Brain tumor surgery patient perioperative period pain assessment system

Abstract

The invention relates to the technical field of medical auxiliary evaluation, in particular to a perioperative pain evaluation system for a brain tumor surgery patient, which comprises a surgery pain data acquisition module for acquiring an HRV signal, an intracranial pressure waveform, a subjective VAS score and pain part data of the patient; the feature extraction module processes the data through abnormal value detection and sliding window filtering, extracts HRV and ICP features in combination with time domain analysis, and outputs standardized data after integrating the HRV and ICP features with subjective feedback features; the AI pain quantification module is used for generating a pain risk probability value based on an AI model constructed by an XGBoost algorithm, triggering three-level early warning and associating pain properties and causes of a patient; the intervention pushing module triggers a multi-terminal prompt and pushes an intervention scheme, and a patient performs graphic interaction and feedback through a mobile phone terminal; and the management and tracing module is used for associating single-time whole-process data, recording an intervention effect and automatically upgrading an unexpected scheme. Therefore, the problems that in the prior art, intracranial pressure and pain coupling is not quantified, intervention lags and the like are solved.

Description

Technical Field

[0001] This invention relates to the field of medical auxiliary assessment technology, specifically a perioperative pain assessment system for patients undergoing brain tumor surgery. Background Technology

[0002] In neurosurgical clinical practice, perioperative pain management for brain tumor surgery patients is always a crucial aspect of ensuring patient safety and promoting postoperative recovery. However, current clinical pain assessment and intervention models have many limitations and cannot meet the needs of precise and dynamic management. Existing pain assessment systems rely heavily on patient subjective reports, with the Visual Analogue Scale (VAS) as the core assessment method. This is easily affected by the patient's cognitive state, emotional fluctuations, and sedative medications, leading to biased assessment results. For example, some patients may be unable to accurately report their pain levels due to postoperative confusion or limited verbal expression; anxiety, fear, and other emotional factors may cause patients to overestimate or underestimate pain intensity, resulting in clinical misjudgment. Furthermore, traditional assessments are often intermittent, with medical staff typically manually recording VAS scores every 4-6 hours. This fails to capture the instantaneous changes and fluctuations in pain, making real-time monitoring difficult and missing opportunities for early intervention.

[0003] More importantly, postoperative pain following brain tumor surgery has a unique neuropathological mechanism. There is a significant "intracranial pressure-pain" coupling effect between intracranial pressure fluctuations and pain perception, but current assessment systems lack quantitative monitoring of this core link. Clinically, elevated intracranial pressure can directly trigger pain-related symptoms such as headache and nausea. However, current monitoring methods mostly focus on single physiological indicators, such as monitoring intracranial pressure only through invasive probes, or collecting routine vital signs like heart rate and blood pressure alone. They fail to combine neurologically specific indicators such as intracranial pressure waveforms and heart rate variability (HRV) with subjective pain scores, making it impossible to construct a multi-dimensional pain assessment model. Furthermore, due to the lack of an AI-based individualized early warning mechanism, medical staff struggle to predict the risk of pain based on individual patient differences. Intervention is often only implemented after moderate to severe pain has occurred, which not only increases patient suffering but may also lead to further increases in intracranial pressure due to pain stress, raising the risk of postoperative complications and prolonging hospital stays and recovery periods. Summary of the Invention

[0004] This application provides a perioperative pain assessment system for patients undergoing brain tumor surgery to address the problems of high pain assessment error, lack of quantification of the coupling between intracranial pressure and pain, and delayed intervention in the prior art.

[0005] The first aspect of this application provides a perioperative pain assessment system for patients undergoing brain tumor surgery, comprising: a pain data acquisition module, a feature extraction module, an AI pain quantification module, an intervention push module, and a management and traceability module; wherein, the pain data acquisition module is used to collect the patient's HRV signal, intracranial pressure waveform, subjective VAS score, and pain location data; the feature extraction module is used to process the HRV signal, intracranial pressure waveform, subjective VAS score, and pain location data through an outlier detection algorithm and a sliding window filtering algorithm, and then extract HRV features, ICP features, and subjective feedback features by combining a time-domain analysis algorithm. After integration, standardized data is output. The AI ​​pain quantification module inputs the standardized data into an AI model built on the XGBoost algorithm to generate a pain risk probability value. Based on the probability value, a three-level warning is triggered, and the nature and cause of the patient's pain are correlated. The intervention push module triggers multi-terminal prompts based on the warning level output by the AI ​​model, and pushes drug or non-drug intervention plans in combination with the nature and cause of pain. At the same time, patients can select plans and provide feedback through graphical interaction on their mobile phones. The management and traceability module is used to correlate the entire process data of a single pain assessment, record the intervention effect, and automatically upgrade plans that do not meet expectations.

[0006] Preferably, the surgical pain data acquisition module includes an HRV signal acquisition unit, an intracranial pressure waveform acquisition unit, a subjective VAS score acquisition unit, and a pain location acquisition unit. The HRV signal acquisition unit continuously acquires HRV signals with a sampling interval of less than or equal to 1 minute. The intracranial pressure waveform acquisition unit acquires real-time intracranial pressure waveforms. The subjective VAS score acquisition unit acquires the patient's subjective score, obtains the VAS score, and simultaneously triggers a photo archive to prevent data falsification. The pain location acquisition unit provides the patient with a 3D head anatomy model for interactive use on their mobile phone, allowing the patient to select the pain area via touchscreen. If the patient is unable to operate the device, a nurse can describe the pain via voice, and the pain location can be automatically labeled using NLP, while simultaneously recording the duration of the pain.

[0007] Preferably, the feature extraction module includes a data cleaning unit, a multi-feature extraction unit, and a standardization output unit. The data cleaning unit is used to remove interference values ​​from the HRV signal and the real-time intracranial pressure waveform using an outlier detection algorithm, and to smooth the VAS score fluctuations using a sliding window filtering algorithm. Simultaneously, it verifies the integrity of the pain location data. The multi-feature extraction unit is used to extract the SDNN and RMSSD features of the HRV signal and the average ICP and amplitude variation coefficient features of the intracranial pressure using a time-domain analysis algorithm, and to convert the VAS score, pain location, and duration data into standardized subjective feedback features. The standardization output unit is used to integrate the SDNN, RMSSD, average ICP, amplitude variation coefficient features, and subjective feedback features into a unified format data, forming standardized data, and then transmit it to the AI ​​pain quantification module.

[0008] Preferably, the AI ​​pain quantification module includes a model calculation unit, a three-level early warning determination unit, and a pain etiology association unit. The model calculation unit inputs the standardized data into an AI model constructed using the XGBoost algorithm to generate a pain risk probability value in the range of 0 to 1. The three-level early warning determination unit triggers a three-level early warning based on the probability value, obtains the risk level, and generates a report on the reason for the early warning trigger. The pain etiology association unit, based on the risk level, combines the pain nature information and pain location from the subjective feedback features, matches a preset rule base, and outputs an etiology label.

[0009] Preferably, the intervention push module includes a multi-terminal prompting unit, a plan matching push unit, and an interactive feedback unit. The multi-terminal prompting unit is used to trigger early warning prompts on the nurse station screen, the patient's mobile phone, and the medical staff's mobile terminal. The plan matching push unit is used to call the corresponding plan from the database according to the risk level and etiology label. The interactive feedback unit is used to provide a graphical interactive interface for the patient, receive the patient's active feedback, and transmit it to the management and traceability module.

[0010] Preferably, the management and traceability module includes a full-process data association unit, an effect evaluation and upgrade unit, a dataset generation and push unit, and a data interaction unit. The full-process data association unit connects data from all stages and stores it in a structured manner along a timeline. The effect evaluation and upgrade unit determines an effective intervention based on a VAS decrease of 30% or more after intervention; if the VAS decrease is less than 30%, the intervention is automatically upgraded. The dataset generation and push unit filters high-quality data with complete feedback weekly, generates a training set, and pushes it to the AI ​​pain quantification module for incremental model optimization. The data interaction unit provides multi-condition data query functionality and bidirectionally interfaces with the hospital's existing system to read basic patient information.

[0011] The second aspect of this application provides a method for perioperative pain assessment in patients undergoing brain tumor surgery, comprising: collecting patient HRV signals, intracranial pressure waveforms, subjective VAS scores, and pain location data; removing interference values ​​from the HRV signals and real-time intracranial pressure waveforms using an outlier detection algorithm, extracting HRV features and ICP features using a time-domain analysis algorithm, smoothing the fluctuations in the VAS scores using a sliding window filtering algorithm, integrating the VAS scores with the pain location data to form subjective feedback features, performing standardized transformation, and generating a feature dataset; inputting the feature dataset into an AI model constructed using the XGBoost algorithm to output pain risk probability values ​​and generate risk levels, combining the pain nature information and pain location of the subjective feedback features with a preset rule base to generate etiology labels; based on the risk level and etiology labels, calling corresponding intervention plans from a preset plan library, automatically pushing them to medical staff mobile terminals and patient terminals for intervention, and simultaneously receiving patient feedback data; if the VAS score decreases by more than or equal to 30%, the plan is deemed effective; if several expected results are not met, the plan is automatically upgraded.

[0012] A third aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the program to implement a perioperative pain assessment method for brain tumor surgery patients as described in the above embodiments.

[0013] A fourth aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement a method for perioperative pain assessment in patients undergoing brain tumor surgery as described in the above embodiments.

[0014] A fifth aspect of this application provides a computer program product, including a computer program or instructions, for implementing a method for perioperative pain assessment in patients undergoing brain tumor surgery as described in the above embodiments.

