A liver and gallbladder and pancreas surgery intraoperative blood vessel blocking risk early warning system

Abstract

The present application belongs to the technical field of medical artificial intelligence, and discloses a liver and gallbladder and pancreas surgery intraoperative blood vessel blocking risk early warning system, which comprises an intraoperative multi-source data acquisition module, a blood vessel blocking state construction module, an integrated agent risk prediction module, a meta-strategy adaptive control module, a blocking scheme candidate optimization module and an intraoperative early warning feedback updating module. The blood vessel blocking state feature, blood vessel blocking risk prediction result, risk uncertainty result, meta-strategy control result and target blocking auxiliary scheme are generated by using the principles of intraoperative multi-source data fusion, blood vessel blocking state feature construction, integrated agent risk prediction, risk uncertainty assessment, meta-strategy adaptive control and candidate blocking scheme optimization, so that the dynamic identification, stable prediction, accurate early warning and auxiliary scheme screening of the blood vessel blocking risk in liver and gallbladder and pancreas surgery are realized, and the problems of experience judgment, insufficient single index, fixed threshold false alarm and missing alarm and lack of dynamic optimization of the blocking scheme in the prior art are solved.

Description

Technical Field

[0001] This invention belongs to the field of medical artificial intelligence technology, and in particular relates to a risk warning system for intraoperative vascular occlusion in hepatobiliary and pancreatic surgery. Background Technology

[0002] In hepatobiliary and pancreatic surgery, the portal vein, hepatic artery, hepatic vein, inferior vena cava, and other important blood vessels around the pancreas are closely related to the location of the lesion, the extent of resection, and the control of bleeding in the surgical field. To ensure a clear surgical field and reduce the risk of intraoperative bleeding, surgeons often need to temporarily occlude, segmentally occlude, or intermittently open the target blood vessels. However, vascular occlusion can alter local tissue perfusion, liver ischemia tolerance, systemic circulatory stability, and postoperative recovery risk. If the timing of occlusion, the sequence of occlusion, the timing of release, or the low perfusion protection measures are not properly judged, it can easily lead to tissue ischemia and damage, aggravated circulatory fluctuations, failure to control intraoperative bleeding, and an increased risk of postoperative complications.

[0003] The existing technology has at least the following problems that need to be improved: (1) Current assessment of intraoperative vascular occlusion risk in hepatobiliary and pancreatic surgery mainly relies on surgeon experience, anesthesia monitoring data, and individual blood flow indicators. There is a lack of a mechanism to integrate preoperative baseline data, intraoperative vital signs data, intraoperative blood flow monitoring data, intraoperative imaging data, vascular occlusion operation data, and laboratory indicator data. This makes it difficult to reflect the patient's baseline risk, surgical anatomical conditions, residual perfusion status, circulatory fluctuations, and changes in tissue ischemia tolerance in a timely manner, resulting in an incomplete identification of vascular occlusion risk.

[0004] (2) Existing intraoperative early warning methods mostly use fixed threshold alarms or manual experience judgment. They lack an integrated proxy risk prediction mechanism that can maintain robustness in the early stage of surgery and improve prediction accuracy in the critical stage of early warning. They also lack a mechanism to adaptively select prediction models and candidate evaluation criteria based on the current surgical stage, risk uncertainty, number of candidate blocking options and intraoperative feedback status. As a result, the vascular blocking auxiliary plan is difficult to take into account bleeding control, tissue ischemia risk, surgeon operation continuity and real-time changes during surgery.

[0005] Therefore, there is a need for an intraoperative vascular occlusion risk warning system in hepatobiliary and pancreatic surgery to perform real-time analysis of multi-source data during surgery and provide dynamic risk warnings and auxiliary scheme selection for the vascular occlusion process. Summary of the Invention

[0006] To address the above issues and overcome the shortcomings of existing technologies, this invention provides an intraoperative vascular occlusion risk early warning system for hepatobiliary and pancreatic surgery. Utilizing principles of intraoperative multi-source data fusion, vascular occlusion status feature construction, integrated proxy risk prediction, risk uncertainty assessment, meta-strategy adaptive control, and candidate occlusion scheme optimization, the system generates vascular occlusion status features, vascular occlusion risk prediction results, risk uncertainty results, meta-strategy control results, and target occlusion auxiliary schemes. This enables dynamic identification, stable prediction, accurate early warning, and auxiliary scheme selection of intraoperative vascular occlusion risks in hepatobiliary and pancreatic surgery. It solves the problems of existing technologies relying on experience-based judgment, insufficient single indicators, false alarms and false negatives due to fixed thresholds, and lack of dynamic optimization of occlusion schemes.

[0007] This invention provides an intraoperative vascular occlusion risk early warning system for hepatobiliary and pancreatic surgery, including an intraoperative multi-source data acquisition module, a vascular occlusion status construction module, an integrated agent risk prediction module, a meta-strategy adaptive control module, an occlusion scheme candidate optimization module, and an intraoperative early warning feedback update module.

[0008] The intraoperative multi-source data acquisition module collects preoperative basic data, intraoperative vital signs data, intraoperative blood flow monitoring data, intraoperative imaging data, vascular occlusion operation data, and laboratory indicator data during hepatobiliary and pancreatic surgery, generates raw intraoperative vascular occlusion data, and sends the raw intraoperative vascular occlusion data to the vascular occlusion status construction module.

