WO2021012203A1 - 基于围术期危险预警的多模型互补增强机器学习平台 - Google Patents
基于围术期危险预警的多模型互补增强机器学习平台 Download PDFInfo
- Publication number
- WO2021012203A1 WO2021012203A1 PCT/CN2019/097415 CN2019097415W WO2021012203A1 WO 2021012203 A1 WO2021012203 A1 WO 2021012203A1 CN 2019097415 W CN2019097415 W CN 2019097415W WO 2021012203 A1 WO2021012203 A1 WO 2021012203A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- model
- patient
- intraoperative
- platform based
- enhanced machine
- Prior art date
Links
- 230000000295 complement effect Effects 0.000 title claims abstract description 15
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 35
- 238000000034 method Methods 0.000 claims abstract description 22
- 230000003068 static effect Effects 0.000 claims abstract description 13
- 208000032984 Intraoperative Complications Diseases 0.000 claims abstract description 9
- 208000035965 Postoperative Complications Diseases 0.000 claims abstract description 9
- 229940079593 drug Drugs 0.000 claims abstract description 9
- 239000003814 drug Substances 0.000 claims abstract description 9
- 208000024891 symptom Diseases 0.000 claims abstract description 9
- 206010057765 Procedural complication Diseases 0.000 claims abstract description 8
- 238000013528 artificial neural network Methods 0.000 claims abstract description 8
- 206010062300 Procedural hypotension Diseases 0.000 claims abstract description 7
- 238000010801 machine learning Methods 0.000 claims description 14
- 230000008602 contraction Effects 0.000 claims description 10
- 238000007635 classification algorithm Methods 0.000 claims description 6
- 238000010586 diagram Methods 0.000 claims description 6
- 238000002483 medication Methods 0.000 claims description 6
- 238000012544 monitoring process Methods 0.000 claims description 6
- 238000013507 mapping Methods 0.000 claims description 5
- 230000000630 rising effect Effects 0.000 claims description 5
- 238000001356 surgical procedure Methods 0.000 claims description 5
- 238000007637 random forest analysis Methods 0.000 claims description 4
- 208000009304 Acute Kidney Injury Diseases 0.000 claims description 3
- 208000012902 Nervous system disease Diseases 0.000 claims description 3
- 208000033626 Renal failure acute Diseases 0.000 claims description 3
- 206010047249 Venous thrombosis Diseases 0.000 claims description 3
- 201000011040 acute kidney failure Diseases 0.000 claims description 3
- 238000002790 cross-validation Methods 0.000 claims description 3
- 201000010099 disease Diseases 0.000 claims description 3
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 claims description 3
- 230000002526 effect on cardiovascular system Effects 0.000 claims description 3
- 238000007477 logistic regression Methods 0.000 claims description 3
- 210000002569 neuron Anatomy 0.000 claims description 3
- 230000002159 abnormal effect Effects 0.000 claims description 2
- 230000003205 diastolic effect Effects 0.000 claims description 2
- 230000010354 integration Effects 0.000 claims 1
- 238000011156 evaluation Methods 0.000 abstract description 4
- 238000007689 inspection Methods 0.000 abstract description 3
- 239000000523 sample Substances 0.000 description 4
- 210000004556 brain Anatomy 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 230000004927 fusion Effects 0.000 description 2
- 239000013610 patient sample Substances 0.000 description 2
- 206010058046 Post procedural complication Diseases 0.000 description 1
- 206010062886 Procedural hypertension Diseases 0.000 description 1
- 230000005856 abnormality Effects 0.000 description 1
- 230000036772 blood pressure Effects 0.000 description 1
- 230000000747 cardiac effect Effects 0.000 description 1
- 238000013210 evaluation model Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/20—Ensemble learning
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/30—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for calculating health indices; for individual health risk assessment
Definitions
- the present invention relates to the field of intelligent medical treatment, in particular to a multi-model complementary enhanced machine learning platform based on perioperative risk warning.
- machine learning fields such as feature extraction and feature selection, artificial neural networks, support vector machines, random forests, and XGboost are developing well and are used in various service industries.