[0015] Therefore, this application has the following beneficial effects:

[0016] This application embodiment continuously acquires HRV signals and captures real-time intracranial pressure waveforms through a surgical pain data acquisition module, simultaneously collecting subjective VAS scores. Combined with a 3D head model and NLP technology, it annotates pain areas and durations, comprehensively collecting multi-dimensional data including objective physiological data, subjective feelings, and location information, thus improving the data integrity of perioperative pain assessment for brain tumor patients. Using a feature extraction module, outlier detection and sliding window filtering algorithms are employed to remove data interference and smooth fluctuations. A time-domain analysis algorithm extracts core features of HRV and intracranial pressure and transforms them into subjective feedback features, integrating and outputting standardized data in a unified format to improve data quality and consistency. The AI ​​pain quantification module inputs the standardized data into a dataset constructed using the XGBoost algorithm. The AI ​​model generates pain risk probability values, triggers tiered early warnings based on these probabilities, and outputs etiology labels by combining subjective feature matching rule bases, improving the objectivity of pain risk assessment and the accuracy of etiology judgment. Through the intervention push module, early warning prompts are triggered on nurse station screens, patient mobile phones, and medical staff mobile terminals. Corresponding intervention plans are invoked based on risk level and etiology, and a graphical interface is provided to receive patient feedback, enhancing the timeliness of pain intervention and the efficiency of doctor-patient interaction. Through the management and traceability module, the entire closed-loop data is stored in a structured timeline, automatically determining the effectiveness of interventions and upgrading substandard plans. High-quality training sets are regularly pushed to optimize the AI ​​model, and multi-condition data queries are performed to connect with the hospital system, ensuring the system's continuous iteration capability and clinical compliance. This solves the problems of high pain assessment error, unquantified coupling of intracranial pressure and pain, and delayed intervention in existing technologies.

[0017] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0018] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0019] Figure 1 This is a schematic diagram of the perioperative pain assessment system for brain tumor surgery patients provided according to an embodiment of this application;

[0020] Figure 2 This is a schematic diagram of a surgical pain data acquisition module according to an embodiment of this application;

[0021] Figure 3 This is a schematic diagram of a feature extraction module provided according to an embodiment of this application;

[0022] Figure 4 This is a schematic diagram of an AI pain quantification module provided according to an embodiment of this application;

[0023] Figure 5 This is a schematic diagram of an intervention push module provided according to an embodiment of this application;

[0024] Figure 6 This is a schematic diagram of a management and traceability module provided according to an embodiment of this application;

[0025] Figure 7 This is a schematic diagram of a perioperative pain assessment system for brain tumor surgery patients according to an embodiment of this application;

[0026] Figure 8 This is a flowchart illustrating a method for assessing perioperative pain in patients undergoing brain tumor surgery, according to an embodiment of this application.

[0027] Figure 9 This is a schematic diagram of a method for assessing perioperative pain in patients undergoing brain tumor surgery according to an embodiment of this application;

[0028] Figure 10 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of this application. Detailed Implementation

[0029] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0030] The following description, with reference to the accompanying drawings, illustrates a perioperative pain assessment system for brain tumor surgery patients according to an embodiment of this application. Addressing the issue of high pain assessment error mentioned in the background art, this application provides a perioperative pain assessment system for brain tumor surgery patients. In this system, a pain data acquisition module continuously acquires HRV signals and captures real-time intracranial pressure waveforms, simultaneously acquiring subjective VAS scores. Combined with a 3D head model and NLP technology, pain areas and durations are labeled, comprehensively collecting multi-dimensional data including objective physiological data, subjective feelings, and location information, thus improving the data integrity of perioperative pain assessment for brain tumor patients. A feature extraction module uses outlier detection and sliding window filtering algorithms to remove data interference and smooth fluctuations. A time-domain analysis algorithm extracts core features of HRV and intracranial pressure and transforms them into subjective feedback features, integrating and outputting standardized data in a unified format to improve data quality and consistency. AI pain measurement is also used in this system. The system employs a modular approach, which inputs standardized data into an AI model built using the XGBoost algorithm to generate pain risk probability values. Based on these probabilities, it triggers tiered warnings and outputs etiology labels by combining subjective feature matching rule bases, improving the objectivity of pain risk assessment and the accuracy of etiology identification. An intervention push module triggers warnings on nurses' station screens, patients' mobile phones, and medical staff's mobile devices, invoking corresponding intervention plans based on risk level and etiology. A graphical interface is provided to receive patient feedback, enhancing the timeliness of pain intervention and the efficiency of doctor-patient interaction. A management and traceability module stores the entire closed-loop data in a structured timeline, automatically determining the effectiveness of interventions and upgrading substandard plans. It regularly pushes high-quality training sets to optimize the AI ​​model and performs multi-condition data queries, integrating with hospital systems to ensure continuous system iteration and clinical compliance. This addresses the problems of high pain assessment errors, unquantified coupling of intracranial pressure and pain, and delayed interventions in existing technologies.

[0031] Figure 1 This is a schematic diagram of the structure of a perioperative pain assessment system for brain tumor surgery patients provided in an embodiment of this application.

[0032] This application provides a perioperative pain assessment system for patients undergoing brain tumor surgery. The system 10 includes:

[0033] The system includes a surgical pain data acquisition module (100), a feature extraction module (200), an AI pain quantification module (300), an intervention push module (400), and a management and traceability module (500).

[0034] The system includes the following modules: The surgical pain data acquisition module 100 collects patient HRV signals, intracranial pressure waveforms, subjective VAS scores, and pain location data; the feature extraction module 200 processes the HRV signals, intracranial pressure waveforms, subjective VAS scores, and pain location data using outlier detection and sliding window filtering algorithms, then extracts HRV and ICP features using time-domain analysis algorithms, integrates these features with subjective feedback features, and outputs standardized data; the AI ​​pain quantification module 300 inputs the standardized data into an AI model built on the XGBoost algorithm to generate pain risk probability values, triggering a three-level warning based on these probability values, and simultaneously associating the nature and cause of the patient's pain; the intervention push module 400 triggers multi-terminal prompts based on the warning levels output by the AI ​​model, pushing drug or non-drug intervention plans based on the nature and cause of the pain; patients can also select and provide feedback through graphical interaction on their mobile phones; and the management and traceability module 500 associates the entire process data of a single pain assessment, records intervention effects, and automatically upgrades plans that do not meet expectations.

[0035] It is understood that the embodiments of this application continuously acquire HRV signals and capture real-time intracranial pressure waveforms through the surgical pain data acquisition module, simultaneously collect subjective VAS scores, and combine 3D head models and NLP technology to label pain areas and durations, thereby comprehensively collecting multi-dimensional data on physiological objective, subjective feelings, and location information, improving the data integrity of perioperative pain assessment for brain tumor patients; with the help of the feature extraction module, outlier detection and sliding window filtering algorithms are used to remove data interference and smooth fluctuations, and the core features of HRV and intracranial pressure are extracted through time-domain analysis algorithms and transformed into subjective feedback features, integrating and outputting standardized data in a unified format to improve data quality and consistency; through the AI ​​pain quantification module, the standardized data is input into the XGBoost algorithm. The AI ​​model constructed using this method generates pain risk probability values, triggers graded early warnings based on probability, and outputs etiology labels by combining subjective feature matching rule bases, improving the objectivity of pain risk assessment and the accuracy of etiology judgment. Through an intervention push module, early warning prompts are triggered on nurse station screens, patient mobile phones, and medical staff mobile terminals. Corresponding intervention plans are invoked based on risk level and etiology, and a graphical interface is provided to receive patient feedback, enhancing the timeliness of pain intervention and the efficiency of doctor-patient interaction. Through a management and traceability module, the entire closed-loop data is stored in a structured manner along a timeline, automatically determining the effectiveness of interventions and upgrading substandard plans. High-quality training sets are regularly pushed to optimize the AI ​​model, and multi-condition data queries are performed to connect with the hospital system, ensuring the system's continuous iteration capability and clinical compliance. This solves the problems of high pain assessment error, unquantified coupling of intracranial pressure and pain, and delayed intervention in existing technologies.

[0036] In this embodiment of the application, the surgical pain data acquisition module 100 further includes: Figure 2As shown, there are HRV signal acquisition unit, intracranial pressure waveform acquisition unit, subjective VAS score acquisition unit, and pain location acquisition unit.

[0037] The system includes: an HRV signal acquisition unit for continuous HRV signal acquisition with a sampling interval of 1 minute or less; an intracranial pressure waveform acquisition unit for real-time intracranial pressure waveform acquisition; a subjective VAS score acquisition unit for acquiring patient subjective scores and obtaining VAS scores, with simultaneous triggering of photo archiving to prevent data falsification; and a pain location acquisition unit for providing patients with interactive 3D head anatomy model graphics on their mobile phones, allowing patients to select pain areas via touchscreen. If the patient is unable to operate the system, a nurse can describe the pain via voice, and the pain location can be automatically labeled using NLP, while simultaneously recording the duration of the pain.

[0038] It should be noted that NLP refers to Natural Language Processing, which is used to understand, analyze, process, and generate natural language used by humans in daily life, enabling smooth language interaction between humans and machines.

[0039] It is understood that the HRV signal acquisition unit in this application embodiment, by continuously acquiring HRV signals with a sampling interval of less than or equal to 1 minute, can capture the dynamic changes in the patient's heart rate variability in real time, providing high-time data support for subsequent analysis of pain-related physiological characteristics; the intracranial pressure waveform acquisition unit, by acquiring real-time intracranial pressure waveforms, can completely retain the fluctuation characteristics of intracranial pressure, avoiding the limitation that a single value cannot reflect the dynamic intracranial response caused by pain; the subjective VAS score acquisition unit obtains VAS scores by acquiring the patient's subjective scores and simultaneously triggers photo archiving, effectively preventing false data entry while retaining the patient's subjective feeling data, and improving the credibility of subjective score data; the pain location acquisition unit provides a 3D head anatomy model on a mobile phone for the patient to select the pain area by touch screen, and when the patient cannot operate, the nurse describes the pain by voice and automatically annotates it through NLP, while recording the duration of pain, which not only achieves accurate positioning and standardized recording of the pain location, but also takes into account the usage needs of patients with different operating abilities, providing clear positioning and time dimension information for pain etiology association analysis.