[0009] The vascular occlusion status construction module receives the original intraoperative vascular occlusion data, extracts the vascular occlusion status features based on the original intraoperative vascular occlusion data, and sends the vascular occlusion status features to the integrated agent risk prediction module, the meta-strategy adaptive control module, and the occlusion scheme candidate optimization module.

[0010] The integrated proxy risk prediction module receives vascular occlusion status characteristics, constructs an intraoperative vascular occlusion risk proxy model based on the vascular occlusion status characteristics, and generates vascular occlusion risk prediction results and risk uncertainty results based on the intraoperative vascular occlusion risk proxy model.

[0011] The meta-strategy adaptive control module receives vascular occlusion status characteristics, vascular occlusion risk prediction results, and risk uncertainty results, and generates meta-strategy control results based on the vascular occlusion status characteristics, vascular occlusion risk prediction results, and risk uncertainty results.

[0012] The blocking scheme candidate optimization module receives vascular blocking state characteristics and meta-policy control results, generates multiple candidate blocking schemes based on the vascular blocking state characteristics and meta-policy control results, and selects the target blocking auxiliary scheme from the multiple candidate blocking schemes.

[0013] The intraoperative early warning feedback update module receives the vascular occlusion risk prediction results, risk uncertainty results, and target occlusion auxiliary plan, generates a risk warning level, and generates vascular occlusion feedback samples based on the surgeon's confirmation results, anesthesia monitoring results, intraoperative outcome results, and postoperative short-term outcome results. Based on the vascular occlusion feedback samples, the integrated agent risk prediction module, the meta-strategy adaptive control module, and the occlusion plan candidate optimization module update the parameters.

[0014] Furthermore, the vascular occlusion status construction module extracts the patient's basic risk characteristics, hepatobiliary and pancreatic surgical anatomical characteristics, target vascular occlusion characteristics, residual perfusion characteristics, intraoperative circulatory fluctuation characteristics, and tissue ischemia tolerance characteristics based on the intraoperative vascular occlusion raw data, and binds the patient's basic risk characteristics, hepatobiliary and pancreatic surgical anatomical characteristics, target vascular occlusion characteristics, residual perfusion characteristics, intraoperative circulatory fluctuation characteristics, and tissue ischemia tolerance characteristics to generate vascular occlusion status characteristics.

[0015] Furthermore, the integrated surrogate risk prediction module constructs an intraoperative vascular occlusion risk surrogate model based on historical vascular occlusion samples. The intraoperative vascular occlusion risk surrogate model includes a robust integrated surrogate model and an accurate integrated surrogate model. The robust integrated surrogate model is trained by half-sampling based on historical vascular occlusion samples and then integrated and averaged. The accurate integrated surrogate model is trained by residual iteration based on historical vascular occlusion samples and then weighted and combined.

[0016] Furthermore, the integrated agent risk prediction module inputs the vascular occlusion status features into a robust integrated agent model or an exact integrated agent model to obtain risk probability distributions corresponding to multiple risk intervals; based on the risk probability distributions and the risk center value of each risk interval, it calculates the vascular occlusion risk prediction value; based on the prediction differences between multiple basic risk agent models, the degree of dispersion of the risk probability distribution, and the degree of missing vascular occlusion status features, it generates a risk uncertainty result; and it binds the vascular occlusion risk prediction value, the risk uncertainty result, the main risk contribution features, and the risk level to generate the vascular occlusion risk prediction result.

[0017] Furthermore, the meta-strategy adaptive control module generates intraoperative search state features based on the current surgical stage, vascular occlusion duration, vascular occlusion risk prediction value, risk uncertainty result, vascular occlusion status feature completeness, number of candidate occlusion schemes, recent risk reduction magnitude, and intraoperative feedback status. Based on the intraoperative search state features, it generates surrogate model selection results and candidate scheme evaluation criterion selection results, and binds the surrogate model selection results and candidate scheme evaluation criterion selection results to generate meta-strategy control results.

[0018] Furthermore, the candidate blocking scheme optimization module generates multiple candidate blocking schemes based on the vascular blocking state characteristics. The candidate blocking schemes include the target vascular blocking sequence, vascular blocking duration, intermittent opening time, segmented blocking method, low perfusion protection measures, and timing of blocking removal. Each candidate blocking scheme is combined with the vascular blocking state characteristics and input into the robust integrated surrogate model or precise integrated surrogate model specified by the meta-policy control result to obtain the candidate risk prediction results corresponding to the candidate blocking schemes.

[0019] Furthermore, the candidate blocking scheme optimization module constructs an objective function for candidate blocking schemes based on candidate risk prediction results, risk uncertainty results, surgical field bleeding control requirements, tissue ischemia tolerance characteristics, and surgeon's operational preferences; according to the candidate scheme evaluation criteria, it sorts and filters multiple candidate blocking schemes, and determines the target blocking auxiliary scheme from the filtered candidate blocking schemes.

[0020] Furthermore, the intraoperative early warning feedback update module generates low-risk alerts, medium-risk alerts, high-risk alerts, and emergency release alerts based on the predicted vascular occlusion risk value, risk uncertainty result, and objective function value of the target occlusion auxiliary plan. It outputs the risk warning level, main risk contribution characteristics, target occlusion auxiliary plan, recommended intermittent opening time, recommended timing for occlusion removal, and the status requiring manual confirmation to the intraoperative display terminal, anesthesia monitoring terminal, and surgical record terminal. It also binds the vascular occlusion status characteristics, vascular occlusion risk prediction result, risk uncertainty result, meta-strategy control result, candidate occlusion plan, target occlusion auxiliary plan, surgeon confirmation result, anesthesia monitoring result, intraoperative outcome result, and postoperative short-term outcome result to generate vascular occlusion feedback samples.

