Multi-source data integration grading early warning method and system for hand surgery vascular crisis

Abstract

The invention relates to the technical field of medical monitoring, in particular to a multi-source data integration grading early warning method and system for hand surgery vascular crisis. The invention discloses a multi-source data integration grading early warning method for hand surgery vascular crisis. The method comprises the following steps: S1, synchronously collecting tissue oxygen saturation, an original and uninjured side temperature difference value, behavioral environment data and clinical auxiliary data of a sentinel area through a multi-source sensor; s2, establishing a behavior recognition algorithm model, recognizing interference behaviors of the patient in real time based on the collected behavior environment data, and quantifying the interference behaviors to obtain influence data; and S3, based on the influence data, carrying out dynamic correction on the original and uninjured side temperature difference value to obtain a corrected and uninjured side temperature difference value, and carrying out dynamic correction on the tissue oxygen saturation to obtain the corrected tissue oxygen saturation. Through multi-source data fusion and intelligent correction, interference of patient behaviors and environmental factors on monitoring data can be effectively recognized and filtered out.

Description

Technical Field

[0001] This invention relates to the field of medical monitoring technology, and in particular to a multi-source data integration and hierarchical early warning method and system for vascular crises in hand surgery. Background Technology

[0002] Early identification of vascular crises in hand surgery is crucial for postoperative monitoring after finger replantation and flap transplantation. Currently, clinical monitoring primarily relies on healthcare professionals regularly observing indicators such as tissue color, tension, capillary refill time, and skin temperature difference. While skin temperature difference is relatively objective, it is easily affected by patient positioning changes, local exposure, or contact with heat sources, leading to a false positive alarm rate as high as 30%–40%. Furthermore, traditional methods often rely on single parameters, making it difficult to distinguish between interference and true pathological changes; and skin temperature changes are delayed relative to blood flow, resulting in delayed warnings and missed optimal intervention windows. Existing monitoring systems mostly use fixed threshold alarms, lacking intelligent multi-source information fusion and risk grading mechanisms. This not only causes alarm fatigue among healthcare professionals but also increases the risk of missing true crises. Moreover, current monitoring methods lack systematic and continuous monitoring and anti-interference analysis of the key physiological parameter of tissue oxygen saturation, failing to achieve multi-dimensional cross-validation with skin temperature monitoring.

[0003] Therefore, there is an urgent need to develop a multi-source data integration and hierarchical early warning method and system for hand surgical vascular crises to overcome the shortcomings of existing technologies in terms of anti-interference ability, data fusion, early warning timeliness and hierarchical response, so as to achieve accurate identification, early warning and hierarchical management of postoperative vascular crises, and improve clinical monitoring efficiency and prognosis. Summary of the Invention

[0004] To overcome the shortcomings of existing hand surgery vascular crisis monitoring methods, such as strong subjectivity, intermittent monitoring, non-quantitative evaluation indicators, delayed early warning, high false positive rate, and lack of intelligent hierarchical response mechanism, this invention provides a multi-source data integration hierarchical early warning method and system for hand surgery vascular crises.

[0005] The technical solution of this invention is: a multi-source data integration and hierarchical early warning method for vascular crises in hand surgery, comprising the following steps: S1: Synchronously collect tissue oxygen saturation, raw and healthy side temperature difference, behavioral environmental data and clinical auxiliary data of the sentinel area through multi-source sensors; S2: Establish a behavior recognition algorithm model to identify the patient's interfering behaviors in real time based on the collected behavioral environment data, and obtain impact data by quantifying the interfering behaviors; S3: Based on the aforementioned impact data, the original temperature difference value between the healthy and unhealthy sides is dynamically corrected to obtain the corrected temperature difference value between the healthy and unhealthy sides, and the tissue oxygen saturation is dynamically corrected to obtain the corrected tissue oxygen saturation. S4: Calculate the confidence score using the clinical auxiliary data, perform a fusion analysis between the corrected temperature difference with the healthy side and the confidence score to calculate the first risk score, and perform a fusion analysis between the corrected tissue oxygen saturation and the confidence score to calculate the second risk score. S5: Establish a comprehensive scoring prediction model, input the first risk score and the second risk score to obtain a comprehensive risk score, and generate a graded early warning strategy based on the comprehensive risk score and clinical rules; S6: Record each warning event, clinical intervention, and final outcome to optimize the behavior recognition algorithm model, adjust warning thresholds, and implement tiered warning strategies.

[0006] Preferably, the simultaneous collection of tissue oxygen saturation, original and healthy side temperature difference, behavioral environment data, and clinical auxiliary data in the sentinel area via multi-source sensors includes: the behavioral environment data includes three-dimensional posture data of the affected limb, local microenvironment temperature, radiant heat flux, patient habitual behavior data, ambient light intensity, and local blood flow change data; the clinical auxiliary data includes patient pain self-assessment data, medication data, and medical staff observation record data.