[0040] For example, in the neurosurgery department of a hospital, a 45-year-old patient who had undergone surgery for a glioma entered the perioperative monitoring phase. The various acquisition units of the system immediately began operating: the HRV signal acquisition unit continuously acquired the patient's HRV signal at a sampling interval of 30 seconds. Within 2 hours post-surgery, the patient's HRV SDNN value decreased from 80ms to 52ms, and the RMSSD value decreased from 45ms to 28ms. These real-time data reflected a decline in the patient's autonomic nervous system regulation ability, suggesting possible pain stress. Simultaneously, the intracranial pressure waveform acquisition unit captured the patient's real-time intracranial pressure waveform. 1.5 hours post-surgery, the waveform showed that the average ICP value increased from the initial 12mmHg to 17mmHg, and the amplitude variation coefficient increased from 5% to 12%, further supporting the possibility of pain-induced intracranial pressure. Fluctuations occurred; at this point, medical staff had the patient score using the subjective VAS score collection unit. The patient gave a VAS score of 8 due to incision and head swelling pain. The system immediately triggered a photo archive, synchronously recording the patient's expression and the scoring interface to avoid disputes during subsequent data tracing. During the pain location collection phase, although the patient was conscious, postoperative limb movement was limited, and the patient could not operate the phone independently. The nurse then described via voice that "the patient reported persistent swelling pain 3 cm to the right of the surgical incision, radiating to the right temporal region." The system automatically labeled the pain location as "the area around the right frontal surgical incision and the right temporal region" using NLP. It also recorded that the pain started 1 hour after surgery and had lasted for 40 minutes. The collection results of these multi-dimensional data provided accurate basis for medical staff to formulate subsequent intervention plans.

[0041] In this embodiment of the application, the feature extraction module 200 includes: Figure 3 As shown, there are a data cleaning unit, a multi-feature extraction unit, and a standardized output unit.

[0042] The data cleaning unit removes interference values ​​from HRV signals and real-time intracranial pressure waveforms using outlier detection algorithms, smooths VAS score fluctuations using sliding window filtering algorithms, and verifies the integrity of pain location data. The multi-feature extraction unit extracts SDNN and RMSSD features of HRV signals and average ICP and amplitude variation coefficient features of intracranial pressure using time-domain analysis algorithms, and converts VAS scores, pain locations, and duration data into standardized subjective feedback features. The standardization output unit integrates SDNN, RMSSD features, average ICP, amplitude variation coefficient features, and subjective feedback features into a unified format data, forms standardized data, and transmits it to the AI ​​pain quantification module.

[0043] It should be noted that time-domain analysis algorithms refer to a method for extracting feature parameters by calculating the statistical properties of time-series signals in the time dimension. The formula is as follows:

[0044]

[0045]

[0046] Wherein, SDN is the overall fluctuation index of heart rate variability; N is the total number of sinus beats within the analysis time window; RR i The interval of the i-th sinus beat; RMSSD is the average of N sinus intervals; RMSSD is an indicator of short-term heart rate variability; RR i+1 The interval between the i-th sinus beat and the next sinus beat; The average value of intracranial pressure is given; M is the total number of intracranial pressure sampling points; ICP j Let be the instantaneous value of intracranial pressure at the j-th sampling point; CV ICP The relative 79BB dispersion of intracranial pressure wave amplitude; SD ICP The standard deviation of intracranial pressure sampling values ​​within the time window; This represents the average intracranial pressure within the time window.

[0047] Understandably, the data cleaning unit in this embodiment removes interference values ​​from HRV signals and real-time intracranial pressure waveforms using an outlier detection algorithm, and smooths VAS score fluctuations using a sliding window filtering algorithm. Simultaneously, it verifies the integrity of pain site data, effectively removing data noise, correcting abnormal fluctuations, and filling information gaps, ensuring the accuracy and integrity of the original data. The multi-feature extraction unit extracts SDNN and RMSSD features from HRV signals using a time-domain analysis algorithm, and extracts average ICP and amplitude variation coefficient features from intracranial pressure waveforms. It also transforms VAS scores, pain sites, and duration data into standardized subjective feedback features, completing the transformation from "raw data" to "effective features," providing analyzable key information for the AI ​​model. The standardized output unit integrates the above multi-dimensional features into a unified format and transmits it to the AI ​​pain quantification module, avoiding model calculation errors caused by inconsistent feature formats, ensuring the smoothness and efficiency of the AI ​​analysis process, and laying a data foundation for the accuracy of subsequent pain risk assessment.

[0048] For example, when a neurosurgery department in a hospital was conducting perioperative pain monitoring on a patient who had undergone surgery for brain metastases, the system's feature extraction module was the first to activate: The data cleaning unit, while processing the patient's HRV signal one hour post-surgery, detected two abnormal values ​​that spiked to 150ms due to loose monitor leads. These abnormal values ​​were immediately removed using an outlier detection algorithm. Simultaneously, while processing the patient's VAS score, short-term fluctuations of 5, 2, and 6 points were observed due to emotional stress. These were smoothed to a stable value of 4.5 using a sliding window filtering algorithm. The system also simultaneously detected unlabeled pain location data and automatically reminded the nurse to record it. The multi-feature extraction unit then processed the cleaned data. The system extracts features with SDNN of 58ms and RMSSD of 32ms from HRV signals using time-domain analysis algorithms, and features with average ICP of 16mmHg and amplitude variation coefficient of 11% from intracranial pressure waveforms. Simultaneously, the patient's supplementary subjective information, including "left temporal pain, lasting 25 minutes, VAS 4.5," is transformed into standardized subjective feedback features containing location coding, time values, and rating levels. The standardized output unit then integrates these features into data with uniform precision and coding format, transmitting it in real-time to the AI ​​pain quantification module. This provides high-quality, directly analyzable standardized data support for subsequent pain risk assessment.

[0049] In this embodiment of the application, the AI ​​pain quantification module 300 includes: Figure 4 As shown, the model calculation unit, the three-level early warning judgment unit, and the pain etiology association unit are included.

[0050] The model operation unit is used to input standardized data into the AI ​​model built by the XGBoost algorithm to generate a pain risk probability value in the range of 0 to 1; the three-level warning judgment unit triggers a three-level warning based on the probability value, obtains the risk level, and generates a warning triggering reason report; the pain etiology association unit, based on the risk level, combines the pain nature information and pain location in the subjective feedback features, matches the preset rule base, and outputs etiology labels.

[0051] It should be noted that the XGBoost algorithm is a machine learning algorithm based on the principle of gradient boosting decision trees. It improves prediction accuracy by integrating multiple weak decision trees and optimizing model complexity. The formula is:

[0052]

[0053] in, x is the overall objective function of the model; i For the i-th patient, the data is standardized; n is the total number of historical perioperative brain tumor patients participating in model training; y i Let f be the true pain label for the i-th patient; t be the total number of decision trees in the model; f k (xi ) represents the original output of the k-th decision tree for the i-th patient; Ω(f k ) represents the regularization term for the k-th decision tree; Let be the final pain risk probability value for the i-th patient; σ be the activation function; and e be the natural constant.

[0054] The three warning levels are as follows: Red warning, with a pain risk probability value greater than or equal to 0.7, corresponding to high pain risk; Yellow warning, with a pain risk probability value less than 0.7 but greater than 0.4, corresponding to medium pain risk; and Green warning, with a pain risk probability value less than or equal to 0.4, corresponding to low pain risk.

[0055] It is understood that the model operation unit in this application generates a pain risk probability value in the range of 0 to 1 by inputting standardized data into the AI ​​model constructed by the XGBoost algorithm. This can transform multi-dimensional physiological and subjective data into intuitive risk quantification results, providing accurate numerical basis for subsequent early warning judgment. The three-level early warning judgment unit triggers a three-level early warning based on this probability value and obtains the risk level. At the same time, it generates an early warning triggering reason report. This allows medical staff to quickly identify the degree of pain risk of patients through graded early warning, and clarifies the key data for early warning triggering through the reason report, reducing clinical judgment time. The pain etiology association unit, based on the risk level, combines the pain nature information and pain location in the subjective feedback characteristics, matches the preset rule base to output etiology labels, and can associate the abstract risk level with specific etiology. This avoids the problem that medical staff only know the risk but not the etiology, and provides clear etiology guidance for the formulation of subsequent targeted intervention plans.

[0056] For example, when a neurosurgery department in a hospital was monitoring perioperative pain in a 55-year-old patient who had undergone surgery for a pituitary tumor, the system's AI pain quantification module was activated: the model processing unit received standardized data transmitted from the feature extraction module—HRV features (HF / LF = 0.8, SDNN = 52ms), ICP features (mean ICP = 18 mmHg, amplitude variation coefficient = 13%), and subjective feedback features (VAS = 7 points, pain type as dull pain, location as right frontal lobe). This data was then input into the AI ​​model constructed using the XGBoost algorithm, ultimately generating a pain risk probability value of 0.82. The three-level early warning judgment unit then used this probability value as a basis for further analysis. A red alert is triggered when the threshold of 0.82 is greater than 0.7, indicating a high risk level of pain. A report on the triggering cause of the alert is generated, clearly indicating that "abnormally low HRV index, elevated ICP value, and VAS score reaching the moderate to severe standard jointly trigger a high-risk alert." The pain etiology association unit combines the high-risk level with the nature of "dull pain" and the location of "right frontal pain" in the subjective feedback, and matches the association rule "dull pain + increased ICP amplitude → intracranial pressure fluctuation pain" in the preset rule base. Finally, the etiology label "intracranial pressure fluctuation causes postoperative pain" is output, providing a clear basis for medical staff to formulate an intervention plan of "mannitol to reduce intracranial pressure + intravenous analgesia".

[0057] In this embodiment of the application, the intervention push module 400 includes, as follows: Figure 5 As shown, there are multi-terminal prompt unit, scheme matching and push unit, and interactive feedback unit.

[0058] The multi-terminal notification unit is used to trigger early warning notifications on the nurse station screen, patient mobile phones, and medical staff mobile terminals; the solution matching and push unit is used to call the corresponding solution from the database according to the risk level and etiology label; the interactive feedback unit is used to provide patients with a graphical interactive interface, receive patients' proactive feedback, and transmit it to the management and traceability module.

[0059] It should be noted that the multi-terminal notification unit is used to trigger warning prompts on the nurse station screen, patient mobile phones, and medical staff mobile terminals. Specifically, for red warnings, the nurse station terminal will display a red pop-up window, a continuous flashing beep, and an emergency analgesia list, while the patient terminal will display a red pop-up window, vibration, and temporary suggestions. For yellow warnings, the nurse station terminal will display a yellow semi-transparent pop-up window, a single beep, and a list of key observation points, while the patient terminal will display a yellow notification and guidance on non-pharmacological interventions. For green warnings, the nurse station terminal will display a green icon and routine observation records, while the patient terminal will display a green notification, self-management guidance, and an entry point to a pain diary.