[0021] The beneficial effects of this invention are as follows: (1) Utilizing the principle of intraoperative multi-source data fusion and vascular occlusion status feature construction, the intraoperative multi-source data acquisition module collects preoperative basic data, intraoperative vital signs data, intraoperative blood flow monitoring data, intraoperative imaging data, vascular occlusion operation data and laboratory indicator data. Then, the vascular occlusion status construction module extracts the patient's basic risk characteristics, hepatobiliary and pancreatic surgical anatomical characteristics, target vascular occlusion characteristics, residual perfusion characteristics, intraoperative circulatory fluctuation characteristics and tissue ischemia tolerance characteristics. The technical effect is to form vascular occlusion status features that can reflect the entire process of vascular occlusion, which solves the problem that the existing intraoperative judgment relies on a single monitoring indicator and cannot fully characterize the source of vascular occlusion risk.

[0022] (2) By utilizing the robust integrated surrogate model, the precise integrated surrogate model and the principle of risk uncertainty assessment, the integrated surrogate risk prediction module constructs an intraoperative vascular occlusion risk surrogate model based on historical vascular occlusion samples, and generates vascular occlusion risk prediction results and risk uncertainty results based on the characteristics of vascular occlusion status. The technical effect is to maintain prediction stability when the intraoperative sample noise is high and the status characteristics are incomplete, and to improve the prediction precision when the vascular occlusion risk is close to the warning threshold. This solves the problem that the existing fixed threshold alarm is prone to false alarms and false alarms and cannot adapt to dynamic changes during the operation.

[0023] (3) By utilizing the principles of meta-strategy adaptive control and candidate blocking scheme optimization, the meta-strategy adaptive control module generates meta-strategy control results based on the current surgical stage, vascular blocking duration, vascular blocking risk prediction value, risk uncertainty result and intraoperative feedback status. Then, the blocking scheme candidate optimization module generates multiple candidate blocking schemes and selects the target blocking auxiliary scheme. The technical effect is to form a dynamic balance between bleeding control requirements, tissue ischemia tolerance, risk uncertainty and surgeon's operational preferences, which solves the problems of existing vascular blocking auxiliary decision-making lacking candidate scheme comparison and difficulty in balancing safety and surgical continuity. Attached Figure Description

[0024] The accompanying drawings are provided to further understand the present invention and form part of the specification. They are used together with the embodiments of the present invention to explain the invention and do not constitute a limitation thereof.

[0025] Figure 1 A flowchart illustrating the construction of vascular occlusion state features proposed in this invention; Figure 2 This is a flowchart of the integrated agent risk prediction process proposed in this invention; Figure 3 This is a flowchart of the candidate optimization and early warning feedback process for the blocking scheme proposed in this invention. Detailed Implementation

[0026] Example 1, see Figures 1-3 The present invention provides an intraoperative vascular occlusion risk early warning system for hepatobiliary and pancreatic surgery, comprising an intraoperative multi-source data acquisition module, a vascular occlusion status construction module, an integrated agent risk prediction module, a meta-strategy adaptive control module, an occlusion scheme candidate optimization module, and an intraoperative early warning feedback update module. The intraoperative multi-source data acquisition module collects preoperative basic data, intraoperative vital signs data, intraoperative blood flow monitoring data, intraoperative imaging data, vascular occlusion operation data, and laboratory indicator data during hepatobiliary and pancreatic surgery, generates intraoperative vascular occlusion raw data, and sends the intraoperative vascular occlusion raw data to the vascular occlusion status construction module. The vascular occlusion status construction module receives the original intraoperative vascular occlusion data, extracts the patient's basic risk characteristics, hepatobiliary and pancreatic surgical anatomical characteristics, target vascular occlusion characteristics, residual perfusion characteristics, intraoperative circulatory fluctuation characteristics, and tissue ischemia tolerance characteristics based on the original intraoperative vascular occlusion data, and binds the patient's basic risk characteristics, hepatobiliary and pancreatic surgical anatomical characteristics, target vascular occlusion characteristics, residual perfusion characteristics, intraoperative circulatory fluctuation characteristics, and tissue ischemia tolerance characteristics to generate vascular occlusion status characteristics; The integrated surrogate risk prediction module receives vascular occlusion status features and constructs an intraoperative vascular occlusion risk surrogate model based on these features. The intraoperative vascular occlusion risk surrogate model includes a robust surrogate model and a precise surrogate model. The robust surrogate model is trained using a semi-sampling method based on historical vascular occlusion samples and then ensembled and averaged. The precise surrogate model is trained using residual iterative methods based on historical vascular occlusion samples and then weighted and combined. The integrated surrogate risk prediction module generates vascular occlusion risk prediction results and risk uncertainty results based on the robust surrogate model and the precise surrogate model. The meta-strategy adaptive control module receives vascular occlusion status characteristics, vascular occlusion risk prediction results, and risk uncertainty results. Based on the current surgical stage, vascular occlusion duration, risk uncertainty results, number of candidate occlusion options, and intraoperative feedback status, it generates meta-strategy control results. The meta-strategy control results include surrogate model selection results and candidate option evaluation criterion selection results. The blocking scheme candidate optimization module receives vascular occlusion status characteristics and meta-strategy control results, and generates multiple candidate blocking schemes based on the meta-strategy control results. The candidate blocking schemes include target vascular occlusion sequence, vascular occlusion duration, intermittent opening time, segmented occlusion method, low perfusion protection measures, and timing of occlusion removal. The blocking scheme candidate optimization module selects the target blocking auxiliary scheme from the multiple candidate blocking schemes based on the candidate scheme evaluation criteria. The intraoperative early warning feedback update module receives the vascular occlusion risk prediction results, risk uncertainty results, and target occlusion auxiliary plans. Based on the vascular occlusion risk prediction results, it generates low-risk prompts, medium-risk warnings, high-risk warnings, and emergency release prompts. It also receives surgeon confirmation results, anesthesia monitoring results, intraoperative bleeding results, tissue ischemia-reperfusion results, and postoperative short-term outcome results, generates vascular occlusion feedback samples, and updates the parameters of the integrated agent risk prediction module, the meta-strategy adaptive control module, and the occlusion plan candidate optimization module based on the vascular occlusion feedback samples.

