A method and system for analyzing pain status in patients with liver cancer
By constructing a multi-dimensional pain abnormality feature matrix and a spatiotemporal constraint density clustering algorithm, the problem of difficult to accurately assess the pain status of liver cancer patients after surgery was solved, personalized pain management and dynamic early warning were achieved, and the patient recovery effect was improved.
Patent Information
- Application Number
- CN202510210042.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-25
- Publication Date
- 2025-09-30
- Estimated Expiration
- 2045-02-25
AI Technical Summary
Existing technologies make it difficult to accurately assess the postoperative pain status of liver cancer patients. Traditional pain assessment methods have low accuracy when the patient is weak or sedated, and frequent assessments increase the burden on patients, affecting the accuracy of analgesia plans.
By collecting patients' postoperative physiological monitoring data, liver function recording data and pain assessment data, a multi-dimensional pain abnormality feature matrix is constructed. A density clustering algorithm with time and space constraints is used to explore the pain evolution law, generate a personalized pain assessment strategy, and combine it with patients' real-time physiological monitoring data for dynamic matching and early warning.
It achieves accurate assessment of patients' pain status in scenarios of blurred consciousness or language disorders, provides personalized pain management plans, and improves postoperative recovery effects.
Smart Images
Figure CN120072327B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of pain state analysis, and in particular to a method and system for analyzing the pain state of liver cancer patients. Background Art
[0002] Liver cancer is one of the most common malignant tumors worldwide, and surgical resection remains the preferred treatment for early-stage and some mid-stage liver cancers. However, traditional open hepatocellular carcinoma resection is highly invasive and often accompanied by moderate to severe postoperative pain, which seriously affects patients' recovery and quality of life. Accurately assessing postoperative pain status and developing personalized intervention strategies have become major challenges in perioperative management.
[0003] Currently, commonly used pain assessment methods in clinical practice mainly include pain recognition technology based on patients' facial expressions and subjective scale scoring methods. However, these methods have certain limitations. When some patients are weak after surgery or affected by sedatives, there may be a deviation between their facial expressions and their actual pain perception. Personal habits and inadvertent control of facial expressions may also lead to a decrease in recognition accuracy. Subjective scoring represented by the Numeric Rating Scale (NRS) or Visual Analog Scale (VAS) requires the patient's active cooperation in filling out and is manually interpreted by medical staff or background programs. Frequent repeated assessments can easily increase the burden on patients and are difficult to implement in scenarios with blurred consciousness or language barriers.
[0004] During the analysis of postoperative pain in patients with liver cancer, the physiological mechanism of postoperative pain is closely related to the patient's postoperative liver function compensation capacity, inflammatory response level, etc. The liver function level will affect the patient's pain level to a certain extent, and the patient's physiological data such as heart rate and respiratory rate may be higher than normal levels in the painful state. Relying solely on a single data dimension can easily lead to a disconnect between the pain assessment results and the actual pathological state, thereby affecting the accuracy of the analgesic plan. Therefore, there is an urgent need for a pain state analysis method based on multi-source data fusion. By mining the correlation between data such as liver function and physiological parameters in the pain process, it is possible to provide patients with high-precision and reliable postoperative pain management services and improve the patient's postoperative recovery effect. Summary of the Invention
[0005] In order to solve the technical problems raised in the above background technology, the present invention proposes a method and system for analyzing the pain status of liver cancer patients.
[0006] As a first aspect of the present invention, a method for analyzing the pain status of a liver cancer patient is provided, comprising:
[0007] Collecting pain status monitoring data from patients after open liver cancer surgery, including physiological monitoring data, liver function recording data, and pain assessment data, and identifying multiple periods of abnormal pain in the patients based on the pain assessment data in the pain status monitoring data;
[0008] Extract local physiological change characteristics, local liver function characteristics, and local pain assessment characteristics during each abnormal pain period, construct a local pain abnormality feature vector for each abnormal pain period, and generate a pain abnormality feature matrix for each patient;
[0009] Performing time evolution analysis on the pain abnormality feature matrices of multiple patients to generate multiple local time evolution patterns, and generating multiple group pain evolution laws based on the multiple local time evolution patterns;
[0010] Matching each patient with multiple group pain evolution patterns to generate pain evolution pattern matching results, generating multiple global pain abnormality association patterns based on the pain evolution pattern matching results, performing pain resistance analysis on multiple patients included in each global pain abnormality association pattern, and generating a pain resistance parameter for each patient;
[0011] A pain assessment strategy for a global pain abnormality association pattern is generated based on the pain abnormality feature matrices and pain resistance parameters corresponding to multiple patients under the global pain abnormality association pattern. The pain state of the object to be analyzed is analyzed based on multiple pain assessment strategies to obtain a pain state analysis result of the object to be analyzed.