[0060] Based on risk level and etiology label, the corresponding solution is retrieved from the database. This means that the system identifies the patient's current pain risk level and etiology label, and directly selects and retrieves suitable drug or non-drug intervention solutions from the intervention solution database based on preset corresponding rules. The preset rules are built based on clinical treatment guidelines, historical effective intervention cases, and pain management experience learned by AI models.

[0061] The intervention protocol library is based on authoritative clinical guidelines and group standards, integrates historical effective cases and data from daily clinical practice scenarios of multiple medical institutions to supplement individualized scenarios, and uses AI models for dynamic iterative optimization. At the same time, it is combined with the three-stage review and regular calibration of the expert committee of neurosurgery, anesthesiology, and pharmacy to ensure that the pushed protocols meet the professional standards of neurosurgical analgesia and are adapted to the individual differences of different patients.

[0062] It is understood that the multi-terminal prompting unit in this application embodiment can ensure that medical staff are aware of the patient's pain risk status in real time by triggering early warning prompts on the nurse station screen, the patient's mobile phone, and the medical staff's mobile terminal, while allowing the patient to understand their own situation, realizing the synchronization of doctor and patient information and improving response efficiency; the plan matching and push unit can call the corresponding plan from the database according to the risk level and etiology label, which can quickly provide targeted intervention suggestions for pain of different risk levels and etiologies, avoiding the blindness of plan formulation and shortening the intervention decision time; the interactive feedback unit provides patients with a graphical interactive interface to receive active feedback and transmit it to the management and traceability module, which can timely capture changes in the patient's feelings after intervention, forming a closed-loop feedback for pain management, providing direct basis for plan adjustment and effect evaluation, and enhancing the accuracy of intervention and patient participation.

[0063] For example, a 48-year-old neurosurgeon in a hospital was diagnosed with a red alert (pain risk probability 0.85) 36 hours post-surgery, with the cause tagged as "intracranial pressure fluctuation pain." At this point, the multi-device notification unit immediately displayed a flashing red alert window on the nurses' station screen (showing the patient's bed number, name, and "high risk - intracranial pressure fluctuation pain" information), simultaneously sending a notification to the patient's mobile phone stating, "Your current pain risk is high; medical staff will intervene as soon as possible." It also sent an alert message containing the patient's real-time ICP data to the attending physician and responsible nurse's mobile devices. The intervention matching and push unit, based on the red alert level and the cause tagged "intracranial pressure fluctuation pain," retrieved the following intervention protocol from the intervention protocol database: "Mannitol 100ml rapid intravenous drip (completed within 30 minutes to avoid extravasation) + head elevation of 30 degrees to maintain a neutral head position (to prevent intraoperative pain)." The combined approach of "combining pressure-induced displacement of dressings with optimization of the ward environment (controlling noise to ≤40 decibels and dimming indoor lights to reduce sensory stimulation)" automatically pushes the details of the plan (including mannitol infusion rate, postural maintenance precautions, and environmental control standards) to the mobile devices of medical staff. The interactive feedback unit provides patients with a graphical interface that displays feedback questions with options such as "Have you received mannitol infusion?", "Has the headache and throbbing pain been relieved after elevating the head of the bed?", and "Is the pain in the current ward environment affected by external interference?". Thirty minutes after the intervention, patients can select "Infusion received, throbbing pain significantly relieved after elevating the head of the bed, pain not aggravated in quiet environment (VAS dropped from 8 to 5)" via touchscreen. This feedback information is transmitted in real time to the management and traceability module, forming a complete data record of this pain intervention, which facilitates subsequent review of the plan's effectiveness by medical staff.

[0064] In this embodiment of the application, the management and traceability module 500 includes, as follows: Figure 6 As shown, the system includes a full-process data association unit, an effect evaluation and upgrade unit, a dataset generation and push unit, and a data interaction unit.

[0065] The system includes a full-process data association unit to connect data from all stages and store it in a structured manner along a timeline; an effectiveness evaluation and upgrade unit to determine an effective intervention if the VAS decreases by 30% or more after the intervention is completed, and to automatically upgrade the intervention if the VAS decreases by less than 30%; a dataset generation and push unit to select high-quality data with complete feedback each week, generate a training set, and push it to the AI ​​pain quantification module for incremental model optimization; and a data interaction unit to provide multi-condition data query functions and to bidirectionally interface with the hospital's existing system to read basic patient information.

[0066] It should be noted that the system provides multi-condition data query functionality and integrates bidirectionally with the hospital's existing systems. Reading basic patient information means using data retrieval algorithms to query corresponding pain management data based on multiple conditions, including patient bed number, surgery type, pain risk level, intervention plan type, and pain location. This allows for rapid retrieval of the target patient's complete pain monitoring and intervention records. Furthermore, it integrates bidirectionally with the hospital's existing hospital information system and EMR electronic medical record system, automatically reading basic information such as patient name, age, surgical history, underlying diseases, and preoperative examination results. This eliminates the need for manual re-entry, reducing data redundancy and operational errors, and ensuring consistency between system data and the hospital's overall diagnostic and treatment data.

[0067] Data retrieval algorithm formula:

[0068]

[0069] Where Score(x) is the total matching score between a pain management data record x to be retrieved and the multiple query conditions entered by the user; M is the total number of query conditions entered by the user; m is the index number of the m-th query condition; w m x is the weight coefficient for the m-th query condition; m To retrieve the field value in record x that corresponds to the m-th query condition; Q m Sim(x) represents the m-th query condition value entered by the user. m Q m ) is x m With Q m The similarity.

[0070] It is understood that the full-process data association unit in this application embodiment, by connecting data from all stages and storing it in a structured manner according to the timeline, can integrate information from various stages such as pain perception, feature extraction, AI prediction, early warning triggering, and intervention execution into a complete and orderly data chain, providing coherent data support for subsequent tracing of the entire pain management process and analysis of intervention effects; the effect evaluation and upgrade unit judges the effectiveness of the plan based on the VAS decline after intervention and automatically upgrades plans that have not met the standards, which can adjust ineffective intervention measures in a timely manner, avoid delays in pain relief due to unsuitable plans, and ensure the timeliness and effectiveness of pain intervention; the dataset generation and push unit selects high-quality data with complete feedback every week to generate training sets and pushes them to the AI ​​pain quantification module, which can continuously provide fresh training data for the AI ​​model built by the XGBoost algorithm, help the model achieve incremental optimization, and continuously improve the accuracy of pain risk prediction; the data interaction unit provides multi-condition data query function, which makes it convenient for medical staff to quickly locate the pain management records of target patients, and at the same time, it connects bidirectionally with the hospital's existing system to read the patient's basic information, reducing data redundancy and operational errors caused by manual repetitive entry, ensuring the consistency of system data with the hospital's overall diagnosis and treatment data, and providing convenience for medical staff to carry out clinical decision-making and data statistics work efficiently.

[0071] For example, when a neurosurgery department of a hospital was managing perioperative pain in a 60-year-old patient who had undergone surgery for brain metastases, the end-to-end data association unit first linked the patient's data from all stages—from the sensory layer recording "dull pain in the right temporal region, VAS score of 8, lasting for 40 minutes, triggered by intracranial pressure fluctuations," to the feature engineering module extracting "HF / LF = 0.6, average ICP = 22 mmHg," and then to the AI ​​model outputting "pain risk probability of 0.83 (red alert)," and the intervention execution "mannitol 100ml IV drip (over 15-30 minutes)." The intervention included monitoring urine output to prevent dehydration and relaxation training (twice daily, 15 minutes each time, using immersive relaxation to distract from pain). This data was then structured and stored chronologically, generating a "PainLoop_ID" record containing complete time points. Thirty minutes after the intervention, the effectiveness assessment upgrade unit compared the patient's VAS score before and after the intervention (from 8 to 6, a decrease of 25%, but less than 30%). If the current plan was deemed ineffective, the plan was automatically upgraded from the intervention strategy library to "Mannitol 100ml IV drip (maintaining intracranial pressure reduction, 8-hour intervals)". The patient's treatment regimen consisted of: "one dose of acetaminophen (0.3g every 8-12 hours, avoiding fasting to reduce gastrointestinal irritation) + music relaxation therapy (selecting soothing piano music, combined with breathing training guidance, 20 minutes each time)." Subsequently, the dataset generation and push unit screened the patient's complete data (including pre-intervention data, two intervention plans, execution results, and two VAS feedbacks) over the weekend, integrated it with other high-quality data of the same type to generate a training set, and pushed it to the AI ​​pain quantification module to provide data support for incremental training of the XGBoost model. During this period, medical staff used the data interaction unit to input a combination of multiple conditions, including "patient bed number + red alert + intracranial pressure fluctuation pain," to quickly retrieve the patient's complete pain management record. Simultaneously, the system bidirectionally connected with the hospital's HIS / EMR system, automatically reading the patient's basic information such as "history of hypertension, preoperative tumor diameter 3cm." Before the intervention plan was invoked, it assisted in individualized adjustments to the plan (e.g., confirming no contraindications to acetaminophen based on the history of hypertension). After the intervention, it assisted in analyzing the correlation between the intervention effect and the patient's individual characteristics, providing a reference for subsequent potential intervention adjustments.

[0072] This application proposes a perioperative pain assessment system for brain tumor surgery patients. The system continuously acquires HRV signals and captures real-time intracranial pressure waveforms through a pain data acquisition module, simultaneously collecting subjective VAS scores. It combines a 3D head model with NLP technology to annotate pain areas and durations, comprehensively collecting multi-dimensional data including objective physiological data, subjective feelings, and location information, thus improving the data integrity of perioperative pain assessment for brain tumor patients. A feature extraction module uses outlier detection and sliding window filtering algorithms to remove data interference and smooth fluctuations. A time-domain analysis algorithm extracts core features of HRV and intracranial pressure and transforms them into subjective feedback features, integrating and outputting standardized data in a unified format to improve data quality and consistency. An AI pain quantification module inputs the standardized data into X... The GBoost algorithm-based AI model generates pain risk probability values, triggers tiered early warnings based on probability, and outputs etiology labels by combining subjective feature matching rule bases, improving the objectivity of pain risk assessment and the accuracy of etiology judgment. Through the intervention push module, early warning prompts are triggered on nurse station screens, patient mobile phones, and medical staff mobile terminals. Corresponding intervention plans are invoked based on risk level and etiology, and a graphical interface is provided to receive patient feedback, enhancing the timeliness of pain intervention and the efficiency of doctor-patient interaction. Through the management and traceability module, the entire closed-loop data is stored in a structured timeline, automatically determining the effectiveness of interventions and upgrading substandard plans. High-quality training sets are regularly pushed to optimize the AI ​​model, and multi-condition data queries are performed to connect with the hospital system, ensuring the system's continuous iteration capability and clinical compliance. This solves the problems of high pain assessment error, unquantified coupling of intracranial pressure and pain, and delayed intervention in existing technologies.