- the prediction of a single type of dangerous complication in the perioperative period is mainly based on the local information of the patient. The accuracy of the prediction is not high, and the previous model is not effective when performing migration. For example, a model for predicting intraoperative hypotension is It often does not have good performance when predicting intraoperative hypertension.
- the existing models are mainly divided into two categories: the first is to use the basic information of the patient before the operation, past medical history, inspection report, medication record and other data to evaluate and early warning.
- the second is to use the electrographic information of the patient's surgical center to predict dangerous symptoms such as intraoperative hypotension.
- the above methods only use the patient's one-sided data, and cannot make full use of the full amount of patient data collected during the perioperative period. Therefore, the accuracy of the prediction of intraoperative and postoperative complications is not high.
- the first method does not input information such as blood pressure and electrocardiogram in the patient's intraoperative life monitor into the model when predicting the patient's state, so it can only provide an assessment of dangerous complications based on preoperative static data.
- the present invention organically integrates the outputs of different algorithms by using an artificial neural network, and organically integrates the above-mentioned two types of methods to form a new model for predicting perioperative complications.
- this method can make full use of the patient's full data, overcome the shortcomings of the original method that can only use part of the patient data to predict a certain specific complication, and the main intraoperative and postoperative complications that may occur in the patient Carry out a static and dynamic comprehensive assessment and early warning, and it has good accuracy and stability.
- the present invention provides a multi-model complementary and enhanced machine learning platform based on perioperative risk warning to realize accurate symptom prediction and comprehensive evaluation and warning.
- the technical scheme adopted by the present invention is as follows:
- a multi-model complementary enhanced machine learning platform based on perioperative risk warning mainly includes the following steps:
- S1 Process the variables used in the model, especially the information related to the intraoperative life monitor;
- the variables in S1 include patient static data, dynamic data of the patient's intraoperative life monitor; the patient's static data includes basic patient information, patient coexisting diseases, admission day medications, type of surgery, preoperative experimental results, intraoperative medications,
- the dynamic data of the patient's intraoperative life monitor includes electrocardiogram and brain wave graph, which are specifically based on actual monitoring results.
- the static data of the patient is a continuous variable, it is directly input into the model after unitization, and if it is a discrete variable, it is converted into "0" and "1" and then coded into the model; the patient's intraoperative life monitor
- the waveform of the dynamic data is split and the five stages of contraction, diastole, contraction rising, contraction decay and total decay are obtained.
- the graphic features of each stage are extracted as the corresponding model input.
- the 9 sub-algorithms described in S2 are: Logistic Regression, XGboost, Random Forest, SVM, ANN, KNN, Naive Bayes, GBDT, LightGBM, and they are further integrated.
- the model needs to predict multiple intraoperative and postoperative complications at the same time. For a sample with a sample size of n, if a total of p complications are predicted, the output result of the k sub-algorithms is an n ⁇ k ⁇ p array. All classification algorithms use grid-search and half-fold cross validation for parameter tuning.
- An ANN model is a vector that only contains "0” and “1” and is mapped into a two-class mapping of "0" or "1".
- the number of neurons in the input layer of the neural network in the ANN algorithm is equal to the number k of the sub-algorithms we used in S2.
- the early warning model includes monitoring and early warning of intraoperative hypotension, acute kidney injury, venous thrombosis, cardiovascular complications, neurological complications, and abnormal operation center rate.
- the invention uses artificial neural networks to organically integrate the output of different algorithms.
- the first is to use the patient’s basic information before surgery, past medical history, inspection reports, medication records and other data for evaluation and early warning.
- the second is to use patient surgery.
- ECG information is used to predict dangerous symptoms such as intraoperative hypotension, and the above two categories of methods are organically integrated to form a new model for predicting perioperative complications.
- this method can make full use of the patient's full data, overcome the shortcomings of the original method that can only use part of the patient data to predict a certain specific complication, and the main intraoperative and postoperative complications that may occur in the patient Carry out a static and dynamic comprehensive assessment and early warning, and it has good accuracy and stability.
- Figure 1 is a schematic diagram of part of the interface of the present invention.
- Figure 2 is a schematic diagram of the stages of an electrocardiogram.
- Figure 3 is a schematic diagram of a neural network for n patient samples.
- Figure 4 is a schematic flow diagram of the present invention.