[0012] Preferably, a time period evolution analysis is performed on the pain abnormality feature matrices of multiple patients to generate multiple local time period evolution patterns, and multiple group pain evolution laws are generated based on the multiple local time period evolution patterns, including:
[0013] Based on a preset sliding window, the abnormal pain feature matrix of each patient is traversed to generate multiple local window feature vectors, the window period parameter of each local window feature vector is determined, the multiple local window feature vectors are clustered using DBSCAN, and the clustering process is temporally constrained based on the window period parameter to generate multiple window feature clusters;
[0014] The timing constraint of the clustering process based on the window period parameter includes calculating the timing constraint distance between any two local window feature vectors based on the window period parameter, and performing DBSCAN clustering on multiple local window feature vectors according to the timing constraint distance. The timing constraint distance is calculated as follows:
[0015] L(i,j)=α·L f (i,j)+(1-α)·L t (i,j)
[0016] Where L(i,j) is the temporal constraint distance between the local window feature vector i and the local window feature vector j, L f(i, j) is the spatial distance parameter between the local window feature vector i and the local window feature vector j, which is calculated based on the local window feature vector i and the local window feature vector j. t (i, j) is the temporal distance parameter between the local window feature vector i and the local window feature vector j, which is calculated based on the window period parameters corresponding to the local window feature vector i and the local window feature vector j respectively, and α is the spatiotemporal adjustment weight;
[0017] Based on the preset sliding window, multiple recovery periods corresponding to the patient are determined, the period coverage of each window feature cluster is counted and multiple pattern transfer parameters of each window feature cluster are calculated, and multiple local period evolution patterns are generated according to the pattern transfer parameters. The multiple local period evolution patterns are screened for evolutionary importance to determine multiple target period evolution patterns, and the group pain evolution law of each target period evolution pattern is generated.
[0018] Preferably, each patient is matched with multiple group pain evolution laws to generate pain evolution law matching results, and multiple global pain abnormality association patterns are generated based on the pain evolution law matching results, including:
[0019] Extract the core pain feature vector of each window feature cluster, generate a pain evolution feature matrix for each group's pain evolution law based on multiple core pain feature vectors, and perform matching analysis on the patient's local window feature vectors in each recovery period using the pain evolution feature matrix. This includes calculating the matching parameters between the core pain feature vector corresponding to the pain evolution feature matrix and the patient's local window feature vector in the same recovery period, and generating a global matching degree between the patient and the pain evolution feature matrix based on the matching parameters across multiple recovery periods.
[0020] The patient's pain evolution law matching result is generated based on the global matching degree corresponding to the patient and multiple group pain evolution laws, multiple global pain abnormality association patterns are determined based on the multiple group pain evolution laws, and multiple patients are classified based on the multiple global pain abnormality association patterns, including assigning the patient to the global pain abnormality association pattern corresponding to the group pain evolution law with the largest global matching degree, thereby generating multiple global pain abnormality association patterns containing multiple patients.
[0021] Preferably, a pain assessment strategy for generating a global pain abnormality association pattern based on the pain abnormality feature matrix and pain resistance parameters corresponding to multiple patients in the global pain abnormality association pattern includes:
[0022] Based on the pain resistance parameter, local pain resistance analysis is performed on multiple patients under the global pain abnormality association model, and multiple patients are divided into multiple resistance category groups;
[0023] Based on the pain abnormality feature matrix, multiple feature abnormality information of patients in each resistance category group in each recovery period is extracted respectively. Based on the multiple abnormal information, multiple physiological state reference ranges and pain state reference ranges of patients in each resistance category group in the recovery period are extracted. A local state analysis strategy for each resistance category group in multiple recovery periods is constructed, and a pain assessment strategy of the global pain abnormality association pattern is generated based on the multiple local state analysis strategies.
[0024] Preferably, a plurality of local period evolution patterns are generated according to the pattern transfer parameters, including:
[0025] For multiple local time period evolution patterns, they include the conversion paths between multiple window feature clusters and the time period transition probability between any two adjacent window feature clusters in the conversion path. Each window feature cluster in the conversion path corresponds to a recovery period. The time period transition probability between two adjacent window feature clusters in the conversion path is calculated using multiple mode transition parameters of the window feature clusters. The calculation formula is as follows:
[0026]
[0027] Where, P ab M represents the mode transfer parameter of the window feature cluster a in the recovery period t being converted to the window feature cluster b in the recovery period t+1. ta represents the time period coverage of window feature cluster a in the recovery period t, M (t+1)b represents the proportion of patients who belong to window feature cluster a in recovery period t and are converted to patients who belong to window feature cluster b in recovery period t+1;
[0028] After calculating multiple mode transfer parameters between any two adjacent recovery periods and between any two window feature clusters using the above formula, multiple local period evolution patterns are constructed based on multiple recovery periods and window feature clusters.
[0029] Preferably, performing evolution importance screening on multiple local period evolution patterns to determine multiple target period evolution patterns includes:
[0030] For any local period evolution pattern, if the period transition probability between any two adjacent window feature clusters in the transition path of the local period evolution pattern is greater than the preset evolution threshold, the transition path of the local period evolution pattern is recorded as the target period evolution pattern.