[0073] The following will illustrate a perioperative pain assessment system for brain tumor surgery patients through a specific embodiment, such as... Figure 7 As shown, it includes:

[0074] Patient Li, male, 52 years old, was diagnosed with a right temporal lobe glioma and was scheduled to undergo tumor resection under general anesthesia. Preoperative evaluation showed he met the inclusion criteria for the perioperative pain assessment system for brain tumor surgery patients, and the system was officially activated one day before surgery for continuous monitoring and management. At 9:00 AM the day before surgery, the pain data acquisition module began collecting data: the HRV signal acquisition unit continuously monitored the patient's condition using a three-lead ECG electrode attached to the chest, with a sampling interval of 20 seconds to ensure three sinus interval data points were acquired per minute and transmitted to the system backend in real time. By 10:00 AM, a total of 180 valid RR interval data points had been collected. The intracranial pressure waveform acquisition unit connected to the monitor via a preoperatively implanted intracranial pressure monitoring probe (located in the anterior horn of the right ventricle), with a sampling frequency of 100Hz, i.e., 100 intracranial pressure data points were collected per second, forming a continuous waveform curve. A total of 36,000 data points were recorded within one hour, including data from 9:15 AM to 9:00 AM. At point 20, a 5-second waveform fluctuation occurred due to brief patient activity. The subjective VAS score acquisition unit displayed an electronic scoring interface while the patient was awake. The interface used a 0-10 point slider design. The patient dragged the slider to the 5-point position based on their right temporal region's throbbing pain. The system immediately triggered the front-facing camera to capture the patient's facial expression, which was archived for subsequent data verification. The pain location acquisition unit displayed a 3D head anatomy model, labeled according to cranial anatomical regions. The patient selected the right temporal region (corresponding to the T4 area in anatomy) using a touchscreen and chose "3 days before surgery" as the duration of pain in the pop-up time input box. Simultaneously, the pain nature was entered as "persistent throbbing pain, occasionally worsening" in the remarks column. This raw data was encrypted and uploaded to the system database in real time, laying the foundation for subsequent feature extraction and analysis.

[0075] After receiving the raw data from the surgical pain data acquisition module, the feature extraction module immediately initiates the data processing flow. The data cleaning unit first runs an outlier detection algorithm to screen the HRV signal point by point, calculating the deviation of each RR interval. It finds two data points at 9:08 with Z values ​​of 3.2 and -3.5 (exceeding the ±3 threshold range). Combined with the synchronously recorded electrode contact status, these are determined to be interference values ​​caused by electrode loosening and are removed. The remaining 178 valid data points undergo a normality test, confirming they meet the analysis requirements. Regarding the intracranial pressure waveform data, at 9:18, three consecutive data points (28 mmHg, 30 mmHg, 29 mmHg) exceed the normal fluctuation range (normal range is 12-18 mmHg, IQR = 4 mmHg, upper limit is 18 + 1.5 × 4 = 24 mmHg). Combined with the patient's activity record, this is confirmed to be interference caused by positional changes and is removed. The removed waveforms are then linearly interpolated to ensure data continuity. Subsequently, a sliding window filtering algorithm was used to process the VAS scores. A 10-minute window was set, and the mean within the window was calculated to smooth out score fluctuations caused by patient emotional changes—the original scores fluctuated between 4 and 6 points, but after filtering, they stabilized at 5 points, with the fluctuation range recorded as 1 point. Simultaneously, the system verified the completeness of the pain location data, confirming that the 3D model point coordinates were accurate (error within 0.5cm), and that the duration and nature of the pain were fully described without missing items, thus passing the completeness check. The multi-feature extraction unit then ran a runtime analysis algorithm to calculate features from the cleaned HRV data: SDNN took 58ms (based on 178 valid RR intervals, with a mean of 820ms; the sum of the squares of the differences between each RR interval and the mean was calculated, divided by 177, and then the square root was taken); RMSSD took 22ms (the sum of the squares of the differences between 177 adjacent RR intervals was calculated, divided by 177, and then the square root was taken). Features were extracted from intracranial pressure data: the mean ICP was 15 mmHg (mean of 35,997 valid sampling points); the amplitude variation coefficient was 0.12 (the standard deviation of intracranial pressure data was calculated as 1.8 mmHg, divided by the mean of 15 mmHg). For subjective feedback data, the VAS score of 5 was standardized to 0.5 (5 / 10), the pain location T4 area was converted to the corresponding code "T4", the pain duration of 3 days was converted to 72 hours and standardized to 0.3 (72 / 240, based on the longest record of 240 hours), and the pain nature "persistent distending pain" was coded as "D02". The standardized output unit integrated these features into a uniform array ([58,22,15,0.12,0.5,0.3,"T4","D02"]), added a timestamp (10:30 AM one day before surgery), and then transmitted it to the AI ​​pain quantification module.

[0076] After receiving standardized data, the AI ​​pain quantification module immediately inputs the data into an AI model built on the XGBoost algorithm. This model, trained on 1000 historical cases, contains 150 decision trees, each with a depth of 6 layers. The model analyzes the input features step-by-step, generating a pain risk probability value of 0.82. The three-level warning unit triggers a red warning based on preset threshold rules (probability ≥ 0.7 for red, 0.3-0.7 for yellow, and ≤ 0.3 for green), simultaneously generating a warning trigger reason report detailing the contribution of each feature: abnormal HRV index contributes 35%, intracranial pressure approaching the threshold contributes 25%, and subjective score contributes 40%, resulting in a comprehensive assessment of high pain risk. The pain etiology association unit retrieves the pain nature code "D02" (persistent distending pain) and location code "T4" (right temporal lobe) from the subjective feedback features, and matches them with the entry in the preset rule base - "temporal lobe region + persistent distending pain → may be related to tumor space-occupying effect and mild increase in intracranial pressure". Combined with the basic information that the patient's tumor is located in the right temporal lobe (approximately 4cm × 3cm in size), the etiology label "tumor space-occupying related pain (with intracranial pressure fluctuation)" is output.

[0077] Upon receiving the red alert signal and etiology label, the intervention push module immediately initiates cross-terminal synchronous notifications: a red full-screen pop-up window appears on the nurse station screen, covering other operating interfaces, accompanied by a buzzer every 3 seconds and a flashing red light. The top of the pop-up displays "Bed 21, Mr. Li, Red Alert," the middle displays the patient's real-time HRV and intracranial pressure waveforms, and the bottom lists an emergency intervention checklist, including "Immediately administer 100ml of mannitol intravenously (complete within 30 minutes, monitor urine output and electrolytes)," "Recheck VAS and level of consciousness after 30 minutes," and "Monitor intracranial pressure, pupillary changes, and vital signs every hour." The patient's mobile phone receives a red pop-up window with a vibration alert, stating, "Your current pain risk is high. Medical staff will assess you within 5 minutes. It is recommended to remain supine and avoid strenuous activity." The attending physician and responsible nurse's mobile apps simultaneously receive push notifications containing the alert level, etiology label, and real-time patient data for rapid response. The intervention matching and push unit retrieves the corresponding level-three intervention plan from the intervention plan database based on the red alert level and the etiology tag of "tumor space-occupying lesion-related pain (with intracranial pressure fluctuations)". The first choice for dehydration and intracranial pressure reduction is "mannitol 100ml rapid intravenous drip (once every 8 hours, observe for electrolyte imbalance)". If intracranial pressure is not well controlled, the alternative is "combined with glycerol fructose 250ml intravenous drip (twice daily, alternating with mannitol to reduce the burden on the kidneys)". The analgesic treatment is "oral acetaminophen tablets 0.3g (once every 8-12 hours, taken after meals to reduce gastrointestinal irritation)". Non-pharmacological interventions include "elevating the head of the bed by 30 degrees to maintain a neutral head position (to help reduce intracranial pressure and avoid pressure on the surgical area)", "avoiding excessive flexion or rotation of the head and neck", "playing soothing and natural sound effects (such as ocean waves or birdsong, 20 minutes each time to distract attention from pain)", and "monitoring consciousness, pupil size and vital signs every hour (to promptly identify changes in the condition and prevent analgesia from masking abnormalities)". The system displays the intervention plan in text and image format on the patient's mobile phone interface. After reading it, the patient clicks "Agree to the intervention plan" and checks "Willing to try listening to natural sound effects" in the non-pharmacological intervention options. Ten minutes later, the responsible nurse administers a rapid intravenous infusion of mannitol and records the administration time (10:40 AM, one day before the procedure), drug batch number, and operator information in the system. The interactive feedback unit prompts the patient to assess changes in pain every 15 minutes. At 11:00 AM, the patient reports "The distending pain has lessened" and selects a VAS score of 3 on the interface.

[0078] The end-to-end data association unit of the management and traceability module integrates all patient data in a timeline: Data collection occurs from 9:00-10:00 AM the day before surgery, recording raw HRV, intracranial pressure, VAS score, pain location, etc., and marking abnormal data points and reasons for removal; from 10:00-10:30 AM, the feature extraction stage stores cleaned indicators, time-domain analysis results, and standardized feature arrays, along with algorithm parameter settings; from 10:30-10:35 AM, the AI ​​quantification stage records model input data, risk probability calculation process, warning levels, and etiology label generation logic; from 10:35-11:30 AM, the intervention stage stores multi-terminal prompts, pushed intervention plans, patient feedback, and medication operation details. All data is named using "Patient ID + Timestamp," structured, and stored in a distributed database, supporting timeline-based review. The efficacy evaluation upgrade unit automatically triggered an efficacy evaluation one hour after medication (11:40). The patient's VAS score was 2, and the calculated decrease was 60%, indicating an effective treatment plan. The system marked the intervention as "effective, maintain the current plan," requiring no upgrade. The dataset generation and push unit conducted a quality screening of the case data on Saturday of the same week, confirming that the data chain was complete (no missing data from collection to intervention feedback), outlier handling was reasonable, and intervention effect records were detailed. The data was marked as high-quality and added to the training set. At 22:00 on Sunday, the system automatically pushed the training set to the AI ​​pain quantification module, initiating incremental model optimization. The decision tree weights were adjusted using the new data to improve the recognition accuracy of "tumor-related pain." The data interaction unit allows medical staff to retrieve the case three days post-surgery using a combination of conditions: inputting "Bed 21 + right temporal lobe glioma + red alert," the system calculates the matching degree using a data retrieval algorithm, prioritizing the return of the patient's complete process record, including data at each stage, intervention plan, and effect evaluation results. Simultaneously, through bidirectional integration with the hospital's HIS and EMR electronic medical record systems, it automatically reads the patient's basic information (age 52, weight 70kg, 5-year history of diabetes, preoperative KPS score 80, tumor pathology grade III), avoiding manual entry errors and ensuring data consistency. On the day of surgery and for seven days post-surgery, the system continuously monitors the patient's pain changes according to the same process. On the third post-surgery day, the patient's pain risk decreased to a green alert, and the intervention plan was adjusted to routine analgesia. Throughout the perioperative period, the system enabled dynamic assessment and precise intervention of pain. All data is included in the management and traceability module, providing a complete basis for subsequent clinical research and model optimization.