[0027] Through the cooperation of the above modules, this system combines multi-source monitoring during hepatobiliary and pancreatic surgery, vascular occlusion status construction, integrated agent risk prediction, meta-strategy adaptive control, candidate occlusion scheme optimization, and intraoperative feedback updates, so that vascular occlusion risk warning is transformed from a single threshold alarm into a dynamic risk warning process based on patient status, anatomical conditions, occlusion method, and intraoperative feedback.

[0028] Example 2: This example is based on all the above examples. The intraoperative multi-source data acquisition module specifically includes: Preoperative basic data collection; Intraoperative multi-source data collection module collects patient age, body mass index, liver function classification, underlying liver disease status, coagulation function, bilirubin level, albumin level, platelet count, previous surgical records, tumor location, tumor diameter and expected resection range, and generates preoperative basic data; Intraoperative vital signs data acquisition; the intraoperative multi-source data acquisition module collects heart rate, mean arterial pressure, central venous pressure, blood oxygen saturation, end-tidal carbon dioxide, urine output, body temperature and anesthetic medication status, and generates intraoperative vital signs data; Intraoperative blood flow monitoring data acquisition; the intraoperative multi-source data acquisition module acquires portal vein blood flow velocity, hepatic artery blood flow velocity, hepatic vein return status, inferior vena cava return status, residual blood flow in the target vessel, collateral circulation blood flow, and local tissue perfusion signals to generate intraoperative blood flow monitoring data; Intraoperative imaging data acquisition; the intraoperative multi-source data acquisition module acquires intraoperative ultrasound images, fluorescence perfusion images, laparoscopic images, and surgical field images to generate intraoperative imaging data; Vascular occlusion operation data acquisition; the intraoperative multi-source data acquisition module collects the target vessel name, vascular occlusion start time, vascular occlusion end time, cumulative occlusion time, single occlusion time, occlusion clamping intensity, occlusion sequence, intermittent opening time and occlusion release time, and generates vascular occlusion operation data; Laboratory indicator data acquisition; the intraoperative multi-source data acquisition module collects blood lactate, hemoglobin, prothrombin time, activated clotting time, blood gas acid-base status, electrolyte status and intraoperative rapid liver function indicators, and generates laboratory indicator data; Intraoperative vascular occlusion raw data generation: The intraoperative multi-source data acquisition module binds preoperative basic data, intraoperative vital signs data, intraoperative blood flow monitoring data, intraoperative imaging data, vascular occlusion operation data, and laboratory indicator data according to the surgical stage number, timestamp, and target vessel number to generate intraoperative vascular occlusion raw data.

[0029] Example 3: This example is based on all the above examples. The blood vessel occlusion state construction module specifically includes: Intraoperative vessel occlusion raw data reception; the vessel occlusion status construction module receives the intraoperative vessel occlusion raw data generated by the intraoperative multi-source data acquisition module; Patient baseline risk characteristics extraction; the vascular occlusion status construction module extracts liver function reserve risk, coagulation abnormality risk, baseline circulatory tolerance risk, previous surgical adhesion risk, and expected resection range risk based on preoperative baseline data, generating patient baseline risk characteristics; Anatomical features of hepatobiliary and pancreatic surgery are extracted. The vascular occlusion status construction module extracts features of gate vein involvement, hepatic artery involvement, hepatic vein involvement, inferior vena cava proximity, bile duct proximity, and pancreatic posterior vessel proximity based on intraoperative imaging data and target vessel number, generating anatomical features of hepatobiliary and pancreatic surgery. Target vessel occlusion feature extraction; The vessel occlusion status construction module extracts the occluded vessel type, cumulative occlusion time, single occlusion time, occlusion sequence, intermittent opening time, occlusion clamping strength, and occlusion release time from the vessel occlusion operation data to generate target vessel occlusion features; Residual perfusion feature extraction; The vascular occlusion status construction module extracts residual perfusion of the outer vein, residual perfusion of the hepatic artery, collateral circulation compensation, fluorescence perfusion delay, the magnitude of decrease in local tissue perfusion, and the degree of restriction of venous return based on intraoperative blood flow monitoring data and intraoperative imaging data, and generates residual perfusion features; Intraoperative circulatory fluctuation feature extraction; The vascular occlusion status construction module extracts the mean arterial pressure decrease, central venous pressure change, heart rate fluctuation, blood lactate increase, urine output decrease, and acid-base status shift based on intraoperative vital signs data and laboratory indicators, generating intraoperative circulatory fluctuation features; Tissue ischemia tolerance feature extraction; The vascular occlusion status construction module extracts tissue ischemia tolerance time, reperfusion recovery trend, low perfusion duration and ischemia cumulative load based on the patient's basic risk characteristics, target vascular occlusion characteristics and residual perfusion characteristics, and generates tissue ischemia tolerance features; The vascular occlusion status feature generation module binds the patient's basic risk characteristics, surgical anatomical characteristics of the hepatobiliary and pancreatic surgery, target vascular occlusion characteristics, residual perfusion characteristics, intraoperative circulatory fluctuation characteristics, and tissue ischemia tolerance characteristics to generate vascular occlusion status features.

