CN-115116605-B - Combined machine learning-based operation critical disease auxiliary evaluation method

CN115116605BCN 115116605 BCN115116605 BCN 115116605BCN-115116605-B

Abstract

The invention relates to an operation critical illness auxiliary evaluation method based on combined machine learning, which comprises the following steps of S1, dividing and artificially marking time sequence data of history monitoring and monitoring, S2, carrying out data complementation by utilizing random forest regression based on a decision tree, S3, carrying out correlation analysis of critical illness common indexes, S4, establishing a pre-operation auxiliary evaluation model and training by utilizing a combined machine learning method, S5, carrying out evaluation analysis on the pre-operation critical illness by an auxiliary doctor, S6, clustering the time sequence data of history monitoring and monitoring in operation according to characteristic values in operation, S7, calculating an optimal monitoring period by utilizing a loss function, S8, calculating a critical illness degree quantification value by utilizing the optimal monitoring period, S9, establishing an operation auxiliary evaluation model and training by utilizing a combined machine learning method, and S10, carrying out evaluation analysis on the critical illness in operation by the auxiliary doctor. The invention can comprehensively and accurately predict the critical period and the critical degree, and is beneficial to assisting medical staff in effective intervention.

Inventors

CHEN YUWEN
WU HAIYANG
CHEN JIAYI
ZHANG JU

Assignees

中国科学院重庆绿色智能技术研究院

Dates

Publication Date: 20260505
Application Date: 20220612

Claims (8)

1. The auxiliary evaluation method for the surgical criticality based on the combined machine learning is characterized by comprising the following steps of: S1, dividing time sequence data of history monitoring and monitoring into time sequence data of history monitoring and monitoring before operation and time sequence data of history monitoring and monitoring during operation according to preoperation stage and intraoperative stage, and carrying out artificial marking of critical symptoms and periods thereof according to doctor diagnosis conclusion; S2, filling the missing monitoring index data of the preoperative and intra-operative historical monitoring and monitoring time sequence data by utilizing a random forest regression algorithm based on a decision tree; S3, carrying out correlation analysis on the critical illness common index according to the preoperative history monitoring and the monitored time sequence data, and extracting the critical illness influence weight and the large monitored index as preoperative factors; s4, taking a preoperative factor as the input of a model, establishing a preoperative auxiliary evaluation model adapting to the unnecessary type critical diseases by using a combined machine learning method, and training the preoperative auxiliary evaluation model by using preoperative historical monitoring and monitored time sequence data; s5, inputting preoperative factors of a patient into a trained preoperative auxiliary evaluation model of the critical illness, and assisting doctors in evaluating and analyzing the preoperative critical illness; s6, carrying out intra-operation characteristic value calculation on data marked as critical illness periods in the time sequence data of the intra-operation history monitoring and monitoring, and similarly, carrying out intra-operation characteristic value calculation on patient data marked as non-critical illness periods in the time sequence data of the intra-operation history monitoring and monitoring, and then carrying out clustering according to the intra-operation characteristic values to obtain a critical illness clustering center and a non-critical illness clustering center; s7, a monitoring period is initially set, intraoperative historical monitoring and monitoring time sequence data are subjected to intraoperative characteristic value calculation according to the monitoring period, classification is carried out according to the distances between the intraoperative characteristic value and critical and non-critical clustering centers, and an optimal monitoring period is calculated by using a loss function; s8, calculating a critical degree quantized value by utilizing intraoperative historical monitoring and monitoring time sequence data in an optimal monitoring period; S9, taking a preoperative factor, an intra-operative characteristic value and a critical degree quantized value as the input of a model, establishing an intra-operative auxiliary evaluation model for adapting to the non-use type critical by using a combined machine learning method, and training the intra-operative auxiliary evaluation model by using historical monitoring and time sequence data before and during the operation; S10, inputting preoperative factors, intra-operative characteristic values and critical degree quantized values of a certain patient into a trained intra-operative auxiliary evaluation model of the critical symptoms, and assisting a doctor in evaluating and analyzing the intra-operative critical symptoms.
2. The auxiliary evaluation method for the surgical critical symptoms based on the combined machine learning according to claim 1 is characterized in that the combined machine learning method in the step S4 and the step S9 is characterized in that firstly, a regression analysis model which takes input to a critical symptom judgment result as output is established from angles of an artificial deep neural network, a XGBOOST regression method, a support vector machine and a random forest regression method respectively by adopting an integrated learning method, then index evaluation is carried out on all regression analysis models, and an optimal regression analysis model is selected to finally judge the critical symptom result.
3. The method for assisted assessment of surgical criticality based on combined machine learning according to claim 2, wherein the integrated learning method comprises Bootstrap Aggregating, boosting algorithm (Boosting), random Forest (Random Forest), and the index in the index evaluation comprises Mean Square Error (MSE), mean Absolute Error (MAE) and determination coefficient (R-square).
4. The method for assisted evaluation of surgical criticality based on combined machine learning according to claim 1, wherein the intra-operative eigenvalue in step S6 is a four-dimensional vector composed of a first moment, i.e., a mean value, a second moment, i.e., a variance, a third moment skewness, and a fourth moment kurtosis of the intra-cycle data.
5. The method for assisted assessment of surgical criticality based on combined machine learning according to claim 1, wherein the clustering method of step S6 is a clustering method using K-means method for two classifications of critical and non-critical cluster centers.
6. The method for assisted assessment of surgical criticality based on combined machine learning of claim 1, wherein the loss function of step S7 is: Wherein, the 、 For critical and non-critical cluster centers respectively, Historical monitoring and time series data for pre-and intra-operative histories of a single patient.
7. The method for assisted evaluation of surgical criticality based on combined machine learning according to claim 1, wherein said step S7 is specifically: s701, setting the sampling period of the history monitoring and time sequence data in operation as the minimum step length, and initially setting the step number of one monitoring period as 2; S702, performing intra-operation characteristic value calculation on the time sequence data of intra-operation history monitoring and monitoring according to a monitoring period, and classifying according to the distances between the time sequence data and critical and non-critical cluster centers; s703, calculating a loss function according to the classification result and the intra-operative characteristic value in the monitoring period; and S704, increasing the step number of the monitoring period by 1, and repeating the steps S702-S703 until N times, and selecting the monitoring period with the minimum loss function as the optimal monitoring period.
8. The combined machine learning based surgical criticality auxiliary assessment method according to claim 1, wherein the criticality quantitative value of step S8 The calculating method comprises that when the result of Chinese angelica is critical, , wherein, Is the enveloping radius of critical illness clustering, when the Chinese angelica is non-critical illness, 。