[0079] In summary, this application's embodiments utilize a multimodal data acquisition and encrypted transmission mechanism to cover the critical perioperative period from 1 day before surgery to 7 days after surgery. This accurately captures fluctuations in patients' pain-related physiological indicators and changes in their subjective feelings. Combined with outlier detection and data cleaning algorithms, it achieves high-quality data input. Based on the AI ​​model constructed using the XGBoost algorithm and a three-level early warning rule, it achieves accurate output of pain risk probability and intelligent correlation between "pain nature and etiology." Combined with multi-terminal synchronous prompts and graded intervention plan pushes, it can flexibly adapt to different pain risk levels and etiological types, quickly respond to high-risk pain events, and precisely adjust intervention measures. Simultaneously, it records the entire lifecycle data chain, improving the accuracy and timeliness of perioperative pain management, reducing the incidence of moderate to severe pain and opioid dosage, shortening the patient recovery period, and providing strong support for subsequent clinical diagnosis and treatment optimization, incremental training of AI models, and standardization of perioperative pain management for brain tumors.

[0080] Next, referring to the accompanying drawings, a method for assessing perioperative pain in patients undergoing brain tumor surgery, based on an embodiment of this application, is described.

[0081] like Figure 8 As shown, this method for assessing perioperative pain in patients undergoing brain tumor surgery includes the following steps:

[0082] In step S101, the patient's HRV signal, intracranial pressure waveform, subjective VAS score, and pain location data are collected.

[0083] It is understood that the embodiments of this application capture patient characteristics under different pain states by collecting patient HRV signals, intracranial pressure waveforms, subjective VAS scores, and pain location data. Under complex conditions such as the influence of sedative drugs and emotional fluctuations, the objective physiological fluctuations related to pain are preserved. When the patient is awake, the subjective intensity and specific location information of pain are accurately obtained, providing high-quality raw data for subsequent feature extraction, AI model calculation, and etiology label matching, thereby improving the adaptability to diverse pain scenarios in the perioperative period and the accuracy of pain assessment.

[0084] In step S102, interference values ​​in the HRV signal and real-time intracranial pressure waveform are removed by an outlier detection algorithm. HRV features and ICP features are extracted by combining a time-domain analysis algorithm. The VAS score fluctuation is smoothed by a sliding window filtering algorithm. The features are integrated with the pain site data into subjective feedback features. After standardization transformation, a feature dataset is generated.

[0085] Among them, outlier detection algorithms refer to algorithms that automatically identify anomalous data points in a dataset that significantly differ from the patterns of most samples and deviate from the normal range by analyzing features such as data distribution, distance, density, or correlation. The formula is:

[0086]

[0087] Where, x i Z represents a single data point to be detected; μ is the overall mean of the class of data; σ is the standard deviation of the data; Z i For the current data point x i The degree of deviation from μ.

[0088] It should be noted that the sliding window filtering algorithm is a signal processing algorithm that uses a fixed-length time window to calculate the average value of the data within the window, thereby smoothing out short-term random fluctuations in discrete data such as VAS scores while preserving their overall trend. The formula is:

[0089]

[0090] in, Vi is the smoothed VAS value at time i; N is the window radius; Vi i+k This represents the score at time i+k in the original VAS score sequence.

[0091] It is understood that the embodiments of this application process outlier detection algorithms and sliding window filtering algorithms to process HRV signals, intracranial pressure waveforms, and subjective VAS scores as key processing steps for subsequent feature extraction and AI model computation. Through the collaborative processing logic of outlier data removal and fluctuation smoothing, the quality of the original data is comprehensively optimized. Under complex conditions such as patient movement, electrode loosening, and emotional fluctuations, interfering data is identified and removed to avoid misleading subsequent feature calculations. When there are slight fluctuations in the original data, the data curve is smoothed by sliding window mean to retain the real trend of pain-related changes, improve the adaptability to diverse data interference scenarios in the perioperative period and the reliability of standardized feature datasets, and ensure the accuracy of AI model output results.

[0092] For example, patient Chen, 54 years old, underwent preoperative monitoring for a craniotomy for a right parietal lobe glioma. The pain data acquisition module collected his HRV signal, intracranial pressure waveform, subjective VAS score, and pain location (right parietal lobe region) data between 9:00 and 10:00 AM. During this period, at 9:05 AM, the patient turned over, resulting in two abnormal RR interval data in the HRV signal (380ms and 1200ms respectively, far exceeding the normal range of 500-1000ms). At 9:18 AM, coughing caused the intracranial pressure waveform to briefly jump to 27 mmHg (normal range 12-18 mmHg), and the VAS score fluctuated frequently between 4 and 6 points due to emotional tension. The system first uses an outlier detection algorithm to remove the two abnormal RR interval values ​​and the 27 mmHg interference value from the real-time intracranial pressure waveform in the HRV signal. Then, it uses a time-domain analysis algorithm to extract features such as SDNN (56 ms) from the processed HRV signal and features such as average ICP (16 mmHg) from the corrected intracranial pressure waveform. At the same time, it uses a sliding window filtering algorithm (calculating the mean with a 10-minute window) to smooth the 4-6 fluctuation of the VAS score to a stable 5, and integrates it with the pain location data (right parietal lobe region) to form subjective feedback features. Finally, it performs standardization transformation on the HRV features, ICP features, and subjective feedback features to generate a feature dataset, providing clean data for subsequent AI model calculations and avoiding errors in pain risk probability calculation caused by outliers.

[0093] In step S103, the feature dataset is input into the AI ​​model constructed by the XGBoost algorithm, which outputs the pain risk probability value and generates a risk level. The pain nature information and pain location of the subjective feedback features are combined with a preset rule base to generate etiology labels.

[0094] The process involves matching a pre-defined rule base to generate etiology labels. First, key information about the nature and location of pain is extracted from subjective feedback features, while the risk level generated by the AI ​​model is used to narrow down the matching range. Then, the pre-defined etiology rule base is invoked, and a weighted matching algorithm is used to comprehensively calculate the nature of the current patient's pain, the location of the pain, other auxiliary features, and the risk level. The algorithm assigns corresponding weights to the risk level (weight 0.35), the nature of pain (weight 0.25), the location of pain (weight 0.25), and other auxiliary features (weight 0.15), calculates the matching score with each item in the rule base, and selects the associated item with the highest score to finally generate etiology labels.

[0095] Weighted matching algorithm formula:

[0096] Matchscore(k) = w1·S risk (k)+w2·S nature (k)+w3·S location (k)+w4·S other (k)

[0097] Where Matchscore(k) is the total matching score between the patient's pain characteristics and the k-th etiology rule in the rule base; w1 is the weighting coefficient of the risk level; S risk (k) is the weighting coefficient for the nature of pain; w3 is the weighting coefficient for the location of pain; w4 is the weighting coefficient for other auxiliary features; S risk (k) represents the degree of match between the patient's current risk level and the risk level requirement of the kth rule in the rule base; S nature (k) represents the degree of match between the patient's current wind-induced pain nature and the k-th pain nature requirement in the rule base; S location (k) represents the match between the patient's current pain location and the k-th pain location requirement in the rule base; S other (k) represents the degree of match between the patient's current auxiliary features and the kth auxiliary feature requirement in the rule base.

[0098] It is understood that the embodiments of this application generate etiology labels by matching a preset rule base, which serves as the core link in the transformation of AI model output results into clinical decision-making. Through the multi-feature quantification logic of the weighted matching algorithm, the mapping from patient characteristics to etiology is performed. In complex cases, the etiology association items with the highest comprehensive scores are selected by calculating the matching degree of multiple dimensions such as risk level, pain nature, pain location, and other auxiliary features. This ensures that the labels are highly consistent with the patient's actual condition, providing standardized etiology basis for subsequent individualized analgesia plan formulation, emergency intervention triggering, and prognostic assessment, thereby improving the efficiency of identifying the etiology of complex perioperative pain and the accuracy of clinical decision-making.

[0099] In step S104, based on the risk level and etiology label, the corresponding intervention plan is retrieved from the preset plan library and automatically pushed to the medical staff's mobile terminal and the patient's terminal for intervention. At the same time, patient feedback data is received. If the VAS score decreases by more than or equal to 30%, the plan is deemed effective. If the expected results are not met, the plan is automatically upgraded.

[0100] It is understood that the embodiments of this application, through the matching logic of risk level and etiology label, call the corresponding intervention plan from the preset plan library and automatically push it to the medical staff's mobile terminal and the patient's terminal to quickly start the pain intervention process; at the same time, it receives patient feedback data in real time, and judges the effectiveness of the plan with a VAS score decrease of greater than or equal to 30% as an objective standard. If the standard is not met, the plan is automatically upgraded and the intervention strategy is dynamically adjusted; this avoids the time cost of medical staff manually searching for plans, shortens the response cycle from etiology confirmation to intervention implementation, ensures that high-risk pain is managed in a timely manner, and reduces ineffective intervention caused by subjective judgment bias through quantitative assessment and automatic upgrade mechanism, thus ensuring the effect of pain control for patients; it improves the accuracy and efficiency of pain management and the patient's treatment experience, and reduces the risk of postoperative recovery being affected by improper pain management.