[0030] Example 4: This example is based on all the above examples. The integrated agent risk prediction module specifically includes: The integrated agent risk prediction module receives the vascular occlusion status features generated by the vascular occlusion status construction module. Construction of historical vascular occlusion samples: The integrated agent risk prediction module receives the characteristics of vascular occlusion status, post-occlusion circulatory fluctuation results, tissue ischemia and injury results, intraoperative bleeding results, occlusion release results, and postoperative short-term outcome results from historical hepatobiliary and pancreatic surgeries, and binds the above data to generate historical vascular occlusion samples; Robust integrated surrogate model construction; the integrated surrogate risk prediction module performs semi-sampling processing based on historical vascular occlusion samples to generate multiple vascular occlusion training subsets, and trains a basic risk surrogate model on each vascular occlusion training subset; the integrated surrogate risk prediction module performs ensemble averaging based on the prediction results of multiple basic risk surrogate models to generate a robust integrated surrogate model; the robust integrated surrogate model outputs robust risk prediction results under conditions of insufficient intraoperative sample size, high monitoring noise, and early surgical stages; The precise integrated surrogate model is constructed as follows: the integrated surrogate risk prediction module trains an initial basic risk surrogate model based on historical samples of vascular occlusion, generates residual samples based on the prediction error of the initial basic risk surrogate model, and iteratively trains subsequent basic risk surrogate models; the integrated surrogate risk prediction module combines multiple basic risk surrogate models with weights to generate a precise integrated surrogate model; the precise integrated surrogate model outputs precise risk prediction results when vascular occlusion data becomes gradually complete, trend characteristics are obvious, and the warning critical stage is reached; Risk prediction distribution generation; The integrated agent risk prediction module inputs the vascular occlusion status features into the robust integrated agent model or the precise integrated agent model to obtain the risk probability distribution corresponding to multiple risk intervals; Risk expectation value calculation: The integrated agent risk prediction module calculates the predicted risk value of vascular occlusion based on the risk probability distribution and the risk center value of each risk interval. The calculation formula is as follows:

[0031] in, It is a predictive value for the risk of vascular occlusion; It refers to the number of risk zones; It is the first The risk probability corresponding to each risk interval; It is the first The risk center value corresponding to each risk interval; Risk uncertainty results are generated; the integrated agent risk prediction module generates risk uncertainty results based on the prediction differences between multiple basic risk agent models, the degree of dispersion of risk probability distribution, and the degree of missing features of vascular occlusion status. The vascular occlusion risk prediction result is generated; the integrated agent risk prediction module binds the vascular occlusion risk prediction value, risk uncertainty result, main risk contribution characteristics and risk level to generate the vascular occlusion risk prediction result.

[0032] Regarding parameter adjustments: Step 1: Adjust the number of robust integrated surrogate models; when the intraoperative monitoring noise increases or the proportion of missing vascular occlusion status features increases, increase the number of basic risk surrogate models to improve the stability of the robust integrated surrogate models; Step 2: Precise integration of surrogate model weight adjustment; when the duration of vascular occlusion is prolonged and the risk trend continues to rise, increase the combined weight of the subsequent basic risk surrogate model so that the prediction results pay more attention to the high-risk changes after residual correction. Step 3: Adjusting the number of risk intervals; when the intraoperative period is in a low-risk and stable phase, reduce the number of risk intervals to increase the calculation speed; when the intraoperative period is in a critical warning phase, increase the number of risk intervals to improve the ability to subdivide risk prediction. Step 4: Adjust the risk uncertainty threshold; when the risk uncertainty result is higher than the uncertainty threshold, mark the vascular occlusion risk prediction result as requiring manual confirmation and send it to the intraoperative early warning feedback update module.

[0033] Through the above processing, the integrated agent risk prediction module can maintain prediction robustness in the early stage of surgery and improve prediction accuracy in the high-risk critical stage during surgery, thereby adapting to the characteristics of rapid changes in vascular occlusion risk, high sample noise, and delayed feedback of actual outcomes in hepatobiliary and pancreatic surgery.

