CN-122004800-A - Method and system for monitoring postoperative bleeding risk of hepatobiliary patient

CN122004800ACN 122004800 ACN122004800 ACN 122004800ACN-122004800-A

Abstract

The invention provides a method and a system for monitoring postoperative bleeding risk of a liver and gall patient. The method comprises the steps of constructing an LSTM-CNN mixed model by using time-frequency decomposition characteristics to obtain local tissue hypoxia index and vasomotor function abnormality probability, fusing low-frequency impedance change rate and albumin level by using a random forest algorithm to obtain ascites occurrence probability and a hydrops volume predicted value, and generating bleeding point positioning coordinates and thermodynamic diagram risk classification. And calculating the comprehensive bleeding risk probability and the bleeding area positioning. Generating a hierarchical early warning signal and recommending a personalized treatment regimen or care advice. According to the invention, different data features are extracted by adopting an LSTM-CNN mixed model, a random forest algorithm and a three-dimensional convolution neural network in a depth mode respectively, so that single index evaluation limitation is avoided. And establishing a causality model through a Bayesian network, improving the calculation accuracy of the comprehensive risk probability, and dynamically adjusting the early warning threshold value by combining individual characteristics of patients to solve the problem of easy misjudgment in the prior art.

Inventors

JIAN HONGMEI
Zheng Dengye
WU YING
JIN SHENGNAN
MAO XUEYUAN

Assignees

中国人民解放军陆军军医大学第一附属医院

Dates

Publication Date: 20260512
Application Date: 20260203

Claims (9)