- Fig. 1 is a schematic diagram of part of the interface of the present invention, which realizes comprehensive evaluation and real-time control.
- a multi-model complementary enhanced machine learning platform based on perioperative risk warning mainly includes the following steps:
- S1 Process the variables used in the model, especially the information related to the intraoperative life monitor;
- the variables in S1 include patient static data, dynamic data of the patient's intraoperative life monitor; the patient's static data includes basic patient information, patient coexisting diseases, admission day medications, type of surgery, preoperative experimental results, intraoperative medications,
- the dynamic data of the patient's intraoperative life monitor includes electrocardiogram and brain wave graph, which are specifically based on actual monitoring results.
- the static data of the patient is a continuous variable, it is directly input into the model after unitization, and if it is a discrete variable, it is converted into "0" and "1" and then coded into the model; the patient's intraoperative life monitor
- the waveform of the dynamic data is split and the five stages of contraction, diastole, contraction rising, contraction decay and total decay are obtained.
- the graphic features of each stage are extracted as the corresponding model input.
- the 9 sub-algorithms described in S2 are: Logistic Regression, XGboost, Random Forest, SVM, ANN, KNN, Naive Bayes, GBDT, LightGBM, and they are further integrated. These algorithms are mature algorithms for modeling and predicting categorical variables and have been widely used in clinical classification and prediction. The specific algorithm composition can be appropriately modified and added from actual data to achieve optimal prediction performance.
- k sub-algorithms denoted as f 1 (x),...,f k (x), where x is the total amount of patient data collected.
- Each sub-algorithm implements a mapping, namely This formula, namely "0" means that the patient has no related complications, and "1" means that the patient has related complications. Therefore, for a sample with a sample size of n, the output of k sub-algorithms is:
- j represents the jth complication to be predicted. Since our model needs to predict multiple intraoperative and postoperative complications at the same time, suppose we want to predict a total of p complications, then based on all complications, the output result of k sub-algorithms is an n ⁇ k ⁇ p array, as shown in the figure below As shown, each of these layers is based on the output of a complication sub-algorithm.
- the various classification algorithms used in this step all use grid-search and half-fold cross-validation for parameter tuning.
- each layer corresponds to a kind of intraoperative or postoperative complication.
- the ANN algorithm is used to classify the signals of each layer to determine whether the patient will have the corresponding symptoms.
- an ANN model be a vector that only contains "0” and "1", for example (10 ⁇ 1) is mapped into a two-class mapping of "0" or "1", as shown below:
- g represents an ANN mapping.
- the number of neurons in the input layer of the neural network is equal to the number k of the sub-algorithms we used in the second step.
- T ij the real output value
- the actual output value is g(f ij (x))
- the parameter tuning of the model is based on the preset loss function as
- the early warning model includes monitoring and early warning of intraoperative hypotension, acute kidney injury, venous thrombosis, cardiovascular complications, neurological complications, and operative center rate abnormalities.
- this method uses the ANN algorithm to fuse other various algorithms.
- the input of the ANN model used for fusion is no longer the traditional original variables, but a series of sub-algorithms to the original
- the output value of variables after preliminary classification and prediction is to obtain a globally robust prediction and evaluation model, which makes up for the shortcomings of traditional methods.
- the fusion predictive model has more stable performance and can predict multiple intraoperative and postoperative complications at the same time.