[0031] As a second aspect of the present invention, a system for analyzing the pain state of a liver cancer patient is provided. The system is used to implement the above-mentioned method for analyzing the pain state of a liver cancer patient, comprising:
[0032] A data acquisition module is used to collect pain status monitoring data of patients after open liver cancer surgery, including physiological monitoring data, liver function recording data, and pain assessment data;
[0033] A local analysis module is used to determine multiple abnormal pain periods of the patient based on the pain assessment data in the pain status monitoring data, extract the local physiological change characteristics, local liver function characteristics and local pain assessment characteristics of each abnormal pain period, construct a local pain abnormality feature vector for each abnormal pain period, and generate a pain abnormality feature matrix for each patient;
[0034] The pain local evolution analysis module is used to perform time evolution analysis on the pain abnormality feature matrices of multiple patients to generate multiple local time evolution patterns, generate multiple group pain evolution laws based on the multiple local time evolution patterns, match each patient with the multiple group pain evolution laws to generate pain evolution law matching results, and generate multiple global pain abnormality association patterns based on the pain evolution law matching results;
[0035] The pain assessment and analysis module is used to perform pain resistance analysis on multiple patients included in each global pain abnormality association pattern, generate pain resistance parameters for each patient, generate a pain assessment strategy for the global pain abnormality association pattern based on the pain abnormality feature matrix and pain resistance parameters corresponding to multiple patients in the global pain abnormality association pattern, analyze the pain status of the object to be analyzed based on multiple pain assessment strategies, and obtain the pain status analysis results of the object to be analyzed.
[0036] The present invention has the following beneficial effects:
[0037] The present invention integrates the liver function indicators, dynamic physiological parameters and pain score data of postoperative patients to construct a multi-dimensional pain abnormality feature matrix, adopts a density clustering algorithm based on time and space constraints to mine the pain evolution law of the patient group, determines multiple local time period evolution patterns, quantifies the correlation between the time period evolution pattern and the pain state to generate multiple group pain evolution laws, and realizes hierarchical strategy modeling in combination with the individual pain resistance level. It integrates the common laws of the group and the individual resistance differences to generate an adaptive pain assessment strategy. It can combine the patient's real-time physiological monitoring data and liver function data to dynamically match and warn the patient's pain development trend, and provide clinical with accurate pain management solutions. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] Figure 1 A flowchart of a method for analyzing pain status in liver cancer patients provided by an embodiment of the present invention.
[0039] Figure 2 A schematic structural diagram of a pain status analysis system for liver cancer patients provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0040] In order to enable those skilled in the art to better understand the technical solutions of 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 of 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 intended to limit the present invention.
[0041] See Figure 1 The present invention provides a method for analyzing the pain status of a liver cancer patient, comprising the following steps:
[0042] Step S1: Collect the patient's pain status monitoring data after open liver cancer surgery, including physiological monitoring data, liver function recording data and pain assessment data, and determine the patient's multiple abnormal pain periods based on the pain assessment data in the pain status monitoring data.
[0043] In this embodiment, the pain status monitoring data of patients after open liver cancer surgery can reflect the dynamic changes in the patient's postoperative pain. Physiological monitoring data includes, but is not limited to, body temperature, heart rate, respiratory rate, blood pressure, and blood oxygen saturation, which can reflect the patient's overall physiological state during the postoperative recovery process. Liver function records include liver function-related indicators collected regularly from patients, such as ALT (alanine aminotransferase), AST (aspartate aminotransferase), bilirubin, and albumin. This can provide insights into changes in liver function. Unstable liver function is closely related to the patient's pain perception and recovery process. Pain assessment data can be used to quantitatively assess the patient's pain status using commonly used pain rating scales such as the Visual Analog Scale (VAS) and the Numerical Rating System (NRS), providing core information for pain status analysis. Multiple abnormal pain periods can be determined based on the pain assessment data, or based on the patient's subjective feedback. The range of different abnormal pain periods can be reasonably determined by the duration of the patient's physiological parameters exceeding normal levels. Abnormal pain periods specifically refer to sudden or persistent pain states that may occur within a certain period of time. In order to record different pain states, the patient's data can be collected through a pain rating scale when the patient's condition is relatively good. Although there is a delay in the collection process, that is, the data may be collected after the patient's pain state is relieved, the collected data can serve as an important basis for patient pain assessment. By collecting a large amount of sample information, the changing patterns of pain states in patients during the recovery period after open liver cancer surgery can be explored.
[0044] Step S2: extract the local physiological change characteristics, local liver function characteristics and local pain assessment characteristics of each abnormal pain period, construct a local pain abnormality feature vector for each abnormal pain period, and generate a pain abnormality feature matrix for each patient.
[0045] In this embodiment, the pain status monitoring data for each abnormal pain period is analyzed separately, representative local features are extracted, and a feature vector for each period is formed. Specifically, by analyzing the changes in physiological data during the abnormal pain period, information such as the mean, rate of change, and variability parameters of various physiological parameters, such as heart rate and respiratory rate, is extracted to obtain local physiological change characteristics for each abnormal pain period. The average levels of various liver function indicators in the liver function recording data are analyzed to obtain local liver function characteristics for the abnormal pain period, thereby better understanding the relationship between pain and liver function changes. Information such as the peak and mean pain scores within the period, as well as the duration of severe pain, can also be extracted from the pain assessment data to obtain local pain assessment characteristics for the abnormal pain period, reflecting the complexity and multidimensionality of pain. Finally, the extracted multiple local features are combined to obtain a local pain abnormality feature vector for each abnormal pain period. These multiple feature vectors are then spliced according to their time sequence to form a pain abnormality feature matrix for each patient. This matrix contains the pain changes of multiple patients over multiple periods and can characterize each patient's personalized pain pattern.
[0046] Step S3: Perform time evolution analysis on the pain abnormality feature matrices of multiple patients to generate multiple local time evolution patterns, generate multiple group pain evolution laws based on the multiple local time evolution patterns, match each patient with the multiple group pain evolution laws to generate pain evolution law matching results, and generate multiple global pain abnormality association patterns based on the pain evolution law matching results.