[0101] For example, on the second day after a patient underwent craniotomy for a left temporal lobe meningioma, the postoperative pain monitoring system analyzed the patient's HRV signal (SDNN = 42ms, indicating pain-related autonomic nerve fluctuations), intracranial pressure waveform (average ICP = 19 mmHg, slightly above the normal range), and subjective VAS score (8 points). The system automatically output a risk level and etiology label of "medium risk - postoperative cerebral edema-related pain." Based on this label, the system retrieved an intervention plan from its preset protocol library: "100ml mannitol rapid intravenous infusion (completed within 30 minutes, monitoring urine output and electrolytes) + 0.3g oral acetaminophen tablets (taken after meals, every 8-12 hours)." This plan was simultaneously pushed to the attending physician's mobile device (including dosage and monitoring frequency) and the patient's device (including medication purpose and feedback instructions) within 10 seconds. After physician confirmation, the intervention was immediately implemented. Thirty minutes later, the patient reported a VAS score reduction to 5 points via the mobile device. The system calculated a reduction of approximately 37.5%, determined the plan was effective, and recorded the data. On the third postoperative day, the patient's VAS score rose to 7 due to a change in body position. The system re-matched the original intervention plan, but one hour later, the VAS score only dropped to 6, a decrease of 14.3%, which did not meet the target. The system automatically triggered a plan upgrade, pushing an adjusted plan: "Based on the original plan, add 100ml mannitol intravenously combined with 250ml glycerol fructose intravenously (twice daily, alternating with mannitol to reduce kidney burden) + enhanced dynamic monitoring of intracranial pressure (every 30 minutes) + continued oral acetaminophen tablets 0.3g (every 8-12 hours as needed)." Two hours after implementing the new plan, the VAS score dropped to 3, and the system confirmed the effect was achieved. Simultaneously, the two intervention data (plan, effect, and adjustment logic) for "medium risk - postoperative cerebral edema-related pain" were stored in the database to provide a reference for optimizing plans for similar cases in the future.

[0102] According to the embodiments of this application, a method for perioperative pain assessment in patients undergoing brain tumor surgery is proposed. This embodiment continuously acquires HRV signals and captures real-time intracranial pressure waveforms through a pain data acquisition module, simultaneously acquiring subjective VAS scores. It combines a 3D head model with NLP technology to label pain areas and durations, comprehensively collecting multi-dimensional data including objective physiological data, subjective feelings, and location information, thus improving the data integrity of perioperative pain assessment for brain tumor patients. A feature extraction module uses outlier detection and sliding window filtering algorithms to remove data interference and smooth fluctuations. A time-domain analysis algorithm extracts core features of HRV and intracranial pressure and transforms them into subjective feedback features, integrating and outputting standardized data in a unified format to improve data quality and consistency. An AI pain quantification module further enhances the standardized data. The AI ​​model, built using the XGBoost algorithm, generates pain risk probability values. Based on these probabilities, it triggers tiered warnings and outputs etiology labels by combining subjective feature matching rule bases, improving the objectivity of pain risk assessment and the accuracy of etiology identification. Through an intervention push module, warning prompts are simultaneously triggered on nurse station screens, patient mobile phones, and medical staff mobile terminals. Corresponding intervention plans are invoked based on risk level and etiology, and a graphical interface is provided to receive patient feedback, enhancing the timeliness of pain intervention and the efficiency of doctor-patient interaction. A management and traceability module stores the entire closed-loop data in a structured timeline, automatically determining the effectiveness of interventions and upgrading substandard plans. High-quality training sets are regularly pushed to optimize the AI ​​model, and multi-condition data queries are performed to connect with hospital systems, ensuring the system's continuous iteration capability and clinical compliance. This addresses the problems of high pain assessment error, unquantified coupling of intracranial pressure and pain, and delayed intervention in existing technologies.

[0103] The following will illustrate a method for assessing perioperative pain in patients undergoing brain tumor surgery through a specific example. Figure 9 As shown, it includes:

[0104] Patient Wang, 58 years old, female, was admitted to the neurosurgery department for "left temporal lobe meningioma" and was scheduled to undergo craniotomy for tumor resection. Two days before the operation, she was transferred to the intensive care unit (ICU). The medical team carried out full-cycle pain management according to the pre-set procedures. The first phase involves data acquisition. Using multimodal acquisition devices and an interactive terminal, four core data types were simultaneously acquired over a 23-hour period from 8:00 AM two days before surgery to 7:00 AM one day before surgery: HRV signals were acquired using a non-invasive ECG monitor. The monitor, with three leads attached to the patient's chest (right midclavicular line second intercostal space, left anterior axillary line fifth intercostal space, left midclavicular line second intercostal space), with a sampling frequency of 250Hz, continuously recording heart rate variability signals, focusing on capturing RR interval data (the time interval between two consecutive heartbeat R waves). One set of data was automatically stored every 5 minutes, generating a total of 276 sets of valid raw HRV data to reflect the patient's autonomic nervous function and pain-related physiological fluctuations. Intracranial pressure waveforms were acquired using an invasive intracranial pressure monitor. The sensor was placed epidurally in the left temporal region of the patient via minimally invasive surgery, with a sampling frequency of 1Hz, recording real-time dynamic changes in intracranial pressure, generating 36 data points per hour. 00 pressure data points were collected, and intracranial pressure waveforms were output synchronously to monitor baseline pressure values ​​and fluctuations, ruling out the risk of abnormally high intracranial pressure before surgery. Subjective VAS scores were collected through the patient's mobile interface. The system automatically popped up a scoring reminder every 2 hours. Patients selected the corresponding score on a slider from 0 to 10 (0 for no pain, 10 for the most severe pain) based on their own pain perception. If the patient did not respond in time due to fatigue or sleep, the system would remind them again after 15 minutes. A total of 12 valid scores were completed within 23 hours, with scores ranging from 3 to 7. Pain location data were collected through a mobile visual human atlas. When the patient clicked on the left temporal region in the atlas (which corresponds to the location of the tumor), the system automatically marked it as "pain in the left temporal lobe area." The system also supported patients to add text descriptions, such as "pain is a throbbing pain, which is aggravated when bending over." This information was stored synchronously with the location data.

[0105] After data acquisition, the system enters the data processing and feature extraction stage, standardizing the four types of raw data: For HRV signals, the system uses an outlier detection algorithm to identify three sets of abnormal data (380ms, 1250ms, and 420ms, respectively), all caused by the patient turning over and raising their arms. The system automatically removes these and retains 273 sets of valid data. Subsequently, the system extracts HRV features through a time-domain analysis algorithm, including SDNN (standard deviation of all sinus RR intervals) 58ms, RMSSD (root mean square of the difference between adjacent RR intervals) 22ms, and NN50 (number of adjacent RR intervals with a difference greater than 50ms) 8, which fully reflect the patient's autonomic nervous system regulation state. For intracranial pressure waveforms, the system uses an outlier detection algorithm to identify two brief jumps in data (25 mmHg and 27 mmHg, respectively), both of which occurred when the patient coughed or expectorated. After removing the outliers, the system extracts ICP features through time-domain analysis, including average ICP = 15 mmHg, peak ICP frequency = 0.2 Hz, and ICP fluctuation amplitude = 3 mmHg, to ensure the stability of intracranial pressure data. For subjective VAS scores, the system uses a sliding window filtering algorithm to calculate smoothing values. For example, the original score sequence "3, 5, 7, 6, 4" is smoothed to "4.6, 5.2, 5.6, 5.0, 4.8", effectively eliminating score fluctuations caused by patients' emotional tension (such as worry about the surgical outcome). Subsequently, the smoothed VAS scores are integrated with the pain location data (distending pain in the left temporal lobe) into subjective feedback features. The Min-Max standardization method (mapping the data to the [0,1] interval) is used to uniformly transform the HRV features, ICP features, and subjective feedback features, finally generating a standardized feature dataset containing 12 dimensions, laying the foundation for subsequent AI model analysis.

[0106] After the feature dataset is generated, the system inputs it into the AI ​​model built using the XGBoost algorithm. The model calculates and outputs a pain risk probability value, and automatically generates a "high-risk" level based on preset risk level classification rules. Simultaneously, the model calls upon the pain nature information (distending pain, aggravated by bending over) and pain location (left temporal lobe region) from the subjective feedback features, matching them against a preset etiology rule base (the rule base contains 28 clinical rules such as "temporal lobe distending pain + aggravated by bending over → tumor space-occupying pain" and "incision area stabbing pain → postoperative incision pain"). Through a weighted matching algorithm, it generates the etiology label "left temporal lobe meningioma space-occupying pain". To ensure the accuracy of the results, the system simultaneously pushes the feature dataset and analysis results to the attending physician's mobile device. The physician, combining the patient's preoperative imaging examination (MRI showing a left temporal lobe meningioma approximately 3.5cm in diameter, compressing surrounding brain tissue) and physical examination (positive tenderness in the left temporal region), confirms that the risk level and etiology label output by the AI ​​model are correct, and clicks "confirm" to enter the intervention plan implementation stage.