Description

Combined machine learning-based operation critical disease auxiliary evaluation method Technical Field The invention relates to an auxiliary evaluation method for surgical critical symptoms based on combined machine learning, belongs to data mining, and is particularly suitable for the auxiliary evaluation method for surgical critical symptoms based on combined machine learning. Background Critical symptoms refer to clinical signs of a critical condition that is severe, variable and life threatening. Vital signs of critically ill patients are unstable, the condition change is rapid, and most of the vital signs are accompanied by one or more organ dysfunction or failure, so that comprehensive judgment can be performed according to vital signs such as body temperature, pulse, respiration, blood pressure and the like. Once critical conditions occur, the condition is critical and changes rapidly, with little carelessness often resulting in irreparable consequences, thus requiring the physician to be able to make a correct judgment in a short period of time and to determine therapeutic measures. With the development of big data technology, the prediction and evaluation of critical symptoms of patients by using monitoring data are possible. However, merely monitoring a moment or a small segment of the procedure on-line to determine whether there is a risk of developing a critical condition is very onesided and inaccurate. In fact, when the doctor determines that the patient is critically ill, the doctor changes the situation to a non-critically ill situation after the doctor effectively intervenes after the situation that the suspected critically ill situation or the critically ill probability is extremely high occurs in a certain period of time in the middle of the situation, not all the time when the patient is critically ill. Currently, the existing research cannot accurately give out what time period and risk level information of critical illness through monitoring data so as to assist medical staff in performing effective intervention in advance. Disclosure of Invention In view of the above, the present invention provides and is based on a combined machine learning method for assisting in the evaluation of critical illness of surgery, which aims to predict accurate critical illness period and critical illness degree by comprehensively analyzing combined machine learning technology by adopting a method of combining preoperative and intra-operative history monitoring and monitoring time sequence data. In order to achieve the above purpose, the present invention provides the following technical solutions: the auxiliary evaluation method for the surgical criticality based on the combined machine learning is characterized by comprising the following steps in combination with fig. 1: S1, dividing time sequence data of history monitoring and monitoring into time sequence data of history monitoring and monitoring before operation and time sequence data of history monitoring and monitoring during operation according to preoperation stage and intraoperative stage, and carrying out artificial marking of critical symptoms and periods thereof according to doctor diagnosis conclusion; S2, filling the missing monitoring index data of the preoperative and intra-operative historical monitoring and monitoring time sequence data by utilizing a random forest regression algorithm based on a decision tree; s3, carrying out correlation analysis on the common indexes of the critical diseases according to the historical monitoring and the monitored time sequence data before operation, and extracting the weight of the critical diseases and the monitoring indexes with larger influence on the critical diseases as preoperation factors; s4, taking a preoperative factor as the input of a model, establishing a preoperative auxiliary evaluation model adapting to the unnecessary type critical diseases by using a combined machine learning method, and training the preoperative auxiliary evaluation model by using preoperative historical monitoring and monitored time sequence data; s5, inputting preoperative factors of a patient into a trained preoperative auxiliary evaluation model of the critical illness, and assisting doctors in evaluating and analyzing the preoperative critical illness; s6, carrying out intra-operation characteristic value calculation on data marked as critical illness periods in the time sequence data of the intra-operation history monitoring and monitoring, and similarly, carrying out intra-operation characteristic value calculation on patient data marked as non-critical illness periods in the time sequence data of the intra-operation history monitoring and monitoring, and then carrying out clustering according to the intra-operation characteristic values to obtain a critical illness clustering center and a non-critical illness clustering center; s7, a monitoring period is initially set, intraoperative historical monitorin