[0047] In this embodiment, by analyzing the evolution of the patient's pain state over time, specifically traversing the patient's pain abnormality feature matrix through a specific sliding window, extracting multi-dimensional statistical features in different windows, and analyzing the correlation between different window features when the data is time-series, multiple local time period evolution patterns existing in multiple patients are identified. For example, in some patients, pain symptoms may worsen with the deterioration of liver function during a specific period, while other patients may show specific patterns in physiological parameters when they are in severe pain. Then, some representative patterns are identified from the multiple local time period evolution patterns to determine the multiple group pain evolution patterns corresponding to the groups formed by multiple patients, and each representative pattern represents one of the pain evolution patterns. After determining the pain evolution laws of multiple groups, the matching between each patient and the pain evolution laws of different groups is analyzed to generate the matching results of the patient's pain evolution laws, and each patient is assigned to the group pain evolution law with the highest matching degree. Finally, based on multiple group pain evolution laws, the time evolution laws of multiple patients based on their pain status can be classified, and multiple global pain abnormality association patterns can be generated. Each pattern contains multiple patients who conform to the pain evolution law corresponding to the pattern. These patterns reflect the evolution trend and influencing factors of the patient's pain status during postoperative pain, thereby providing data support for the mining of group laws.
[0048] Step S4: Perform pain resistance analysis on multiple patients included in each global pain abnormality association pattern to generate pain resistance parameters for each patient; generate a pain assessment strategy for the global pain abnormality association pattern based on the pain abnormality feature matrix and pain resistance parameters corresponding to multiple patients in the global pain abnormality association pattern; perform pain status analysis on the object to be analyzed based on multiple pain assessment strategies to obtain a pain status analysis result for the object to be analyzed.
[0049] In this embodiment, after completing the assignment of multiple patients to the corresponding group patterns, for multiple patients in the same pattern, their individual differences in the group can be further analyzed, taking into account the individual differences of the patients such as pain sensitivity and tolerance, and finally calculating the pain resistance parameter of each patient, that is, the resistance characteristics to the pain state under relatively similar liver function and physiological data change patterns, that is, the possible perception differences of the same degree of pain in different individuals.
[0050] For each global pain abnormality association pattern, by further mining the pain abnormality feature matrix corresponding to multiple patients, the correlation level between the pain status and liver function and physiological data of patients in the group pattern at different time periods is extracted. At the same time, taking into account the personalized pain resistance of each patient, a pain assessment strategy for the group is finally generated to describe the pain levels that multiple objects in the group may exhibit under different physiological states and liver function levels. Based on the generated multiple pain assessment strategies, real-time pain status analysis of the analyzed objects can be performed. The pain status that the patient may exhibit can be predicted in advance based on the patient's real-time physiological test data and the liver function data collected in stages. It is not limited to the patient's active assessment based on the scale. Even in scenarios where the patient is in confusion or language disorders, the patient's possible severe pain status can be promptly diagnosed and auxiliary treatment can be provided to promptly relieve the patient's possible postoperative pain and avoid affecting the patient's postoperative recovery.
[0051] The above scheme analyzes the dynamic change patterns of postoperative pain in a large number of patients, extracts group rules and generates pain assessment strategies based on the personalized resistance of each patient. It can effectively combine individual differences of patients such as pain resistance with group rules, provide more accurate decision support for clinical pain management, and thus realize personalized and dynamic pain status analysis.
[0052] As an optional real-time process, in step S3 above, a time evolution analysis is performed on the pain abnormality feature matrices of multiple patients to generate multiple local time evolution patterns, and multiple group pain evolution laws are generated based on the multiple local time evolution patterns, including:
[0053] Based on the preset sliding window, the pain abnormality feature matrix of each patient is traversed to generate multiple local window feature vectors, the window period parameters of each local window feature vector are determined, DBSCAN clustering is performed on the multiple local window feature vectors, and the clustering process is temporally constrained based on the window period parameters to generate multiple window feature clusters.
[0054] In this embodiment, the size of the preset sliding window is determined by the time step and feature dimension, where the feature dimension specifically refers to the dimension of each local pain abnormality feature vector in the pain abnormality feature matrix. The time step can be based on a specific short-term postoperative monitoring period. For example, if the 24 hours after surgery are the key observation period, then each observation period can be 6 hours, and the time step within the preset sliding window is 6. After determining the size of the preset sliding window based on actual monitoring needs, each window may contain one or more pain abnormality periods. In this case, statistical features such as the mean, slope, and variance of the local features within multiple pain periods can be taken to form a local window feature vector within the window. Finally, after traversing the patient's pain abnormality feature matrix using the preset sliding window to generate multiple local window feature vectors, the window period parameter of each local window feature vector is further determined. The window period parameter is used to indicate the temporal position of the local window feature vector. For example, a series of parameters such as 1, 2, and 3 are used to determine the window period parameters of different local window feature vectors according to the temporal sequence. DBSCAN clustering is then performed on the multiple local window feature vectors. At the same time, in order to explore the temporal evolution of the patient's pain state, the clustering process is temporally constrained based on the window period parameter.