[0034] Example 5: This example is based on all the above examples. The meta-policy adaptive control module specifically includes: The vascular occlusion status feature is received; the meta-strategy adaptive control module receives the vascular occlusion status feature generated by the vascular occlusion status construction module. The module receives the vascular occlusion risk prediction results; the meta-strategy adaptive control module receives the vascular occlusion risk prediction results generated by the integrated agent risk prediction module. Risk uncertainty results reception; the meta-strategy adaptive control module receives the risk uncertainty results generated by the integrated agent risk prediction module; Intraoperative search state feature generation: The meta-strategy adaptive control module generates intraoperative search state features based on the current surgical stage, duration of vascular occlusion, predicted vascular occlusion risk, risk uncertainty results, completeness of vascular occlusion state features, number of candidate occlusion schemes, recent risk reduction magnitude, and intraoperative feedback status. The proxy model selection result is generated; the meta-policy adaptive control module generates the proxy model selection result based on the intraoperative search state characteristics; when the integrity of the vascular occlusion state characteristics is low, the risk uncertainty result is high, or it is in the early stage of surgery, the proxy model selection result is a robust integrated proxy model; when the integrity of the vascular occlusion state characteristics is high, the risk uncertainty result is low, and the vascular occlusion risk prediction value is close to the risk threshold, the proxy model selection result is an exact integrated proxy model. The candidate protocol evaluation criteria selection results are generated; the meta-strategy adaptive control module generates the candidate protocol evaluation criteria selection results based on the intraoperative search state characteristics; when the risk uncertainty result is high, the candidate protocol evaluation criteria selection results are biased towards exploratory evaluation criteria; when the predicted risk value of vascular occlusion is close to the risk threshold, the candidate protocol evaluation criteria selection results are biased towards convergent evaluation criteria; when the risk difference between candidate occlusion protocols is small and the safety margin is insufficient, the candidate protocol evaluation criteria selection results are biased towards diversity maintenance evaluation criteria. Meta-policy control results are generated; the meta-policy adaptive control module binds the agent model selection results and the candidate scheme evaluation criterion selection results to generate meta-policy control results, and sends the meta-policy control results to the integrated agent risk prediction module and the blocking scheme candidate optimization module.

[0035] Regarding parameter adjustments: Step 1: Adjust the model switching threshold; when the intraoperative early warning feedback update module reports a large number of false alarms for the robust integrated proxy model, lower the switching threshold for the precise integrated proxy model; when the precise integrated proxy model fluctuates significantly in high-noise scenarios, increase the switching threshold for the precise integrated proxy model. Step 2: Adjusting the weight of exploratory evaluation criteria; when the number of candidate blocking solutions is insufficient or the risk uncertainty results increase, the weight of exploratory evaluation criteria is increased; Step 3: Adjusting the weights of convergent evaluation criteria; when the predicted risk value of vascular occlusion continues to approach the risk threshold, increase the weights of convergent evaluation criteria; Step 4: Adjust the weight of diversity maintenance evaluation criteria; when multiple candidate blocking schemes have similar risks and none of them meet the safety margin, increase the weight of diversity maintenance evaluation criteria to avoid candidate schemes from concentrating on a single blocking method too early.

[0036] Through the above processing, the meta-policy adaptive control module can dynamically select a robust integrated surrogate model or an exact integrated surrogate model according to the intraoperative state, and simultaneously select candidate scheme evaluation criteria, so that the system can balance prediction stability, prediction accuracy, scheme exploration ability and scheme convergence ability at different stages of the operation.

[0037] Example 6: This example is based on all the above examples. The blocking scheme candidate optimization module specifically includes: The vascular occlusion status feature is received; the occlusion scheme candidate optimization module receives the vascular occlusion status features generated by the vascular occlusion status construction module. Meta-policy control result reception; Blocking scheme candidate optimization module receives meta-policy control results generated by meta-policy adaptive control module; Candidate blocking scheme generation; The blocking scheme candidate optimization module generates multiple candidate blocking schemes based on the characteristics of the vascular blocking state. The candidate blocking schemes include the target vascular blocking sequence, vascular blocking duration, intermittent opening time, segmented blocking method, low perfusion protection measures, and timing of blocking removal. Risk prediction of candidate blocking schemes; The blocking scheme candidate optimization module combines each candidate blocking scheme with the characteristics of the vascular blocking state, and inputs the robust integrated surrogate model or the precise integrated surrogate model specified by the meta-policy control result to obtain the candidate risk prediction results corresponding to the candidate blocking scheme; Objective function construction for candidate blocking strategies: The candidate blocking strategy optimization module constructs the objective function for candidate blocking strategies based on candidate risk prediction results, risk uncertainty results, surgical field bleeding control requirements, tissue ischemia tolerance characteristics, and surgeon's operational preferences. The calculation formula is as follows:

[0038] in, It is the objective function value of the candidate blocking scheme; It is a predictive value for the risk of vascular occlusion; It is a result of risk and uncertainty; It is the cumulative burden of tissue ischemia; It is a benefit of controlling bleeding in the surgical field; It is the deviation of the surgeon's operational preferences; , , , and These are the corresponding scheme weight coefficients; Candidate blocking scheme screening; the blocking scheme candidate optimization module sorts and filters multiple candidate blocking schemes based on the candidate scheme evaluation criteria selection results; when the candidate scheme evaluation criteria selection result is an exploratory evaluation criterion, priority is given to retaining candidate blocking schemes with large risk differences and reduced uncertainty; when the candidate scheme evaluation criteria selection result is a convergent evaluation criterion, priority is given to retaining candidate blocking schemes with lower objective function values; when the candidate scheme evaluation criteria selection result is a diversity preservation evaluation criterion, candidate blocking schemes with different target vessel blocking sequences, intermittent opening times, and timing of blocking removal are retained. The target blocking auxiliary plan is generated; the blocking plan candidate optimization module determines the target blocking auxiliary plan from the screened candidate blocking plans and sends the target blocking auxiliary plan to the intraoperative early warning feedback update module.