[0007] Preferably, the step of establishing a behavior recognition algorithm model to identify interfering behaviors of the patient in real time based on collected behavioral environment data, and obtaining impact data by quantifying the interfering behaviors, includes: extracting features from the behavioral environment data according to the behavior recognition algorithm model to obtain thermal interference behavior data, postural habit interference data, and oxygenation interference data; and defining a thermal interference behavior set based on the thermal interference behavior data. Each thermal disturbance behavior Including the The local microenvironment temperature and radiative heat flux at the occurrence of secondary thermal disturbances; based on the aforementioned postural behavior disturbance data, the set of postural habit disturbance behaviors is defined as follows: Each body position habit interferes with behavior Including the Three-dimensional posture data of the affected limb and patient habitual behavior data at the time of occurrence of postural habit interference behavior; the oxygenation interference behavior set is defined based on the oxygenation interference data. Each oxygenation disturbance behavior Including the Data on changes in ambient light intensity and local blood flow during the occurrence of secondary oxygenation interference behaviors were used. Based on the acquired set of interference behaviors, a data model was employed for quantitative analysis to obtain data on the impact of interference behaviors on the temperature difference between the affected and healthy sides. The set of interference behaviors included thermal interference behaviors, postural habit interference behaviors, and oxygenation interference behaviors.

[0008] Preferably, the step of using a data model to perform quantitative analysis based on the acquired set of interfering behaviors to obtain data on the impact of interfering behaviors on the temperature difference between the healthy and unaffected sides includes: the data model comprising a thermodynamic model, a hemodynamic model, and an oxygenation kinetic model; the impact data comprising temperature impact data, postural habit impact data, and oxygenation interference impact data; the thermodynamic model taking the identified set of thermal interfering behaviors as input, calculating using an algorithm based on the energy conservation-based heat balance equation, and outputting quantified temperature impact data; the hemodynamic model taking the identified set of postural habit interfering behaviors as input, calculating using a gravity-blood flow coupling algorithm based on fluid dynamics and circulatory physiology, and outputting quantified postural habit impact data; and the oxygenation kinetic model taking the identified set of oxygenation interference behaviors as input, outputting quantified oxygenation interference impact data using an algorithm based on a light propagation model and the blood flow-oxygenation coupling relationship.

[0009] Preferably, the step of dynamically correcting the original temperature difference with the healthy side to obtain a corrected temperature difference with the healthy side based on the influence data, and dynamically correcting the tissue oxygen saturation to obtain a corrected tissue oxygen saturation, includes: obtaining the corrected temperature difference with the healthy side according to the influence data using a correction formula for the original temperature difference with the healthy side, and obtaining the corrected tissue oxygen saturation according to the influence data using a tissue oxygen saturation correction formula, wherein the correction formula is: ; ; in This is the corrected temperature difference value between the healthy and unhealthy sides; This represents the original temperature difference between the healthy and unhealthy sides. For the first The data includes data on the effects of temperature and data on the effects of body position habits. For the first The weight parameters that affect the data by class; This is the corrected tissue oxygen saturation; Tissue oxygen saturation; Data affected by oxygenation interference; To adjust the parameters.

[0010] Preferably, the step of calculating a confidence score using the clinical auxiliary data, fusing the corrected temperature difference with the healthy side and the confidence score to calculate a first risk score, and fusing the corrected tissue oxygen saturation with the confidence score to calculate a second risk score, includes: calculating a confidence score based on the clinical auxiliary data using the clinical auxiliary data. Obtain confidence score ,in For normalized patient pain self-rating data, For normalized medication data, For the normalized observation records of medical staff, The non-linear weighting parameter is defined as: [The clinical standard deviation is then defined as...] A first risk score is obtained using a first risk scoring formula based on the original temperature difference between the healthy and unhealthy sides, the confidence score, and the corrected temperature difference between the healthy and unhealthy sides; a second risk score is obtained using a second risk scoring formula based on the tissue oxygen saturation, the confidence score, and the corrected tissue oxygen saturation; wherein the first risk scoring formula is: ; in Assign the first risk score; This is the corrected temperature difference value between the healthy and unhealthy sides; This represents the original temperature difference between the healthy and unhealthy sides. Clinical standard deviation; Score the confidence level; It is a very small constant to prevent the denominator from being zero.

[0011] Preferably, the step of obtaining the second risk score based on the tissue oxygen saturation, confidence score, and corrected tissue oxygen saturation using a second risk scoring formula includes: defining the normal standard deviation of oxygen saturation fluctuation as... The second risk score is obtained through a second risk scoring formula, which is as follows: ; in This is the second risk score; This is the corrected tissue oxygen saturation; Tissue oxygen saturation; This represents the standard deviation of normal fluctuations in oxygen saturation. Score the confidence level; It is a very small constant to prevent the denominator from being zero.

[0012] Preferably, the step of establishing a comprehensive risk score prediction model, which involves inputting the first risk score and the second risk score to obtain a comprehensive risk score, and generating a graded early warning strategy based on the comprehensive risk score and clinical rules, includes: the comprehensive risk score prediction model is a nonlinear regression model containing interaction terms, which, by inputting the first risk score and the second risk score, employs... Obtain a comprehensive risk score The tiered early warning strategy includes behavioral intervention-level early warning, enhanced observation-level early warning, clinical assessment-level early warning, and emergency response-level early warning. A behavioral intervention-level early warning is issued when a significant interfering behavior is detected and the overall risk score is below the Level 1 risk threshold. An enhanced observation-level early warning is issued when the overall risk score is above the Level 1 risk threshold but below the Level 2 risk threshold. A clinical assessment-level early warning is issued when the overall risk score is above the Level 2 risk threshold but below the Level 3 risk threshold. An emergency response-level early warning is issued when the overall risk score is above the Level 3 risk threshold.

[0013] Preferably, the recording of each early warning event, clinical intervention, and final outcome for optimizing the behavior recognition algorithm model, adjusting the early warning threshold, and the graded early warning strategy includes: when an early warning is triggered, collecting and processing early warning data, wherein the processed early warning data includes the graded early warning strategy, clinical intervention, and the tissue oxygen saturation and the original temperature difference between the original and healthy side after intervention; optimizing the behavior recognition model and the weight parameters of the influencing data based on the processed early warning data, and adjusting the graded early warning threshold and the graded logic.