[0055] In the process of timing constraints on the clustering process based on the window period parameters, the timing constraint distance between any two local window feature vectors is calculated based on the window period parameters. DBSCAN clustering is performed on multiple local window feature vectors based on the timing constraint distance. The timing constraint distance is calculated as follows:
[0056] L(i,j)=α·L f (i,j)+(1-α)·L t (i,j)
[0057] Where L(i,j) is the temporal constraint distance between the local window feature vector i and the local window feature vector j, L f (i, j) is the spatial distance parameter between the local window feature vector i and the local window feature vector j, which is calculated based on the local window feature vector i and the local window feature vector j. For example, the spatial distance is measured by Euclidean distance, L t (i, j) is the temporal distance parameter between local window eigenvector i and local window eigenvector j. It is calculated based on the window period parameters corresponding to local window eigenvector i and local window eigenvector j, respectively. For example, the difference between the window period parameters is used as the temporal distance parameter. Specifically, the temporal distance parameter between the eigenvectors in the first and third sliding windows is 3 and 1, which is 2. A larger temporal distance parameter indicates a lower temporal correlation between the two eigenvectors. α is the spatiotemporal adjustment weight, which is used to balance the influence of spatial and temporal distances. By properly setting the spatiotemporal adjustment weight, you can select a preference for spatial and temporal distances.
[0058] In the above formula, the closer the spatial and temporal distances between the eigenvectors are, the easier it is for them to be classified into the same high-density area during the density clustering process, so as to better consider the association between adjacent time periods. For eigenvectors with a longer temporal distance, considering that the distance of the time period evolution is too far and the reference is low, the temporal distance between the eigenvectors is penalized to avoid clustering the eigenvectors separated by multiple time periods into the same group due to the similarity of the spatial distance. In this way, the evolution law of the pain state between adjacent time periods can be efficiently captured.
[0059] Based on the preset sliding window, multiple recovery periods corresponding to the patient are determined, the period coverage of each window feature cluster is counted and multiple pattern transfer parameters of each window feature cluster are calculated, and multiple local period evolution patterns are generated according to the pattern transfer parameters. The multiple local period evolution patterns are screened for evolutionary importance to determine multiple target period evolution patterns, and the group pain evolution law of each target period evolution pattern is generated.
[0060] In this embodiment, after completing the clustering of the window features, multiple recovery time periods corresponding to the patient's postoperative recovery period are determined based on a preset sliding window. For example, every 6 hours in the early postoperative period is considered a recovery time period. For the multiple window feature clusters obtained by clustering, the time period coverage rate is statistically analyzed for each window feature cluster. This includes determining the multiple patients involved in the window feature cluster, counting the number of these patients distributed in each recovery time period, and calculating the ratio of the number of patients distributed in each recovery time period to the total number of patients involved in the window feature cluster, which serves as the time period coverage rate of the window feature cluster in different recovery time periods.
[0061] Then, based on the time period coverage of the window feature cluster in different recovery periods, multiple mode transfer parameters of the window feature cluster are calculated.
[0062] In this embodiment, the mode transfer parameter is calculated as follows:
[0063]
[0064] Where, P ab M represents the mode transfer parameter of the window feature cluster a in the recovery period t being converted to the window feature cluster b in the recovery period t+1. ta represents the time period coverage of window feature cluster a in the recovery period t, M (t+1)bIt represents the proportion of patients who belong to window feature cluster a in recovery period t and are converted to patients who belong to window feature cluster b in recovery period t+1. The pattern transfer parameter is used to characterize the changes in the window feature cluster to which the patient belongs during the evolution of two adjacent recovery periods. For example, after each patient enters the next recovery period, the window feature cluster to which he belongs may remain unchanged or change to another window feature cluster. By counting the changes in the window feature cluster to which each patient belongs, the temporal evolution law of the patient's local pain state between different recovery periods can be identified, and the local change law of one or more pain states that appear frequently in multiple patients during postoperative recovery can be further revealed, so as to realize the efficient mining of implicit patterns in patient historical data, reveal the multi-stage dynamic path of pain development, and support the evaluation, analysis and early warning of early pain states.
[0065] After using the above formula to calculate the multiple mode transfer parameters between any two adjacent recovery periods and between any two window feature clusters, multiple local period evolution patterns are constructed based on the multiple recovery periods and window feature clusters. Among them, any local period evolution pattern includes a conversion path between multiple window feature clusters, and a period transfer probability between any two adjacent window feature clusters in the conversion path. The period transfer probability is specifically a mode transfer parameter between specific recovery periods and window feature clusters. Each window feature cluster in the conversion path corresponds to a recovery period. Exemplarily, in multiple consecutive recovery periods, each window feature cluster represents a local pain evolution pattern, then the possible conversion path is mode A→mode B→mode D→mode D→mode F, and the directed edge between two adjacent modes in the conversion path is the period transfer probability between the corresponding two window feature clusters.
[0066] For the multiple local period evolution patterns identified, considering that some transition paths are not representative, these local period evolution patterns are further screened for evolutionary importance to determine multiple target period evolution patterns. In this process, for any local period evolution pattern, if the period transition probability between any two adjacent window feature clusters in the transition path of the local period evolution pattern is greater than the preset evolution threshold, then the transition path of the local period evolution pattern is recorded as the target period evolution pattern; otherwise, it is directly discarded. In this way, multiple highly representative transition paths are determined, and thus multiple target period evolution patterns are determined.