1. A method for monitoring risk of post-operative bleeding in a hepatobiliary patient, the method comprising: S1, continuously acquiring and monitoring microcirculation blood flow, tissue bioimpedance and body surface temperature distribution of a postoperative region of a patient in a noninvasive manner, and generating real-time multi-mode data comprising capillary vessel perfusion state, peritoneal effusion risk and potential hemorrhagic thermodynamic diagram; S2, constructing an LSTM-CNN mixed model by utilizing time-frequency decomposition characteristics based on the real-time data of the capillary perfusion state and the vasomotor function in the real-time multi-mode data to obtain local tissue hypoxia index and vasomotor function abnormality probability; S3, based on the quantitative data of the abdominal cavity effusion and the blood coagulation function in the real-time multi-mode data, adopting a random forest algorithm to fuse the low-frequency impedance change rate and the albumin level, and obtaining the ascites occurrence probability and the effusion predicted value; S4, processing temperature gradient time sequence data through a three-dimensional convolutional neural network based on thermodynamic diagram data in the real-time multi-mode data to generate bleeding point positioning coordinates and thermodynamic diagram risk classification; S5, constructing a Bayes network to analyze causal relationship among various risk factors based on the local tissue hypoxia index, vasomotor function abnormality probability, ascites occurrence probability, effusion volume predicted value, bleeding point positioning coordinates and thermodynamic diagram risk classification and the basic blood coagulation function of a patient, and calculating comprehensive bleeding risk probability and bleeding area positioning based on the causal relationship; s6, dynamically adjusting a trigger threshold value of the early warning signal based on the comprehensive bleeding risk probability and the individual characteristics of the patient, and generating a grading early warning signal; And S7, recommending personalized treatment schemes or nursing suggestions based on the comprehensive bleeding risk probability, and early warning signals, wherein a personalized intervention suggestion library contains intervention measures preset for different bleeding risk grades.
2. The method for post-operative bleeding risk monitoring of a hepatobiliary patient of claim 1, wherein generating real-time multimodal data in step S1 includes: S101, acquiring capillary blood flow signals in real time through a laser Doppler effect, and extracting endothelial source and non-endothelial source blood flow frequency band characteristics through time-frequency decomposition to generate real-time data comprising capillary perfusion state and relaxation function; s102, applying multi-frequency alternating current to measure tissue impedance and phase angle, evaluating extracellular fluid accumulation through low-frequency impedance change, and generating quantitative real-time data comprising abdominal cavity fluid accumulation and blood coagulation function by reflecting intracellular fluid fluctuation through high-frequency impedance change; s103, utilizing a non-contact scanning postoperative incision area, removing dressing interference through dynamic area identification, extracting local temperature gradient and heating rate characteristics, and generating thermodynamic diagram real-time data comprising potential bleeding point positioning and bleeding types; S104, preprocessing the real-time data acquired in the steps S101 to S103 to generate real-time multi-mode data comprising capillary perfusion state, abdominal dropsy risk and potential hemorrhagic thermodynamic diagram.
3. The method for post-operative bleeding risk monitoring of a hepatobiliary patient of claim 1, wherein obtaining a local tissue hypoxia index and a probability of vasomotor dysfunction in step S2 comprises: S201, extracting features of real-time data of the capillary vessel perfusion state and the relaxation function to obtain a time-frequency decomposition feature matrix; s202, constructing the LSTM-CNN mixed model, and obtaining local tissue hypoxia indexes and vasomotor function abnormality probabilities by using the LSTM-CNN mixed model based on the time-frequency decomposition feature matrix.
4. The method for monitoring post-operative bleeding risk in a hepatobiliary patient according to claim 1, wherein obtaining the ascites occurrence probability and the fluid accumulation volume prediction value in step S3 includes: s301, carrying out feature extraction on quantitative data and albumin data of the abdominal cavity effusion and the blood coagulation function to obtain an ascites feature vector; S302, constructing a random forest algorithm model, and obtaining the ascites occurrence probability and the hydrops volume predicted value by utilizing the random forest algorithm model based on the ascites characteristic vector.
5. The method of post-operative bleeding risk monitoring for a hepatobiliary patient of claim 1, wherein generating bleeding point location coordinates and thermodynamic diagram risk classification in step S4 includes: S401, removing dressing interference through dynamic region identification based on the thermodynamic diagram data, extracting local temperature gradient and heating rate characteristics, and generating a standardized temperature gradient diagram and a heating rate time sequence; S402, carrying out space-time feature extraction on the standardized temperature gradient map and the temperature rise rate time sequence by adopting a three-dimensional convolutional neural network to obtain a high-order space-time feature vector; s403, constructing a classification and regression double-task model, and obtaining bleeding point positioning coordinates and thermodynamic diagram risk classification by utilizing the classification and regression double-task model based on the high-order space-time feature vector.
6. The method of post-operative bleeding risk monitoring for a hepatobiliary patient of claim 1, wherein calculating the integrated bleeding risk probability and bleeding area localization based on the causality in step S5 includes: S501, generating a standardized multisource data matrix based on the local tissue hypoxia index, ascites occurrence probability, bleeding point positioning coordinates and thermodynamic diagram risk classification and basic blood coagulation function of a patient through time stamp alignment, missing value interpolation and dimension normalization; S502, constructing a Bayesian network structure and learning a condition probability table, and defining causal relationship and condition dependence among risk factors; s503, based on causal relation and condition dependence among the risk factors, calculating comprehensive bleeding risk probability and bleeding area positioning through forward reasoning and backward reasoning algorithms and combining real-time monitoring data to update node probability.
7. The method of post-operative bleeding risk monitoring for a hepatobiliary patient of claim 1, wherein generating a hierarchical early warning signal in step S6 includes: s601, dynamically adjusting a grading early warning threshold value by using a reinforcement learning model and a rule engine based on the comprehensive bleeding risk probability and individual characteristics of a patient; s602, comparing the current bleeding risk probability with a dynamic threshold based on the grading early warning threshold, and generating a grading early warning signal comprising early warning level, risk value and trend change rate.
8. The method of post-operative bleeding risk monitoring for a hepatobiliary patient of claim 1, wherein recommending personalized treatment regimens or care advice in step S7 includes: S701, integrating expert experience and clinical guidelines, and constructing an intervention measure library aiming at different bleeding risk grades and individual characteristics of patients; s702, matching initial intervention measures matched with the risk level of a patient in the intervention measure library by using a rule engine and a clinical guideline based on the comprehensive bleeding risk probability; s703, optimizing the initial intervention measures based on the real-time monitoring data and the historical intervention effect data.
9. A post-operative bleeding risk monitoring system for a hepatobiliary patient, the system comprising: A processor; A memory for storing processor-executable instructions; wherein the processor is configured to implement the post-operative bleeding risk monitoring method for a hepatobiliary patient of any one of claims 1 to 8 when executing the executable instructions.