[0014] Preferably, the multi-source data integration and hierarchical early warning system for hand surgical vascular crises further includes: The sensing and acquisition module synchronously collects tissue oxygen saturation, raw and healthy side temperature difference, behavioral environment data, and clinical auxiliary data in the sentinel area through multi-source sensors. The behavior recognition module establishes a behavior recognition algorithm model, identifies interfering behaviors of patients in real time based on collected behavioral environment data, and obtains impact data by quantifying the interfering behaviors. The data correction and confidence assessment module, based on the impact data, dynamically corrects the original temperature difference between the original and healthy sides to obtain the corrected temperature difference between the original and healthy sides, dynamically corrects the tissue oxygen saturation to obtain the corrected tissue oxygen saturation, calculates a confidence score using the clinical auxiliary data, fuses the corrected temperature difference between the original and healthy sides with the confidence score to calculate a first risk score, and fuses the corrected tissue oxygen saturation with the confidence score to calculate a second risk score. The intelligent early warning module establishes a comprehensive scoring prediction model, inputs the first risk score and the second risk score to obtain a comprehensive risk score, generates a graded early warning strategy based on the comprehensive risk score and clinical rules, and records each early warning event, clinical intervention measures and final outcome, which is used to optimize the behavior recognition algorithm model, adjust the early warning threshold and graded early warning strategy.

[0015] The beneficial effects are as follows: This invention, through multi-source data fusion and intelligent correction, can effectively identify and filter out interference from patient behavior and environmental factors on monitoring data, significantly reducing the false positive alarm rate; through multi-parameter trend analysis, it identifies early risk changes before traditional threshold alarms, achieving true early warning; it provides a graded response strategy based on comprehensive risk scores, optimizing clinical resource allocation and intervention priorities; while reducing the workload of medical staff and improving patient compliance, the accumulated high-quality data also supports clinical research and continuous model optimization. The overall solution has good clinical applicability and scalability, enabling precise, timely, and graded management of surgical vascular crises at a high cost-performance ratio. This invention introduces an independent anti-interference assessment channel for tissue oxygen saturation and intelligently integrates it with a temperature difference assessment channel, achieving multi-dimensional, cross-validated early warning for vascular crises, further improving the specificity and reliability of early identification. Attached Figure Description

[0016] Figure 1 A flowchart of a multi-source data integration and hierarchical early warning method for vascular crises in hand surgery; Figure 2 This is a schematic diagram of a multi-source data integration and hierarchical early warning system for vascular crises in hand surgery. Detailed Implementation

[0017] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0018] Example 1: A multi-source data integration and hierarchical early warning method for vascular crises in hand surgery, such as... Figure 1 As shown, it includes the following steps: S1: Synchronously collect tissue oxygen saturation, raw and healthy side temperature difference, behavioral environmental data and clinical auxiliary data of the sentinel area through multi-source sensors; The behavioral environment data includes three-dimensional posture data of the affected limb, local microenvironment temperature, radiant heat flux, patient habitual behavior data, ambient light intensity, and local blood flow change data; the clinical auxiliary data includes patient pain self-assessment data, medication data, and medical staff observation record data.

[0019] It should also be noted that, to achieve accurate early warning of vascular crises, this invention integrates multi-source heterogeneous data for analysis. Tissue oxygen saturation and the initial temperature difference between the affected and healthy sides are used as the primary criteria, combined with behavioral and environmental data recordings of the affected limb's positioning, local temperature and humidity, radiant heat flux, patient habitual behavior data, ambient light intensity, and local blood flow changes. The affected limb's positioning is used to obtain parameters required for the hemodynamic model; the local temperature and humidity refer to the temperature and humidity data of the affected limb area; the radiant heat flux refers to the heat flux emitted by heat sources affecting the temperature of the affected limb; the patient habitual behavior data refers to the temperature changes caused by the patient's habitual behaviors; the ambient light intensity refers to the amount of light radiation in the environment surrounding the monitoring probe installed in the affected limb area; and the local blood flow change data refers to data reflecting instantaneous changes in the blood flow velocity or volume of the subcutaneous capillary bed in the monitoring area. Furthermore, clinical auxiliary data includes the patient's subjective feedback, medication history, and bedside observation records from medical staff, providing necessary humanistic and clinical context for the assessment.

[0020] S2: Establish a behavior recognition algorithm model to identify the patient's interfering behaviors in real time based on the collected behavioral environment data, and obtain impact data by quantifying the interfering behaviors; Based on the behavior recognition algorithm model, feature extraction is performed on the behavioral environment data to obtain thermal interference behavior data, postural habit interference data, and oxygenation interference data; based on the thermal interference behavior data, a thermal interference behavior set is defined as follows. Each thermal disturbance behavior Including the The local microenvironment temperature and radiative heat flux at the occurrence of secondary thermal disturbances; based on the aforementioned postural behavior disturbance data, the set of postural habit disturbance behaviors is defined as follows: Each body position habit interferes with behavior Including the Three-dimensional posture data of the affected limb and patient habitual behavior data at the time of occurrence of postural habit interference behavior; the oxygenation interference behavior set is defined based on the oxygenation interference data. Each oxygenation disturbance behavior Including the Data on changes in ambient light intensity and local blood flow during the occurrence of secondary oxygenation interference behaviors were used. Based on the acquired set of interference behaviors, a data model was employed for quantitative analysis to obtain data on the impact of interference behaviors on the temperature difference between the affected and healthy sides. The set of interference behaviors included thermal interference behaviors, postural habit interference behaviors, and oxygenation interference behaviors.

[0021] The data model includes a thermodynamic model, a hemodynamic model, and an oxygenation kinetic model; the influence data includes temperature influence data, postural habit influence data, and oxygenation disturbance influence data; the thermodynamic model takes the identified set of thermal disturbance behaviors as input, calculates using an algorithm based on the energy conservation-based heat balance equation, and outputs quantified temperature influence data; the hemodynamic model takes the identified set of postural habit disturbance behaviors as input, calculates using a gravity-blood flow coupling algorithm based on fluid mechanics and circulatory physiology, and outputs quantified postural habit influence data; the oxygenation kinetic model takes the identified set of oxygenation disturbance behaviors as input, and outputs quantified oxygenation disturbance influence data using an algorithm based on a light propagation model and the blood flow-oxygenation coupling relationship.

[0022] It should also be noted that the behavior recognition algorithm model collects historical patient behavior environment data videos and synchronized skin temperature data. Experts, based on abnormal skin temperature fluctuations and video recordings, categorize time periods into categories such as no interference, thermal interference, postural interference, and oxygenation interference. Features are extracted from the raw data, and a multi-class machine learning model, such as a random forest or lightweight convolutional neural network, is trained using the labeled feature dataset. The dataset is divided into training and testing sets in a ratio (e.g., 7:3). The model is trained using the training set, aiming to minimize classification error. The trained model is integrated into the system to extract features and classify real-time collected behavioral environment data, outputting the current interference behavior category and corresponding data subset. ).

[0023] In the quantitative analysis phase, three professional computational models were used to deconstruct the aforementioned interference effects one by one. First, based on the thermodynamic energy balance equation, a simplified skin heat balance equation was used, treating the affected side measurement point as a thermal system. Its heat balance equation is expressed as: ,in Data on the effect of temperature; The values ​​are the Min-Max normalized values ​​of the local microenvironment temperature. The values ​​are the normalized values ​​of the radiative heat flux Min-Max. The empirical coefficients are obtained by simultaneously collecting local microenvironment temperature rise, radiant heat flux, and corresponding skin temperature changes under typical scenarios simulating direct sunlight and hot compresses. With skin temperature change as the dependent variable and temperature rise and radiant flux as independent variables, a multiple linear regression is performed using the least squares method. The coefficients of each independent variable in the regression equation are the empirical coefficients required for the thermodynamic model. The steady-state temperature difference change is solved by substituting the preset heat transfer coefficient and the radiant heat flux, ambient temperature, and wind speed (estimated from microenvironment data) read by the sensor into the above equation.

[0024] Secondly, combining the principles of fluid mechanics and circulatory physiology to establish a hemodynamic model, a hydrostatic pressure correction model based on gravity effects is adopted. Based on the three-dimensional posture data of the affected limb, the height difference of a certain measurement point of the affected limb relative to the level of the heart is defined as... Changes in venous pressure at this point due to changes in body position Approximately ,in Blood density, The pressure change is due to gravity; this pressure change affects the perfusion pressure of local microcirculation, and thus affects skin temperature. The model establishes an empirical relationship curve of "height difference - perfusion pressure - skin temperature change" and uses a lookup table method to map the "height difference - skin temperature effect". The empirical relationship table is established by measuring the skin temperature difference between the back of the hand and the chest wall at the heart level of 10 healthy volunteers after placing their arms at different heights (relative to the level of the heart, from -40cm to +40cm, in 10cm intervals) for 5 minutes. The average temperature difference value of all volunteers is taken. The model is calculated based on real-time three-dimensional posture data. The query retrieves the corresponding data on the impact of body position habits. .

[0025] Finally, a tissue optical model was established based on the modified Lambert-Beer light propagation law. This model uses real-time ambient light intensity and local blood flow change data collected from the oxygenation interference behavior dataset as input. By calculating the deviation between ambient light intensity and standard measured illuminance, the optical interference component of the external light source on the tissue oxygen saturation detection signal was quantified. Combined with the blood flow-oxygenation coupling equation, the theoretical impact on tissue oxygen saturation was dynamically calculated based on local blood flow change data, ultimately quantifying and obtaining oxygenation interference behavior data. .

[0026] S3: Based on the aforementioned impact data, the original temperature difference value between the healthy and unhealthy sides is dynamically corrected to obtain the corrected temperature difference value between the healthy and unhealthy sides, and the tissue oxygen saturation is dynamically corrected to obtain the corrected tissue oxygen saturation. Based on the aforementioned impact data, the corrected temperature difference with the healthy side is obtained using the original temperature difference correction formula. Similarly, the corrected tissue oxygen saturation is obtained using the aforementioned impact data using the tissue oxygen saturation correction formula, where the correction formula is: ; ; in This is the corrected temperature difference value between the healthy and unhealthy sides; This represents the original temperature difference between the healthy and unhealthy sides. For the first The data includes data on the effects of temperature and data on the effects of body position habits. For the first The weight parameters that affect the data by class; This is the corrected tissue oxygen saturation; Tissue oxygen saturation; Data affected by oxygenation interference; To adjust the parameters.