[0067] As an optional real-time process, in step 3 above, each patient is matched with multiple group pain evolution patterns to generate pain evolution pattern matching results. Based on the pain evolution pattern matching results, multiple global pain abnormality association patterns are generated, specifically including:
[0068] The core pain feature vector of each window feature cluster is extracted, and a pain evolution feature matrix of the pain evolution law of each group is generated based on multiple core pain feature vectors. The local window feature vectors of the patient in each recovery period are matched and analyzed through the pain evolution feature matrix.
[0069] In this embodiment, the core pain feature vector can be generated based on the local window feature vectors of multiple individual patients in the window feature cluster. For example, the mean level of the local window feature vectors of multiple individual patients is taken to represent the core feature of the window feature cluster. Thus, based on the multiple window feature clusters involved in the conversion path of the group pain evolution law, a pain evolution feature matrix of the group pain evolution law is constructed. Based on the pain evolution feature matrix, the local window feature vectors of the patient in each recovery period are matched and analyzed. In this process, the core pain feature vector corresponding to the pain evolution feature matrix in each recovery period and the local window feature vector of the patient are determined, and the matching parameters between the two vectors are calculated. For example, distance measurement formulas such as Euclidean distance and cosine similarity are selected to calculate the matching parameters. Then, the matching parameters of multiple recovery periods are fused. For example, the average level of the matching parameters of multiple recovery periods is taken to obtain the global matching degree between the patient and the pain evolution feature matrix.
[0070] Finally, a pain evolution pattern matching result is generated for each patient based on the global matching degree between the patient and the multiple group pain evolution patterns. The multiple group pain evolution patterns determined in the previous steps are simultaneously recorded as a global pain abnormality association pattern, and multiple global pain abnormality association patterns are determined. Multiple patients are then classified based on the multiple global pain abnormality association patterns. Specifically, the group pain evolution pattern with the highest global matching degree is determined for each patient based on the patient pain evolution pattern matching results. The global pain abnormality association pattern corresponding to this group pain evolution pattern is then used as the patient's assignment target, thereby completing the assignment of multiple patients to multiple global pain abnormality association patterns. Each global pain abnormality association pattern is derived from the analysis of local pain state evolution patterns and can well characterize the various specific recovery state change patterns exhibited by patients during their recovery process after open liver cancer surgery. This classification method allows for more accurate identification of the patient's pain state change patterns, enabling precise analysis and prediction of the patient's pain state based on the local change patterns of the group, facilitating the provision of personalized pain treatment plans for clinical care.
[0071] As an optional real-time process, in the above step S4, a pain assessment strategy for the global pain abnormality association pattern is generated based on the pain abnormality feature matrix and pain resistance parameters corresponding to multiple patients in the global pain abnormality association pattern, including:
[0072] Based on the pain resistance parameters, local pain resistance analysis was performed on multiple patients under the global pain abnormality association pattern, and multiple patients were divided into multiple resistance category groups.
[0073] It is worth noting that while the aforementioned steps deeply explored the evolutionary patterns of pain states in different patients and achieved classification of patient groups, given the potential differences in pain tolerance among different patients, the group approach more comprehensively considered the correlation between changes in liver function, physiological parameters, and pain assessment data. To further develop a more personalized pain assessment strategy, in addition to considering the patient's pain state and physiological characteristics, the patient's pain tolerance and changes during the recovery period were also considered as key factors. This approach generates an accurate pain assessment strategy based on the individual differences of each patient and combined with group patterns to provide a tailored treatment plan for each patient.
[0074] In this embodiment, local pain resistance analysis is performed on multiple patients under the global pain abnormality association pattern based on the pain resistance parameter, wherein the pain resistance parameter can be specifically calculated based on the difference between the patient's own average level and the average level of multiple individual patients in the group. For example, the ratio between the mean of the patient's subjective pain score and the mean of the overall score of multiple patients in the group is calculated as the pain resistance parameter of each patient in the group, and according to the distribution of the pain resistance parameter, multiple patients are divided into multiple resistance category groups, for example, 0.9 to 1.1 is used as the average level, those below this range are recorded as high resistance groups, and those above this range are recorded as low resistance groups, thereby dividing into multiple resistance category groups.
[0075] Then, based on the abnormal pain feature matrix, multiple characteristic abnormal information of each patient in each resistance category group in each recovery period is extracted, and based on the multiple abnormal information, multiple patient physiological state reference ranges and pain state reference ranges of each patient in each resistance category group in the recovery period are extracted. Among them, the characteristic abnormal information can specifically be the abnormal levels of different parameter items in the recovery period, and is mainly analyzed based on the abnormal pain feature matrix, that is, the degree of difference between the liver function indicators, physiological indicators, and pain scores of patients in one or more abnormal pain periods in the recovery period compared with the normal levels. For example, the difference between the average level and the normal level of the liver function indicators of each patient in the early recovery group in the abnormal pain period is calculated, so that based on these characteristic abnormal information, multiple patient physiological state reference ranges and pain state reference ranges of each patient in the resistance category group in the recovery period are extracted, such as the excess or reduction of different liver function indicators and physiological indicators compared with the normal level, and the state reference ranges corresponding to different parameters and the state reference ranges corresponding to the pain scores are obtained. The state reference range can be reasonably set after calculating the mean and standard deviation based on the differential data of multiple patients. For example, the value range corresponding to twice the standard deviation centered on the mean can be used as the state reference range of the parameter. This can take into account the local pain resistance levels of different patients in the group and obtain the local state analysis strategy for each resistance category group in multiple recovery periods. Finally, a pain assessment strategy for the global pain abnormality association pattern is generated based on multiple local state analysis strategies.