[0039] Regarding parameter adjustments: Step 1: Adjusting the weight of cumulative tissue ischemia load; when the increase in blood lactate, the decrease in local tissue perfusion, and the duration of low perfusion all increase simultaneously, increase the weight corresponding to the cumulative tissue ischemia load. Step 2: Adjust the weight of surgical field bleeding control benefits; when the intraoperative bleeding volume increases and the bleeding source is close to the target vessel, increase the corresponding weight of surgical field bleeding control benefits; Step 3: Adjusting the weight of the surgeon's operational preference deviation; when the target blocking auxiliary plan differs significantly from the surgeon's current operating path, increase the weight corresponding to the surgeon's operational preference deviation to avoid the suggested plan affecting the continuity of the surgery; Step 4: Adjust the number of candidate blocking strategies; when the risk uncertainty is high and the candidate strategy evaluation criterion is an exploratory evaluation criterion, increase the number of candidate blocking strategies; when the predicted risk value of vascular blocking is in a low-risk stable stage, decrease the number of candidate blocking strategies.

[0040] Through the above processing, the candidate optimization module for the blocking scheme can achieve a balance between bleeding control, tissue ischemia, risk uncertainty and surgeon continuity, and output a target blocking auxiliary scheme suitable for the current hepatobiliary and pancreatic surgical status.

[0041] Example 7: This example is based on all the above examples. The intraoperative early warning feedback update module specifically includes: The intraoperative early warning feedback update module receives the vascular occlusion risk prediction results generated by the integrated agent risk prediction module. The target blocking auxiliary plan is received; the intraoperative early warning feedback update module receives the target blocking auxiliary plan generated by the blocking plan candidate optimization module. Risk warning level generation; the intraoperative warning feedback update module generates low-risk alerts, medium-risk warnings, high-risk warnings, and emergency release of blockage alerts based on the predicted risk value of vascular occlusion, the result of risk uncertainty, and the objective function value of the target occlusion auxiliary plan. Intraoperative early warning information output; The intraoperative early warning feedback update module outputs the risk warning level, main risk contribution characteristics, target blockade auxiliary plan, recommended intermittent opening time, recommended timing for removal of blockade and status requiring manual confirmation to the intraoperative display terminal, anesthesia monitoring terminal and surgical record terminal; The surgeon confirms receipt of the results; the intraoperative early warning feedback update module receives information from the surgeon regarding the adoption, postponement, modification, and rejection of the target blocking auxiliary plan, and generates the surgeon confirmation result. The anesthesia monitoring results are received; the intraoperative early warning feedback update module receives the circulatory stability assessment, blood pressure intervention record, volume management record and blood gas retest results input from the anesthesia terminal, and generates the anesthesia monitoring results. Intraoperative outcome results reception; the intraoperative early warning feedback update module receives intraoperative bleeding results, tissue ischemia-reperfusion results, blockage release results, and postoperative short-term outcome results; The vascular occlusion feedback sample is generated. The intraoperative early warning feedback update module binds the vascular occlusion status characteristics, vascular occlusion risk prediction results, risk uncertainty results, meta-strategy control results, candidate occlusion schemes, target occlusion auxiliary schemes, surgeon confirmation results, anesthesia monitoring results, intraoperative outcome results, and postoperative short-term outcome results to generate vascular occlusion feedback samples. Parameter updates; the intraoperative early warning feedback update module updates the parameters of the integrated agent risk prediction module, the meta-strategy adaptive control module, and the blockade scheme candidate optimization module based on the vascular occlusion feedback samples.

[0042] Through the above processing, the intraoperative early warning feedback update module can bind intraoperative risk prediction, blocking plan suggestions, surgeon confirmation, anesthesia feedback and actual outcome in a closed loop, so that the system can continuously correct the vascular blocking risk prediction and candidate blocking plan screening process in consecutive surgical cases.

Claims

1. A risk warning system for intraoperative vascular occlusion in hepatobiliary and pancreatic surgery, characterized in that: It includes an intraoperative multi-source data acquisition module, a vessel occlusion status construction module, an integrated agent risk prediction module, a meta-strategy adaptive control module, an occlusion scheme candidate optimization module, and an intraoperative early warning feedback update module; The intraoperative multi-source data acquisition module collects preoperative basic data, intraoperative vital signs data, intraoperative blood flow monitoring data, intraoperative imaging data, vascular occlusion operation data, and laboratory indicator data during hepatobiliary and pancreatic surgery, generates intraoperative vascular occlusion raw data, and sends the intraoperative vascular occlusion raw data to the vascular occlusion status construction module. The vascular occlusion status construction module receives the original intraoperative vascular occlusion data, extracts the vascular occlusion status features based on the original intraoperative vascular occlusion data, and sends the vascular occlusion status features to the integrated agent risk prediction module, the meta-strategy adaptive control module, and the occlusion scheme candidate optimization module. The integrated proxy risk prediction module receives vascular occlusion status characteristics, constructs an intraoperative vascular occlusion risk proxy model based on the vascular occlusion status characteristics, and generates vascular occlusion risk prediction results and risk uncertainty results based on the intraoperative vascular occlusion risk proxy model. The meta-strategy adaptive control module receives vascular occlusion status characteristics, vascular occlusion risk prediction results, and risk uncertainty results, and generates meta-strategy control results based on the vascular occlusion status characteristics, vascular occlusion risk prediction results, and risk uncertainty results. The blocking scheme candidate optimization module receives vascular blocking state characteristics and meta-policy control results, generates multiple candidate blocking schemes based on the vascular blocking state characteristics and meta-policy control results, and selects the target blocking auxiliary scheme from the multiple candidate blocking schemes. The intraoperative early warning feedback update module receives the vascular occlusion risk prediction results, risk uncertainty results, and target occlusion auxiliary plan, generates a risk warning level, and generates vascular occlusion feedback samples based on the surgeon's confirmation results, anesthesia monitoring results, intraoperative outcome results, and postoperative short-term outcome results. Based on the vascular occlusion feedback samples, the integrated agent risk prediction module, the meta-strategy adaptive control module, and the occlusion plan candidate optimization module update the parameters.