[0027] It should also be noted that the calculation relationship used in skin temperature difference correction is expressed as follows: In this formula, the corrected temperature difference with the healthy side... The aim is to eliminate non-pathological interference, thereby more accurately reflecting the temperature difference between the affected and healthy sides determined by the state of blood vessels. This is the raw temperature difference reading obtained through direct measurement. This represents the various interference effects that have been quantitatively assessed, primarily summarizing data on the effects of temperature and body position habits. Each item... All results are calculated using corresponding professional analytical models. The weighting coefficients corresponding to the interference effect are based on historical clinical data. Through multiple linear regression or machine learning algorithms, the actual vascular state is used as a label to backfit the weighting parameters that make the corrected temperature difference closest to the physiological true value.

[0028] Tissue oxygen saturation correction formula In the middle, adjust the parameters In a controlled laboratory environment, data on the impact of oxygenation disturbances on patients during oxygenation disturbance behaviors are recorded simultaneously. Corresponding original change in tissue oxygen saturation Based on the linear relationship between the two, the least squares method is used to fit the data points, and the slope of the resulting regression equation is the initial value of the adjustment parameter γ.

[0029] S4: Calculate the confidence score using the clinical auxiliary data, perform a fusion analysis between the corrected temperature difference with the healthy side and the confidence score to calculate the first risk score, and perform a fusion analysis between the corrected tissue oxygen saturation and the confidence score to calculate the second risk score. Based on the aforementioned clinical auxiliary data, Obtain confidence score ,in For normalized patient pain self-rating data, For normalized medication data, For the normalized observation records of medical staff, The non-linear weighting parameter is defined as: [The clinical standard deviation is then defined as...] A first risk score is obtained using a first risk scoring formula based on the original temperature difference between the healthy and unhealthy sides, the confidence score, and the corrected temperature difference between the healthy and unhealthy sides; a second risk score is obtained using a second risk scoring formula based on the tissue oxygen saturation, the confidence score, and the corrected tissue oxygen saturation; wherein the first risk scoring formula is: ; Assign the first risk score; This is the corrected temperature difference value between the healthy and unhealthy sides; This represents the original temperature difference between the healthy and unhealthy sides. Clinical standard deviation; Score the confidence level; It is a very small constant to prevent the denominator from being zero.

[0030] Define the standard deviation of normal fluctuations in oxygen saturation as The second risk score is obtained through a second risk scoring formula, which is as follows: ; in This is the second risk score; This is the corrected tissue oxygen saturation; Tissue oxygen saturation; This represents the standard deviation of normal fluctuations in oxygen saturation. Score the confidence level; It is a very small constant to prevent the denominator from being zero.

[0031] It should also be noted that the formula used in generating the confidence score is... A comprehensive calculation was performed. The model integrates three dimensions of clinical information: patient-reported pain self-rating data, current medication records, and bedside observations and comments from healthcare professionals. Before calculation, all data were normalized using Min-Max to eliminate dimensional differences and ensure consistency in the scoring scale. Weighting parameters. Embedded in a non-linear form, its value is dynamically calculated based on the historical correlation between the data source and the diagnosis results of the crisis and the weight of data quality indicators. First, the early warning accuracy rate (true positive rate / total number of early warnings) of each data source is calculated as the basic weight, and then dynamically adjusted according to the degree of deviation between the current data and the patient's baseline - the greater the deviation, the higher the weight.

[0032] The clinical standard deviation is based on publicly available clinical studies. Data on the original and healthy side temperature differences during the stable period in patients with good postoperative outcomes and no vascular crisis were collected and labeled. A normal temperature difference database was constructed, and the standard deviation of all temperature difference data in the database was calculated to obtain the clinical standard deviation. The value essentially characterizes the normal range of skin temperature fluctuations under physiological conditions, providing an objective statistical benchmark for subsequent identification of deviations from normal values ​​in individual patients.

[0033] The normal fluctuation standard deviation of oxygen saturation is obtained by collecting a large amount of time-series data of tissue oxygen saturation from patients who have recovered well after surgery and have not experienced vascular crisis during a stable monitoring period. After removing obviously disturbed segments, the fluctuation standard deviation of all valid data points around their individual mean is calculated. Then, the standard deviation of all patients is statistically aggregated to establish a population benchmark parameter that characterizes the natural variation range of oxygen saturation under physiological conditions.

[0034] Risk Score Given by the following formula: , This formula combines the magnitude of data change before and after correction with the aforementioned confidence score. The exponential term... Used to amplify the significance of abnormal temperature differences relative to normal fluctuations; while the confidence adjustment part in the multiplication term Then it will be at the confidence level When the risk level is low, proactively increase the risk score to incorporate data uncertainty into risk assessment. The formula introduces a minimal constant. This is designed to prevent mathematical errors such as a zero denominator when C approaches zero, thus ensuring the numerical stability of the calculation.

[0035] S5: Establish a comprehensive scoring prediction model, input the first risk score and the second risk score to obtain a comprehensive risk score, and generate a graded early warning strategy based on the comprehensive risk score and clinical rules; The comprehensive score prediction model is a nonlinear regression model with interaction terms. It is calculated by inputting a first risk score and a second risk score. Obtain a comprehensive risk score The tiered early warning strategy includes behavioral intervention-level early warning, enhanced observation-level early warning, clinical assessment-level early warning, and emergency response-level early warning. A behavioral intervention-level early warning is issued when a significant interfering behavior is detected and the overall risk score is below the Level 1 risk threshold. An enhanced observation-level early warning is issued when the overall risk score is above the Level 1 risk threshold but below the Level 2 risk threshold. A clinical assessment-level early warning is issued when the overall risk score is above the Level 2 risk threshold but below the Level 3 risk threshold. An emergency response-level early warning is issued when the overall risk score is above the Level 3 risk threshold.