[0076] In the process of analyzing the pain state of the analyzed subject using multiple pain assessment strategies based on global pain abnormality association patterns, the local variation patterns of the patient's early pain state can be extracted to determine its matching degree with the global pain abnormality association pattern. The global pain abnormality association pattern with the highest matching degree is selected. Then, based on the patient's current recovery period, the local state analysis strategy for the corresponding recovery period is extracted from the pain assessment strategy based on the global pain abnormality association pattern. Then, based on the patient's actual physiological state monitoring data and liver function test data, the patient's pain state is preliminarily assessed. When physiological parameters and liver function levels reach the warning level, a timely warning is issued, allowing medical staff to observe and analyze the patient's possible severe pain state in advance and take timely measures to reduce the pain state. The pain assessment strategy can be used to assist medical staff in assessing the patient's pain state at different recovery periods and making corresponding treatment decisions, avoiding the patient's difficulty in subjectively expressing their own pain state due to confusion or language barriers, which affects the patient's postoperative recovery.
[0077] See Figure 2Based on the same inventive concept, an embodiment of the present invention further provides a pain status analysis system for liver cancer patients, the system being used to implement the above-mentioned pain status analysis method for liver cancer patients, the system comprising:
[0078] A data acquisition module is used to collect pain status monitoring data of patients after open liver cancer surgery, including physiological monitoring data, liver function recording data, and pain assessment data;
[0079] A local analysis module is used to determine multiple abnormal pain periods of the patient based on the pain assessment data in the pain status monitoring data, extract the local physiological change characteristics, local liver function characteristics and local pain assessment characteristics of each abnormal pain period, construct a local pain abnormality feature vector for each abnormal pain period, and generate a pain abnormality feature matrix for each patient;
[0080] The pain local evolution analysis module is used to perform time evolution analysis on the pain abnormality feature matrices of multiple patients to generate multiple local time evolution patterns, generate multiple group pain evolution laws based on the multiple local time evolution patterns, match each patient with the multiple group pain evolution laws to generate pain evolution law matching results, and generate multiple global pain abnormality association patterns based on the pain evolution law matching results;
[0081] The pain assessment and analysis module is used to perform pain resistance analysis on multiple patients included in each global pain abnormality association pattern, generate pain resistance parameters for each patient, generate a pain assessment strategy for the global pain abnormality association pattern based on the pain abnormality feature matrix and pain resistance parameters corresponding to multiple patients in the global pain abnormality association pattern, analyze the pain status of the object to be analyzed based on multiple pain assessment strategies, and obtain the pain status analysis results of the object to be analyzed.
[0082] The foregoing description is merely a detailed description of the present invention, which is intended to enable those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art. Portions not described in detail in this specification are well known to those skilled in the art.
Claims
1. A method for analyzing the pain status of patients with liver cancer, characterized in that: include: Collecting pain status monitoring data from patients after open liver cancer surgery, including physiological monitoring data, liver function recording data, and pain assessment data, and identifying multiple periods of abnormal pain in the patients based on the pain assessment data in the pain status monitoring data; Extract local physiological change characteristics, local liver function characteristics, and local pain assessment characteristics during each abnormal pain period, construct a local pain abnormality feature vector for each abnormal pain period, and generate a pain abnormality feature matrix for each patient; Performing time evolution analysis on the abnormal pain feature matrices of multiple patients to generate multiple local time evolution patterns, and generating multiple group pain evolution laws based on the multiple local time evolution patterns, including traversing the abnormal pain feature matrix of each patient based on a preset sliding window to generate multiple local window feature vectors, determining the window time parameter of each local window feature vector, performing DBSCAN clustering on the multiple local window feature vectors, and imposing time constraints on the clustering process based on the window time parameter to generate multiple window feature clusters; The timing constraint of the clustering process based on the window period parameter includes calculating the timing constraint distance between any two local window feature vectors based on the window period parameter, and performing DBSCAN clustering on multiple local window feature vectors according to the timing constraint distance. The timing constraint distance is calculated as follows: ； Where, is the local window feature vector and the local window feature vector The timing constraint distance between is the local window feature vector and the local window feature vector The spatial distance parameter between is the local window feature vector and the local window feature vector The time distance parameter between is the spatiotemporal adjustment weight; Determine multiple recovery periods corresponding to the patient based on a preset sliding window, count the period coverage of each window feature cluster, and calculate multiple pattern transfer parameters for each window feature cluster. Generate multiple local period evolution patterns based on the pattern transfer parameters, screen the multiple local period evolution patterns based on their evolutionary importance to determine multiple target period evolution patterns, and generate a group pain evolution law for each target period evolution pattern. Matching each patient with multiple group pain evolution patterns to generate pain evolution pattern matching results, generating multiple global pain abnormality association patterns based on the pain evolution pattern matching results, performing pain resistance analysis on multiple patients included in each global pain abnormality association pattern, and generating a pain resistance parameter for each patient; A pain assessment strategy for a global pain abnormality association pattern is generated based on the pain abnormality feature matrices and pain resistance parameters corresponding to multiple patients under the global pain abnormality association pattern. The pain state of the object to be analyzed is analyzed based on multiple pain assessment strategies to obtain a pain state analysis result of the object to be analyzed.