[0107] Based on the matching results of "high risk + pain caused by left temporal lobe meningioma", the system retrieves the corresponding intervention plan from the preset plan library: drug intervention is "oral ibuprofen sustained-release capsules 0.3g, once every 12 hours, taken after meals" and "intravenous infusion of mannitol injection 100ml, once every 8 hours, rapid infusion (completed within 30 minutes)"; non-drug intervention is "relaxation training (20 minutes each time, 1 hour apart)" and "postural guidance (elevate the head of the bed to 30°, avoid bending over and lowering the head)"; monitoring requirements are "record VAS score every 1 hour, monitor HRV and ICP data every 2 hours". Within 10 seconds of the plan being generated, it is automatically pushed to the attending physician, the responsible nurse's mobile terminal (including medication dosage, execution time, and precautions), and the patient's mobile terminal. After receiving the treatment plan, the responsible nurse administered the first drug intervention (oral ibuprofen and intravenous mannitol) to the patient at 8:30 AM the day before surgery. At 9:30 AM, the nurse assisted the patient with relaxation training and instructed the patient to adjust the head of the bed to 30°. One hour later (9:30 AM), the patient reported via mobile device that their VAS score had dropped to 5 points. The system calculated the reduction as (preoperative baseline score of 7 points - current score of 5 points) / 7 points ≈ 28.6%, which did not meet the 30% effective standard, and automatically triggered a plan upgrade. The upgraded plan added the following to the original plan: **"Rapid intravenous infusion of 250ml of glycerol fructose injection twice daily (alternating with mannitol to reduce the burden on the kidneys)", "Oral administration of 0.3g of acetaminophen tablets every 8-12 hours (taken after meals to enhance analgesia)", and "Transcutaneous electrical stimulation therapy to the painful area (using low-frequency current, 20 minutes each time, 3 times daily, stimulation parameters: frequency 100Hz, intensity as tolerated by the patient)"**, and increased the VAS score monitoring frequency to once every 30 minutes. At 10:00, the nurse administered a rapid intravenous infusion of 250ml of glycerol fructose injection to the patient (controlling the infusion rate to ensure completion within approximately 2 hours for stable dehydration), and instructed the patient to take 0.3g of acetaminophen tablets orally (immediately after meals to reduce gastrointestinal irritation); at 10:30, the patient underwent their first transcutaneous electrical nerve stimulation (TENS) treatment on the painful area (using a low-frequency current mode, setting the frequency to 100Hz, and adjusting the intensity to the patient's subjective tolerance and without discomfort, for 20 minutes). At 11:00, the patient reported that their VAS score had decreased to 3 points, a reduction of approximately 57.1% (7-3) / 7, meeting the ≥30% effective standard. The system determined the treatment to be effective and automatically recorded the intervention process data (drug name, dosage, execution time, and effect).Over the next 12 hours, the system continuously monitored patient data. The VAS score remained stable at 2-3 points, the SDNN in the HRV feature increased to 65ms (improved autonomic nervous system regulation), and the average ICP in the ICP feature decreased to 13mmHg (reduced cerebral edema), with no abnormal fluctuations. At 20:00 one day before surgery, the system compiled and generated information on the entire process of data collection, processing, model analysis, and intervention effects, providing a reference for the surgery the next day and postoperative pain management, ensuring that the patient faced surgery with a stable pain state and reducing the surgical risks caused by pain stress.

[0108] In summary, this application's embodiments, based on multimodal precision data, rationally process and eliminate interference, extract effective features, utilize AI models to accurately determine risks and causes, match and dynamically upgrade appropriate intervention plans, standardize and intelligentize the entire pain management process, shorten intervention response time, ensure timely control of high-risk pain, reduce ineffective interventions, stabilize patients' preoperative state, reduce surgical risks associated with pain stress, and provide a reliable practical paradigm for perioperative pain management.

[0109] Figure 10 A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include:

[0110] The memory 1001, the processor 1002, and the computer program stored on the memory 1001 and capable of running on the processor 1002.

[0111] When the processor 1002 executes the program, it implements a perioperative pain assessment method for patients undergoing brain tumor surgery provided in the above embodiments.

[0112] Furthermore, electronic devices also include:

[0113] Communication interface 1003 is used for communication between memory 1001 and processor 1002.

[0114] The memory 1001 is used to store computer programs that can run on the processor 1002.

[0115] The memory 1001 may include high-speed RAM (Random Access Memory) memory, and may also include non-volatile memory, such as at least one disk storage.

[0116] If the memory 1001, processor 1002, and communication interface 1003 are implemented independently, then the communication interface 1003, memory 1001, and processor 1002 can be interconnected via a bus to complete communication between them. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 10 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0117] Optionally, in a specific implementation, if the memory 1001, processor 1002, and communication interface 1003 are integrated on a single chip, then the memory 1001, processor 1002, and communication interface 1003 can communicate with each other through an internal interface.

[0118] The processor 1002 may be a CPU (Central Processing Unit), an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of this application.

[0119] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for assessing perioperative pain in patients undergoing brain tumor surgery.

[0120] Furthermore, this application also provides a computer program product, including a computer program or instructions, which, when executed, implement the aforementioned method for perioperative pain assessment in patients undergoing brain tumor surgery.

[0121] In the description of this specification, the references to "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0122] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0123] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0124] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0125] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0126] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A perioperative pain assessment system for patients undergoing brain tumor surgery, characterized in that, include: The module includes a surgical pain data acquisition module, a feature extraction module, an AI pain quantification module, an intervention push module, and a management and traceability module; among which, The surgical pain data acquisition module is used to collect patient HRV signals, intracranial pressure waveforms, subjective VAS scores, and pain location data. The feature extraction module is used to process the HRV signal, intracranial pressure waveform, subjective VAS score and pain site data through outlier detection algorithm and sliding window filtering algorithm, and then extract HRV features and ICP features by combining time domain analysis algorithm. After integrating with subjective feedback features, standardized data is output. The AI ​​pain quantification module is used to input the standardized data into an AI model built on the XGBoost algorithm, generate a pain risk probability value, and trigger a three-level warning based on the probability value. At the same time, it associates the nature and cause of the patient's pain. The intervention push module triggers multi-terminal prompts based on the warning level output by the AI ​​model, and pushes drug or non-drug intervention plans in combination with the nature and cause of pain. At the same time, patients can select plans and provide feedback through graphical interaction on their mobile phones. The management and traceability module is used to link the entire process data of a single pain assessment, record the intervention effect, and automatically upgrade solutions that do not meet expectations.

2. The perioperative pain assessment system for brain tumor surgery patients according to claim 1, characterized in that, The surgical pain data acquisition module includes an HRV signal acquisition unit, an intracranial pressure waveform acquisition unit, a subjective VAS score acquisition unit, and a pain location acquisition unit. The HRV signal acquisition unit continuously acquires HRV signals with a sampling interval of less than or equal to 1 minute. The intracranial pressure waveform acquisition unit acquires real-time intracranial pressure waveforms. The subjective VAS score acquisition unit acquires the patient's subjective score, obtaining a VAS score and simultaneously triggering a photo archive to prevent data falsification. The pain location acquisition unit provides the patient with a 3D head anatomy model for interactive viewing on their mobile phone, allowing the patient to select the pain area via touchscreen. If the patient is unable to operate the device, a nurse can describe the pain via voice, and NLP will automatically annotate the pain location, while simultaneously recording the duration of the pain.

3. The perioperative pain assessment system for brain tumor surgery patients according to claim 1, characterized in that, The feature extraction module includes a data cleaning unit, a multi-feature extraction unit, and a standardization output unit. The data cleaning unit removes interference values ​​from the HRV signal and real-time intracranial pressure waveform using an outlier detection algorithm, smooths VAS score fluctuations using a sliding window filtering algorithm, and verifies the integrity of the pain location data. The multi-feature extraction unit extracts the SDNN and RMSSD features of the HRV signal and the average ICP and amplitude variation coefficient features of the intracranial pressure using a time-domain analysis algorithm, and converts the VAS score, pain location, and duration data into standardized subjective feedback features. The standardization output unit integrates the SDNN, RMSSD, average ICP, amplitude variation coefficient features, and subjective feedback features into a unified format, forming standardized data, and transmits it to the AI ​​pain quantification module.

4. The perioperative pain assessment system for brain tumor surgery patients according to claim 1, characterized in that, The AI ​​pain quantification module includes a model computation unit, a three-level early warning determination unit, and a pain etiology association unit. The model computation unit is used to input the standardized data into an AI model constructed by the XGBoost algorithm to generate a pain risk probability value in the range of 0 to 1. The three-level early warning determination unit triggers a three-level early warning based on the probability value, obtains the risk level, and generates an early warning triggering reason report. The pain etiology association unit, based on the risk level, combines the pain nature information and pain location in the subjective feedback features, matches a preset rule base, and outputs etiology labels.

5. The perioperative pain assessment system for brain tumor surgery patients according to claim 1, characterized in that, The intervention push module includes a multi-terminal prompting unit, a plan matching and push unit, and an interactive feedback unit. The multi-terminal prompting unit is used to trigger early warning prompts on the nurse station screen, the patient's mobile phone, and the medical staff's mobile terminal. The plan matching and push unit is used to call the corresponding plan from the database according to the risk level and etiology label. The interactive feedback unit is used to provide a graphical interactive interface for patients, receive patients' active feedback, and transmit it to the management and traceability module.

6. The perioperative pain assessment system for brain tumor surgery patients according to claim 1, characterized in that, The management and traceability module includes a full-process data association unit, an effect evaluation and upgrade unit, a dataset generation and push unit, and a data interaction unit. The full-process data association unit connects data from all stages and stores it in a structured manner along a timeline. The effect evaluation and upgrade unit determines an effective intervention based on a VAS decrease of 30% or more after intervention; if the VAS decrease is less than 30%, the intervention is automatically upgraded. The dataset generation and push unit filters high-quality data with complete feedback weekly, generates a training set, and pushes it to the AI ​​pain quantification module for incremental model optimization. The data interaction unit provides multi-condition data query functionality and bidirectionally interfaces with the hospital's existing systems to read basic patient information.

7. A method for assessing perioperative pain in patients undergoing brain tumor surgery according to any one of claims 1-6, characterized in that, include: Collect patient HRV signals, intracranial pressure waveforms, subjective VAS scores, and pain location data; The outlier detection algorithm removes interference values ​​from the HRV signal and the real-time intracranial pressure waveform. The time-domain analysis algorithm is used to extract HRV features and ICP features. The sliding window filtering algorithm is used to smooth the fluctuation of the VAS score. The features are then integrated with the pain site data to form subjective feedback features. After standardization transformation, a feature dataset is generated. The feature dataset is input into the AI ​​model constructed by the XGBoost algorithm, which outputs a pain risk probability value and generates a risk level. The pain nature information and pain location of the subjective feedback features are combined with a preset rule base to generate etiology labels. Based on the risk level and the etiology label, the corresponding intervention plan is retrieved from the preset plan library and automatically pushed to the medical staff's mobile terminal and the patient's terminal for intervention. At the same time, patient feedback data is received. If the VAS score drops by more than or equal to 30%, the plan is deemed effective. If the expected results are not met, the plan is automatically upgraded.

8. An electronic device, characterized in that, include: The device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the program to implement the perioperative pain assessment method for patients undergoing brain tumor surgery as described in claim 7.

9. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed, they implement the perioperative pain assessment method for patients undergoing brain tumor surgery as described in claim 7.

10. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed, they implement the perioperative pain assessment method for patients undergoing brain tumor surgery as described in claim 7.

Images