[0036] It should also be noted that, based on the progressive relationship of risk assessment, this invention divides the early warning mechanism into four levels with clear operational directions.

[0037] When a significantly disruptive behavior affecting monitoring accuracy (such as inappropriate limb movement) is detected, but the overall risk is below the Level 1 risk threshold, Level 1, or behavioral intervention warning, will be activated. This level does not trigger clinical alerts; it primarily provides real-time behavioral correction guidance to patients or caregivers, aiming to eliminate risk triggers and serving as a preventative measure. The Level 1 risk threshold is determined by collecting comprehensive risk score data from historical cases of patients ultimately clinically assessed as low-risk or crisis-free. The distribution of this dataset is analyzed, and the 95th percentile is selected as the initial Level 1 risk threshold. Finally, this threshold is fine-tuned based on its warning accuracy on the test set and clinical expert opinions to balance sensitivity and specificity.

[0038] After effectively excluding behavioral interference, a Level 2 warning—enhanced observation—is issued when the comprehensive risk score reaches the Level 1 risk threshold but falls below the Level 2 risk threshold. The Level 2 risk threshold is determined by collecting comprehensive risk score data from historical cases of patients clinically assessed as requiring enhanced observation but not yet requiring emergency treatment. The distribution characteristics of this dataset are analyzed, and the 85th percentile is selected as the initial value for the Level 2 risk threshold. This threshold is then calibrated based on its warning performance on an independent validation set and clinical feedback to optimize the accuracy of intermediate risk identification.

[0039] When the comprehensive risk score reaches the Level 2 risk threshold but falls below the Level 3 risk threshold, Level 3, or clinical assessment warning, will be activated immediately. The Level 3 risk threshold is based on cases clinically diagnosed with severe vascular crisis requiring emergency treatment. The comprehensive risk score data is extracted and its typical range is analyzed. The lower quartile of this dataset or a risk threshold set according to clinical guidelines is used as the initial benchmark for the Level 3 risk threshold. Through retrospective analysis and prospective validation, it is ensured that this threshold can reliably trigger the highest level warning and meets the actual needs of clinical emergency treatment.

[0040] The highest level, Level 4, is an emergency response warning, triggered under a specific high-risk situation: that is, when the comprehensive risk score is higher than the Level 3 risk threshold.

[0041] Obtain the comprehensive risk score formula In the middle, weight Based on the historical case database, we collected the first risk score, second risk score, and corresponding clinical outcome labels of vascular crisis. Using a supervised learning algorithm with clinical outcome as the optimization objective, we trained the comprehensive score prediction model and fitted the optimal initial value.

[0042] S6: Record each warning event, clinical intervention, and final outcome to optimize the behavior recognition algorithm model, adjust warning thresholds, and implement tiered warning strategies.

[0043] When an alert is triggered, alert data is collected and processed. The processed alert data includes a graded alert strategy, clinical intervention measures, and the tissue oxygen saturation and the original temperature difference between the original and healthy sides after intervention. Based on the processed alert data, the behavior recognition model and the weight parameters of the influencing data are optimized, and the graded alert threshold and grading logic are adjusted.

[0044] It should also be noted that once the alert is triggered, a closed-loop data collection process will be initiated simultaneously. This process not only records the details of the specific grading strategy activated and the subsequent actual clinical interventions, but also continuously collects the raw skin temperature difference readings on the affected side after the intervention.

[0045] Based on this series of processed early warning data, the correlation between intervention records and changes in physiological parameters is analyzed to optimize the behavior recognition model and the weight parameters of the influencing data. The optimization process is based on the correspondence between historical early warning records and final clinical outcomes, constructing a labeled training dataset. The behavior recognition classifier is retrained using supervised learning methods, and the weight parameters of each influencing data are dynamically adjusted through gradient descent or Bayesian optimization. .

[0046] The tiered early warning thresholds and tiered logic are adjusted. This adjustment process involves periodically analyzing the trigger frequency of each level of warning, subsequent clinical validation results (true / false positives), and patient outcomes. Based on the statistical results and clinical expert consensus, grid search or optimization algorithms are used to fine-tune the risk thresholds at each level, and A / B testing is performed to validate the combinations of conditions in the tiered logic. The validated thresholds and logic are then updated to the system rule base, completing the closed-loop optimization.

[0047] Example 2: Based on Example 1, a multi-source data integration and hierarchical early warning system for vascular crises in hand surgery, such as... Figure 2 As shown, it also includes: The sensing and acquisition module synchronously collects tissue oxygen saturation, raw and healthy side temperature difference, behavioral environment data and clinical auxiliary data in the sentinel area through multi-source sensors; The behavior recognition module establishes a behavior recognition algorithm model, identifies interfering behaviors of patients in real time based on collected behavioral environment data, and obtains impact data by quantifying the interfering behaviors. The data correction and confidence assessment module, based on the impact data, dynamically corrects the original temperature difference between the original and healthy sides to obtain the corrected temperature difference between the original and healthy sides, dynamically corrects the tissue oxygen saturation to obtain the corrected tissue oxygen saturation, calculates a confidence score using the clinical auxiliary data, fuses the corrected temperature difference between the original and healthy sides with the confidence score to calculate a first risk score, and fuses the corrected tissue oxygen saturation with the confidence score to calculate a second risk score. The intelligent early warning module establishes a comprehensive scoring prediction model, inputs the first risk score and the second risk score to obtain a comprehensive risk score, generates a graded early warning strategy based on the comprehensive risk score and clinical rules, and records each early warning event, clinical intervention measures and final outcome, which is used to optimize the behavior recognition algorithm model, adjust the early warning threshold and graded early warning strategy.