2. The method for analyzing the pain status of a liver cancer patient according to claim 1, wherein: Each patient is matched with multiple groups of pain evolution patterns to generate pain evolution pattern matching results. Based on the pain evolution pattern matching results, multiple global pain abnormality association patterns are generated, including: Extract the core pain feature vector of each window feature cluster, generate a pain evolution feature matrix for each group's pain evolution law based on multiple core pain feature vectors, and perform matching analysis on the patient's local window feature vectors in each recovery period using the pain evolution feature matrix. This includes calculating the matching parameters between the core pain feature vector corresponding to the pain evolution feature matrix and the patient's local window feature vector in the same recovery period, and generating a global matching degree between the patient and the pain evolution feature matrix based on the matching parameters across multiple recovery periods. The patient's pain evolution law matching result is generated based on the global matching degree corresponding to the patient and multiple group pain evolution laws, multiple global pain abnormality association patterns are determined based on the multiple group pain evolution laws, and multiple patients are classified based on the multiple global pain abnormality association patterns, including assigning the patient to the global pain abnormality association pattern corresponding to the group pain evolution law with the largest global matching degree, thereby generating multiple global pain abnormality association patterns containing multiple patients.
3. The method for analyzing pain status of liver cancer patients according to claim 1, characterized in that: A pain assessment strategy for generating a global pain abnormality association pattern based on the pain abnormality feature matrix and pain resistance parameters corresponding to multiple patients under the global pain abnormality association pattern includes: Based on the pain resistance parameter, local pain resistance analysis is performed on multiple patients under the global pain abnormality association model, and multiple patients are divided into multiple resistance category groups; Based on the pain abnormality feature matrix, multiple feature abnormality information of patients in each resistance category group in each recovery period is extracted respectively. Based on the multiple abnormal information, multiple physiological state reference ranges and pain state reference ranges of patients in each resistance category group in the recovery period are extracted. A local state analysis strategy for each resistance category group in multiple recovery periods is constructed, and a pain assessment strategy of the global pain abnormality association pattern is generated based on the multiple local state analysis strategies.
4. The method for analyzing pain status of a liver cancer patient according to claim 2, wherein: Based on the mode transfer parameters, multiple local time evolution modes are generated, including: For multiple local time period evolution patterns, they include the conversion paths between multiple window feature clusters and the time period transition probability between any two adjacent window feature clusters in the conversion path. Each window feature cluster in the conversion path corresponds to a recovery period. The time period transition probability between two adjacent window feature clusters in the conversion path is calculated using multiple mode transition parameters of the window feature clusters. The calculation formula is as follows: ； Where, Indicates the recovery period Window feature clusters in During the recovery period Convert to window feature cluster The mode transfer parameter, Represents a window feature cluster During the recovery period The time period coverage in Indicates that during the recovery period Belongs to the window feature cluster Patients in the recovery period Convert to a window feature cluster the proportion of patients with After calculating multiple mode transfer parameters between any two adjacent recovery periods and between any two window feature clusters using the above formula, multiple local period evolution patterns are constructed based on multiple recovery periods and window feature clusters.
5. The method for analyzing pain status of liver cancer patients according to claim 4, characterized in that: Evolution importance screening is performed on multiple local period evolution patterns to determine multiple target period evolution patterns, including: For any local period evolution pattern, if the period transition probability between any two adjacent window feature clusters in the transition path of the local period evolution pattern is greater than the preset evolution threshold, the transition path of the local period evolution pattern is recorded as the target period evolution pattern.
6. A pain status analysis system for liver cancer patients, characterized in that: The system is used to implement the method for analyzing the pain status of a liver cancer patient according to any one of claims 1 to 5, comprising: A data acquisition module is used to collect pain status monitoring data of patients after open liver cancer surgery, including physiological monitoring data, liver function recording data, and pain assessment data; A local analysis module is used to determine multiple abnormal pain periods of the patient based on the pain assessment data in the pain status monitoring data, extract the local physiological change characteristics, local liver function characteristics and local pain assessment characteristics of each abnormal pain period, construct a local pain abnormality feature vector for each abnormal pain period, and generate a pain abnormality feature matrix for each patient; The pain local evolution analysis module is used to perform time evolution analysis on the pain abnormality feature matrices of multiple patients to generate multiple local time evolution patterns, generate multiple group pain evolution laws based on the multiple local time evolution patterns, match each patient with the multiple group pain evolution laws to generate pain evolution law matching results, and generate multiple global pain abnormality association patterns based on the pain evolution law matching results; The pain assessment and analysis module is used to perform pain resistance analysis on multiple patients included in each global pain abnormality association pattern, generate pain resistance parameters for each patient, generate a pain assessment strategy for the global pain abnormality association pattern based on the pain abnormality feature matrix and pain resistance parameters corresponding to multiple patients in the global pain abnormality association pattern, analyze the pain status of the object to be analyzed based on multiple pain assessment strategies, and obtain the pain status analysis results of the object to be analyzed.