2. The intraoperative vascular occlusion risk warning system for hepatobiliary and pancreatic surgery according to claim 1, characterized in that: The vascular occlusion status construction module extracts the patient's basic risk characteristics, hepatobiliary and pancreatic surgical anatomical characteristics, target vascular occlusion characteristics, residual perfusion characteristics, intraoperative circulatory fluctuation characteristics, and tissue ischemia tolerance characteristics from the intraoperative vascular occlusion raw data, and binds the patient's basic risk characteristics, hepatobiliary and pancreatic surgical anatomical characteristics, target vascular occlusion characteristics, residual perfusion characteristics, intraoperative circulatory fluctuation characteristics, and tissue ischemia tolerance characteristics to generate vascular occlusion status characteristics.

3. The intraoperative vascular occlusion risk warning system for hepatobiliary and pancreatic surgery according to claim 2, characterized in that: The integrated surrogate risk prediction module constructs an intraoperative vascular occlusion risk surrogate model based on historical vascular occlusion samples. The intraoperative vascular occlusion risk surrogate model includes a robust integrated surrogate model and an accurate integrated surrogate model. The robust integrated surrogate model is trained by half-sampling based on historical vascular occlusion samples and then integrated and averaged. The accurate integrated surrogate model is trained by residual iteration based on historical vascular occlusion samples and then weighted and combined.

4. The intraoperative vascular occlusion risk warning system for hepatobiliary and pancreatic surgery according to claim 3, characterized in that: The integrated agent risk prediction module inputs the vascular occlusion status features into a robust integrated agent model or an exact integrated agent model to obtain the risk probability distribution corresponding to multiple risk intervals; based on the risk probability distribution and the risk center value of each risk interval, the predicted value of vascular occlusion risk is calculated. Based on the prediction differences among multiple basic risk proxy models, the degree of dispersion of risk probability distribution, and the degree of missing characteristics of vascular occlusion status, risk uncertainty results are generated; and the predicted value of vascular occlusion risk, risk uncertainty results, main risk contribution characteristics, and risk level are bound together to generate vascular occlusion risk prediction results.

5. The intraoperative vascular occlusion risk warning system for hepatobiliary and pancreatic surgery according to claim 4, characterized in that: The meta-strategy adaptive control module generates intraoperative search state features based on the current surgical stage, vascular occlusion duration, vascular occlusion risk prediction value, risk uncertainty result, vascular occlusion status feature completeness, number of candidate occlusion schemes, recent risk reduction magnitude, and intraoperative feedback status. Based on the intraoperative search state characteristics, the surrogate model selection results and the candidate solution evaluation criterion selection results are generated, and the surrogate model selection results and the candidate solution evaluation criterion selection results are bound together to generate the meta-strategy control results.

6. The intraoperative vascular occlusion risk warning system for hepatobiliary and pancreatic surgery according to claim 5, characterized in that: The candidate blocking scheme optimization module generates multiple candidate blocking schemes based on the characteristics of the vascular blocking state. The candidate blocking schemes include the target vascular blocking sequence, vascular blocking duration, intermittent opening time, segmented blocking method, low perfusion protection measures, and timing of blocking removal. Each candidate blocking scheme is combined with the characteristics of the vascular blocking state and input into the robust integrated surrogate model or precise integrated surrogate model specified by the meta-policy control result to obtain the candidate risk prediction results corresponding to the candidate blocking schemes.

7. The intraoperative vascular occlusion risk warning system for hepatobiliary and pancreatic surgery according to claim 6, characterized in that: The candidate blocking scheme optimization module constructs an objective function for candidate blocking schemes based on candidate risk prediction results, risk uncertainty results, surgical field bleeding control requirements, tissue ischemia tolerance characteristics, and surgeon's operational preferences. Based on the candidate scheme evaluation criteria, multiple candidate blocking schemes are sorted and screened, and the target blocking auxiliary scheme is determined from the screened candidate blocking schemes.

8. The intraoperative vascular occlusion risk warning system for hepatobiliary and pancreatic surgery according to claim 7, characterized in that: The intraoperative early warning feedback update module generates low-risk alerts, medium-risk alerts, high-risk alerts, and emergency release alerts based on the predicted risk value of vascular occlusion, the result of risk uncertainty, and the objective function value of the target occlusion auxiliary plan. The risk warning level, main risk contribution characteristics, target occlusion auxiliary plan, recommended intermittent opening time, recommended timing for occlusion removal, and status requiring manual confirmation are output to the intraoperative display terminal, anesthesia monitoring terminal, and surgical record terminal. The vascular occlusion status characteristics, vascular occlusion risk prediction results, risk uncertainty results, meta-strategy control results, candidate occlusion plans, target occlusion auxiliary plans, surgeon confirmation results, anesthesia monitoring results, intraoperative outcome results, and postoperative short-term outcome results are bound together to generate vascular occlusion feedback samples.

Images