[0048] Although the invention has been described with reference to exemplary embodiments, it should be understood that the invention is not limited to the disclosed exemplary embodiments. The scope of the following claims should be given the broadest interpretation so as to cover all variations and equivalent structures and functions.

Claims

1. A multi-source data integration and hierarchical early warning method for vascular crises in hand surgery, characterized in that, Includes the following steps: S1: Synchronously collect tissue oxygen saturation, raw and healthy side temperature difference, behavioral environmental data and clinical auxiliary data of the sentinel area through multi-source sensors; S2: Establish a behavior recognition algorithm model to identify the patient's interfering behaviors in real time based on the collected behavioral environment data, and obtain impact data by quantifying the interfering behaviors; S3: Based on the aforementioned impact data, the original temperature difference value between the healthy and unhealthy sides is dynamically corrected to obtain the corrected temperature difference value between the healthy and unhealthy sides, and the tissue oxygen saturation is dynamically corrected to obtain the corrected tissue oxygen saturation. S4: Calculate the confidence score using the clinical auxiliary data, perform a fusion analysis between the corrected temperature difference with the healthy side and the confidence score to calculate the first risk score, and perform a fusion analysis between the corrected tissue oxygen saturation and the confidence score to calculate the second risk score. S5: Establish a comprehensive scoring prediction model, input the first risk score and the second risk score to obtain a comprehensive risk score, and generate a graded early warning strategy based on the comprehensive risk score and clinical rules; S6: Record each warning event, clinical intervention, and final outcome to optimize the behavior recognition algorithm model, adjust warning thresholds, and implement tiered warning strategies.

2. The multi-source data integration and hierarchical early warning method for vascular crises in hand surgery according to claim 1, characterized in that, The method involves synchronously collecting tissue oxygen saturation, raw and healthy side temperature difference, behavioral environment data, and clinical auxiliary data in the sentinel area using multi-source sensors. The behavioral environment data includes three-dimensional posture data of the affected limb, local microenvironment temperature, radiant heat flux, patient habitual behavior data, ambient light intensity, and local blood flow change data. The clinical auxiliary data includes patient pain self-assessment data, medication data, and observation records from medical staff.

3. The multi-source data integration and hierarchical early warning method for vascular crises in hand surgery according to claim 1, characterized in that, The establishment of the behavior recognition algorithm model, which identifies interfering behaviors of patients in real time based on collected behavioral environment data, and obtains impact data by quantifying the interfering behaviors, includes: extracting features from the behavioral environment data according to the behavior recognition algorithm model to obtain thermal interference behavior data, postural habit interference data, and oxygenation interference data; and defining a thermal interference behavior set based on the thermal interference behavior data. Each thermal disturbance behavior Including the The local microenvironment temperature and radiative heat flux at the occurrence of secondary thermal disturbances; based on the aforementioned postural behavior disturbance data, the set of postural habit disturbance behaviors is defined as follows: Each body position habit interferes with behavior Including the Three-dimensional posture data of the affected limb and patient habitual behavior data at the time of occurrence of postural habit interference behavior; the oxygenation interference behavior set is defined based on the oxygenation interference data. Each oxygenation disturbance behavior Including the Data on changes in ambient light intensity and local blood flow during the occurrence of secondary oxygenation interference behaviors were used. Based on the acquired set of interference behaviors, a data model was employed for quantitative analysis to obtain data on the impact of interference behaviors on the temperature difference between the affected and healthy sides. The set of interference behaviors included thermal interference behaviors, postural habit interference behaviors, and oxygenation interference behaviors.

4. The multi-source data integration and hierarchical early warning method for vascular crises in hand surgery according to claim 3, characterized in that, The process involves quantitative analysis using a data model based on the acquired set of interfering behaviors to obtain data on the impact of these behaviors on the temperature difference between the affected and healthy sides. This includes: the data model comprising a thermodynamic model, a hemodynamic model, and an oxygenation kinetic model; the impact data comprising temperature impact data, postural habit impact data, and oxygenation interference impact data; the thermodynamic model taking the identified set of thermal interference behaviors as input, calculating using an algorithm based on the energy conservation-based heat balance equation, and outputting quantified temperature impact data; the hemodynamic model taking the identified set of postural habit interference behaviors as input, calculating using a gravity-blood flow coupling algorithm based on fluid dynamics and circulatory physiology, and outputting quantified postural habit impact data; and the oxygenation kinetic model taking the identified set of oxygenation interference behaviors as input, calculating using an algorithm based on a light propagation model and the blood flow-oxygenation coupling relationship, and outputting quantified oxygenation interference impact data.

5. The multi-source data integration and hierarchical early warning method for vascular crises in hand surgery according to claim 1, characterized in that, The step of dynamically correcting the original temperature difference with the healthy side to obtain a corrected temperature difference with the healthy side based on the influence data, and dynamically correcting the tissue oxygen saturation to obtain a corrected tissue oxygen saturation, includes: obtaining the corrected temperature difference with the healthy side using the original temperature difference with the healthy side correction formula based on the influence data, and obtaining the corrected tissue oxygen saturation using the tissue oxygen saturation correction formula based on the influence data, wherein the correction formula is: ； ； in This is the corrected temperature difference value between the healthy and unhealthy sides; This represents the original temperature difference between the healthy and unhealthy sides. For the first The data includes data on the effects of temperature and data on the effects of body position habits. For the first The weight parameters that affect the data by class; This is the corrected tissue oxygen saturation; Tissue oxygen saturation; Data affected by oxygenation interference; To adjust the parameters.