Description

Method and system for monitoring postoperative bleeding risk of hepatobiliary patient Technical Field The invention relates to the technical field of bleeding risk monitoring, in particular to a method and a system for monitoring bleeding risk of a liver and gall patient after operation. Background Post-surgical hepatobiliary hemorrhage is a clinically common and critical complication, and early warning and intervention are critical to improving patient prognosis. In the prior art, the monitoring of postoperative bleeding risk mainly depends on the experience judgment and manual intervention of medical staff, and the abdominal cavity effusion amount is usually estimated by measuring vital signs (such as blood pressure and heart rate) at regular time, observing the bleeding condition of incision dressing or combining with imaging means such as ultrasound. Although the method can provide basic risk assessment, the method has obvious limitation that continuous non-invasive acquisition cannot be realized by manual monitoring, and the judgment standard is greatly influenced by subjective experience, so that missed diagnosis or misjudgment is easily caused. For example, a recessive bleeding symptom such as abnormal microcirculation perfusion or early abdominal dropsy may delay the intervention time due to not being recognized in time. In addition, the integration capability of the prior art on multi-mode data is insufficient, and the correlation of heterogeneous data such as microcirculation blood flow, tissue biological impedance, body surface temperature and the like is difficult to be systematically analyzed. Manual judgment often evaluates a single index in isolation (e.g., only regarding blood pressure drop), and ignores causal relationships between risk factors (e.g., hypoproteinemia exacerbates abdominal dropsy, thereby inducing coagulation dysfunction). At the same time, traditional intervention advice lacks personalized adaptations, for example, for elderly patients or hypofibrinogenemic patients, unified adoption of "observe-conservation" strategies may lead to risk escalation. Disclosure of Invention The invention aims at least solving the technical problem of low accuracy of bleeding risk judgment in the prior art, and particularly creatively provides a method and a system for monitoring bleeding risk of a liver and gall patient after operation. To achieve the above object, the present invention provides a method for monitoring risk of postoperative bleeding of a hepatobiliary patient, the method comprising: S1, continuously acquiring and monitoring microcirculation blood flow, tissue bioimpedance and body surface temperature distribution of a postoperative region of a patient in a noninvasive manner, and generating real-time multi-mode data comprising capillary vessel perfusion state, peritoneal effusion risk and potential hemorrhagic thermodynamic diagram; S2, constructing an LSTM-CNN mixed model by utilizing time-frequency decomposition characteristics based on the real-time data of the capillary perfusion state and the vasomotor function in the real-time multi-mode data to obtain local tissue hypoxia index and vasomotor function abnormality probability; S3, based on the quantitative data of the abdominal cavity effusion and the blood coagulation function in the real-time multi-mode data, adopting a random forest algorithm to fuse the low-frequency impedance change rate and the albumin level, and obtaining the ascites occurrence probability and the effusion predicted value; S4, processing temperature gradient time sequence data through a three-dimensional convolutional neural network based on thermodynamic diagram data in the real-time multi-mode data to generate bleeding point positioning coordinates and thermodynamic diagram risk classification; S5, constructing a Bayes network to analyze causal relationship among various risk factors based on the local tissue hypoxia index, vasomotor function abnormality probability, ascites occurrence probability, effusion volume predicted value, bleeding point positioning coordinates and thermodynamic diagram risk classification and the basic blood coagulation function of a patient, and calculating comprehensive bleeding risk probability and bleeding area positioning based on the causal relationship; s6, dynamically adjusting a trigger threshold value of the early warning signal based on the comprehensive bleeding risk probability and the individual characteristics of the patient, and generating a grading early warning signal; And S7, recommending personalized treatment schemes or nursing suggestions based on the comprehensive bleeding risk probability, and early warning signals, wherein a personalized intervention suggestion library contains intervention measures preset for different bleeding risk grades. In another aspect, the present invention also provides a post-operative bleeding risk monitoring system for a hepatobiliary patient, the system comprising: