient's postoperative pain intensity and pain type respectively, and then obtaining the pain peak weights and pain type weights of age, gender, BMI, underlying diseases, gene test results, operation duration, and operation type; A pain prediction module for using the patient's preoperative data, intraoperative data, as well as the pain peak weights and pain type weights as inputs to a pain prediction model, and outputting the predicted pain intensity peak and predicted pain type of the patient after surgery; A pain monitoring module for recording the patient's pain data, including the actual pain intensity and actual pain type, at any monitoring time point during the patient's postoperative recovery. If the patient's actual pain intensity exceeds the predicted pain intensity peak or the actual pain type deviates from the predicted pain type, it indicates that the patient's condition is abnormal, and then calculates the patient's pain anomaly score based on the patient's pain data; based on the patient's pain anomaly score, implements differential pain management measures for the patient.

[0026] It should be understood that those of ordinary skill in the art can make improvements or transformations based on the above description, and all such improvements and transformations shall fall within the protection scope of the appended claims of the present invention. The parts not described in detail in this specification belong to the prior art well-known to those of ordinary skill in the art.

Claims

1. A method for analyzing perioperative pain data of patients undergoing interventional therapy, characterized in that: include: Obtain the patient's preoperative and intraoperative data, including age, gender, BMI, underlying diseases, and genetic test results; Intraoperative data included operation duration, operation type, heart rate variability time series data, skin electrode activity time series data, and device energy release time series data; Calculate the impact of preoperative and intraoperative data on the patient's postoperative pain intensity and pain type, and then obtain the pain peak weight and pain type weight based on age, gender, BMI, underlying disease, genetic test results, operation duration, and operation type; The patient's preoperative data, intraoperative data, pain peak weight and pain type weight are used as inputs to the pain prediction model, and the patient's postoperative predicted pain intensity peak and predicted pain type are output; During the patient's postoperative recovery process, the patient's pain data is recorded at any monitoring time point, including the actual pain intensity and the actual pain type. If the patient's actual pain intensity exceeds the predicted pain intensity peak or the actual pain type deviates from the predicted pain type, it means that the patient's condition is abnormal, and then the patient's pain abnormality score is calculated based on the patient's pain data; Differentiated pain management measures are implemented for patients based on their pain abnormality scores.

2. A method for analyzing perioperative pain data of interventional therapy patients according to claim 1, characterized in that: Age is directly expressed using an integer value; BMI is expressed as a continuous value; Gender is represented by binary value encoding, where `0` represents male and `1` represents female; The underlying disease is represented by a multi-label binary vector; The results of gene testing are classified codes for key loci, and each gene locus is independently coded according to the phenotypic type; the GG, AG, and AA genotypes of the OPRM1 gene are coded as `0`, `1`, and `2`, respectively, and the Val / Val, Val / Met, and Met / Met genotypes of the COMT gene are coded as `0`, `1`, and `2`; The duration of surgery is a numerical parameter for plastic surgery; The type of surgery is represented by a serial number code, with radiofrequency ablation, cryoablation, balloon dilatation, and other types being `0`, `1`, `2`, and `3`, respectively; Pain intensity was expressed as a continuous numerical value, and pain intensity was obtained based on a visual analogue scale; The pain type is represented by multi-label binary coding.

3. A method for analyzing perioperative pain data of interventional therapy patients according to claim 2, characterized in that: Calculate the influence of preoperative data and intraoperative data on the patient's postoperative pain intensity and pain type respectively. The specific operation is as follows: Obtain several historical normal patient samples, each of which contains the age, gender, BMI, underlying disease, genetic test results, operation duration and type, and postoperative pain intensity peak and pain type of a normal patient; Numerize the underlying diseases, genetic test results, and pain types in any historically normal patient sample; Independent variable data based on all historical normal patient samples And the dependent variable data , =1, 2, …, ; represents the number of historical normal patient samples; =1, 2, ..., 7; independent variable data to They represent age, sex, BMI, underlying diseases, genetic test results, duration of surgery, and type of surgery, respectively; =1,2; dependent variable data Indicates the peak postoperative pain intensity; dependent variable data Indicates the type of pain; For each independent variable data With each dependent variable data Formation of independent analysis group ; For each analysis group Do the following: Step 1: Set up multiple grid combinations , , ; Traverse all grid combinations, for each historical normal patient sample, determine the value combination of the historical normal patient sample in the current grid combination position; using the formula Compute joint probability distribution ,in, express The value combinations of historical normal patient samples fall into the grid The number of; then use the formula and Calculate marginal probability distribution and ,in, and Respectively represent the current grid combination Value and value; finally, using the formula Calculate the current grid combination Time independent variable data With dependent variable data The mutual information coefficient ; Step 2: Based on the mutual information coefficients corresponding to all grid combinations obtained, select the largest mutual information coefficient as the independent variable data For dependent variable data degree of impact.

4. The method for analyzing perioperative pain data of an interventional therapy patient according to claim 3, characterized in that: Obtain the pain peak weight and pain type weight of age, gender, BMI, underlying diseases, genetic test results, surgery duration, and surgery type. The specific operations are as follows: Any independent variable data The degree of influence on the peak pain intensity was used as the independent variable data Peak pain weight ; Any independent variable data The degree of influence on pain type as independent variable data Weight of pain type .

5. The method for analyzing perioperative pain data of an interventional therapy patient according to claim 4, characterized in that: The pain prediction model is built based on the LSTM model, including input layer, static feature extraction layer, static feature weighting layer, temporal feature extraction layer, feature concatenation layer, fully connected layer and output layer.

6. A method for analyzing perioperative pain data of interventional therapy patients according to claim 5, characterized in that: The patient's pain abnormality score is calculated based on the patient's pain data. The specific operation is as follows: Based on the patient's predicted peak pain intensity and actual pain intensity , using the formula Calculate the abnormal value of pain intensity ; Get predicted pain type and actual pain type , respectively The predicted and actual existence of the pain types; the actual pain types are obtained respectively Predicting pain type The size of the intersection and the size of the union, and calculate the ratio of the intersection to the union to obtain the degree of overlap ; Then use the formula Calculate the abnormal value of pain type ; Finally, using the formula Calculate the pain score , and are weight coefficients, , and The value of is optimized by genetic algorithm.

7. A perioperative pain data analysis system for interventional therapy patients, characterized in that: The system is applied to a method for analyzing perioperative pain data of an interventional treatment patient as described in any one of claims 1 to 6, comprising: A data acquisition module, used to acquire the patient's preoperative data and intraoperative data; A weight acquisition module, used to obtain the pain peak weight and pain type weight of age, gender, BMI, underlying diseases, genetic test results, surgery duration and surgery type; A pain prediction module is used to use the patient's preoperative data, intraoperative data, pain peak weight and pain type weight as inputs to the pain prediction model, and output the patient's postoperative predicted pain intensity peak and predicted pain type; The pain monitoring module is used to record the patient's pain data at any monitoring time point during the patient's postoperative recovery process, including the actual pain intensity and the actual pain type. If the patient's actual pain intensity exceeds the predicted pain intensity peak or the actual pain type deviates from the predicted pain type, it means that the patient's condition is abnormal, and the patient's abnormal pain score is calculated based on the patient's pain data; based on the patient's abnormal pain score, differentiated pain management measures are implemented for the patient.

Images