6. The multi-source data integration and hierarchical early warning method for vascular crises in hand surgery according to claim 1, characterized in that, The process of calculating a confidence score using the clinical auxiliary data, fusing the corrected temperature difference with the healthy side with the confidence score to calculate a first risk score, and fusing the corrected tissue oxygen saturation with the confidence score to calculate a second risk score, includes: calculating a confidence score based on the clinical auxiliary data using the clinical auxiliary data. Obtain confidence score ,in For normalized patient pain self-rating data, For normalized medication data, For the normalized observation records of medical staff, The non-linear weighting parameter is defined as: [Clinical standard deviation is defined as...] A first risk score is obtained using a first risk scoring formula based on the original temperature difference between the healthy and unhealthy sides, the confidence score, and the corrected temperature difference between the healthy and unhealthy sides; a second risk score is obtained using a second risk scoring formula based on the tissue oxygen saturation, the confidence score, and the corrected tissue oxygen saturation; wherein the first risk scoring formula is: ； in Assign the first risk score; This is the corrected temperature difference value between the healthy and unhealthy sides; This represents the original temperature difference between the healthy and unhealthy sides. Clinical standard deviation; Score the confidence level; It is a very small constant to prevent the denominator from being zero.

7. The multi-source data integration and hierarchical early warning method for vascular crises in hand surgery according to claim 6, characterized in that, The step of obtaining the second risk score based on the tissue oxygen saturation, confidence score, and corrected tissue oxygen saturation using a second risk scoring formula includes: defining the standard deviation of normal fluctuations in oxygen saturation as... The second risk score is obtained through a second risk scoring formula, which is as follows: ； in This is the second risk score; This is the corrected tissue oxygen saturation; Tissue oxygen saturation; This represents the standard deviation of normal fluctuations in oxygen saturation. Score the confidence level; It is a very small constant to prevent the denominator from being zero.

8. The multi-source data integration and hierarchical early warning method for vascular crises in hand surgery according to claim 1, characterized in that, The step of establishing a comprehensive risk score prediction model involves inputting the first risk score and the second risk score to obtain a comprehensive risk score, and generating a tiered early warning strategy based on the comprehensive risk score and clinical rules. This includes: the comprehensive risk score prediction model is a nonlinear regression model containing an interaction term; by inputting the first risk score and the second risk score, a tiered early warning strategy is generated. Obtain a comprehensive risk score The tiered early warning strategy includes behavioral intervention-level early warning, enhanced observation-level early warning, clinical assessment-level early warning, and emergency response-level early warning. A behavioral intervention-level early warning is issued when a significant interfering behavior is detected and the overall risk score is below the Level 1 risk threshold. An enhanced observation-level early warning is issued when the overall risk score is above the Level 1 risk threshold but below the Level 2 risk threshold. A clinical assessment-level early warning is issued when the overall risk score is above the Level 2 risk threshold but below the Level 3 risk threshold. An emergency response-level early warning is issued when the overall risk score is above the Level 3 risk threshold.

9. The multi-source data integration and hierarchical early warning method for vascular crises in hand surgery according to claim 1, characterized in that, The recording of each early warning event, clinical intervention, and final outcome is used to optimize the behavior recognition algorithm model, adjust the early warning threshold, and the graded early warning strategy. This includes: when an early warning is triggered, collecting and processing early warning data, which includes the graded early warning strategy, clinical intervention, and the tissue oxygen saturation and the original temperature difference with the healthy side after intervention; optimizing the behavior recognition model and the weight parameters of the influencing data based on the processed early warning data, and adjusting the graded early warning threshold and the graded logic.

10. A multi-source data integration and hierarchical early warning system for hand surgical vascular crises, used to implement the multi-source data integration and hierarchical early warning method for hand surgical vascular crises as described in any one of 1-9, characterized in that, Also includes: The sensing and acquisition module synchronously collects tissue oxygen saturation, raw and healthy side temperature difference, behavioral environment data and clinical auxiliary data in the sentinel area through multi-source sensors; The behavior recognition module establishes a behavior recognition algorithm model, identifies interfering behaviors of patients in real time based on collected behavioral environment data, and obtains impact data by quantifying the interfering behaviors. The data correction and confidence assessment module, based on the impact data, dynamically corrects the original temperature difference between the original and healthy sides to obtain the corrected temperature difference between the original and healthy sides, dynamically corrects the tissue oxygen saturation to obtain the corrected tissue oxygen saturation, calculates a confidence score using the clinical auxiliary data, fuses the corrected temperature difference between the original and healthy sides with the confidence score to calculate a first risk score, and fuses the corrected tissue oxygen saturation with the confidence score to calculate a second risk score. The intelligent early warning module establishes a comprehensive scoring prediction model, inputs the first risk score and the second risk score to obtain a comprehensive risk score, generates a graded early warning strategy based on the comprehensive risk score and clinical rules, and records each early warning event, clinical intervention measures and final outcome, which is used to optimize the behavior recognition algorithm model, adjust the early warning threshold and graded early warning strategy.

Images