Landscapes
- Engineering & Computer Science (AREA)
- Medical Informatics (AREA)
- Public Health (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Databases & Information Systems (AREA)
- General Health & Medical Sciences (AREA)
- Pathology (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Epidemiology (AREA)
- Physics & Mathematics (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Biomedical Technology (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
本发明是设计一种基于围术期危险预警的多模型互补增强机器学习平台,本发明通过使用人工神经网络将不同算法的输出进行有机整合,第一种是利用病人手术前的基本信息、既往病史、检验检查报告、用药记录等数据进行评估预警,第二种是利用病人术中心电图信息来预测例如术中低血压等危险症状,本发明将上述的两大类方法进行了有机的融合,形成一个全新的预测围手术期并发症的模型。通过整合不同模型的输出,本方法能充分利用患者全量数据,克服了原有方法只能利用部分患者数据做出某种特定并发症预测的不足,对患者可能出现的术中术后主要并发症进行一个静态加动态的综合评估预警,并且具有良好的准确性和稳定性。
Description
本发明涉及智能医疗领域,具体而言是一种基于围术期危险预警的多模型互补增强机器学习平台。
当前特征提取和特征选择、人工神经网络、支持向量机、随机森林、XGboost等机器学习领域发展态势良好并应用于各服务业。目前主要基于患者的局部信息对围手术期单一种类的危险并发症进行预测,预测准确度不高,且以往模型在进行迁移时效果很不好,例如,一个预测术中低血压的模型,在预测术中高血压时往往不具备良好的表现。针对围手术期的预警,现有模型主要分为两类:第一种是利用病人手术前的基本信息、既往病史、检验检查报告、用药记录等数据进行评估预警。第二种是利用病人术中心电图信息来预测例如术中低血压等危险症状。以上方法都只运用了病人的片面数据,无法充分利用围手术期所能收集到的病人全量数据,因此对于病人术中术后并发症的预测准确度不高。例如,第一种方法在预测病人状态时,没有将病人术中生命监护仪中的诸如血压、心电等信息输入模型,因而只能给出一个基于术前静态数据的危险并发症评估。
同时,因为两种方法所用的变量种类完全不同,将两种方法进行组合也是困难的。本发明通过使用人工神经网络将不同算法的输出进 行有机整合,将上述的两大类方法进行了有机的融合,形成一个全新的预测围手术期并发症的模型。通过整合不同模型的输出,本方法能充分利用患者全量数据,克服了原有方法只能利用部分患者数据做出某种特定并发症预测的不足,对患者可能出现的术中术后主要并发症进行一个静态加动态的综合评估预警,并且具有良好的准确性和稳定性。
发明内容
针对上述问题,本发明提供一种基于围术期危险预警的多模型互补增强机器学习平台,实现症状预测准确、综合评估预警。本发明采用的技术方案如下:
一种基于围术期危险预警的多模型互补增强机器学习平台,主要包括以下步骤:
S1:对模型所采用的变量,尤其是术中生命监护仪相关信息进行处理;
S2:选取9个常用的分类算法,作为我们模型的子算法,以提高整个模型的稳定性和精确度;
S3:利用S2中所得数组中的每一层作为ANN算法的输入信号,利用ANN算法对各层信号进行分类,判别病人是否会出现相应的症状,建立预警模型。
S1中所述变量包括患者静态数据、患者术中生命监护仪的动态数据;所述患者静态数据包括患者基本信息、患者并存疾病、入院日 药物、手术类型、术前实验结果、术中药物,所述患者术中生命监护仪的动态数据包括心电图、脑电波图,具体以实际监测结果为根据。
所述的患者静态数据,若为连续型变量则经单位化后直接输入模型,若为离散型变量则转化为“0”、“1”编码后输入模型;所述的患者术中生命监护仪的动态数据,其波形予以拆分并得到收缩阶段、舒张阶段、收缩上升阶段、收缩衰减阶段以及总衰减阶段五个阶段,提取各个阶段的图形学特征作为相应的模型输入。
S2中所述9个子算法分别是:逻辑回归、XGboost、随机森林、SVM、ANN、KNN、朴素贝叶斯、GBDT、LightGBM,并经过进一步的融合组成。
模型要同时预测多种术中及术后并发症,针对一个样本量为n的样本,若预测总共p种并发症,k个子算法的输出结果为一个n·k·p数组,所采用的各种分类算法均采用grid-search以及五折cross validation进行参数调优。
一个ANN模型为一个将只含有“0”和“1”的向量,映射为“0”或“1”二分类的映射。ANN算法中神经网络的输入层神经元个数等于我们在S2中所采用的子算法个数k。
所述的预警模型包括对术中低血压、急性肾损伤、静脉血栓、心血管并发症、神经系统并发症、术中心率失常进行监测预警。
该发明通过使用人工神经网络将不同算法的输出进行有机整合,第一种是利用病人手术前的基本信息、既往病史、检验检查报告、用药记录等数据进行评估预警,第二种是利用病人术中心电图信息来预 测例如术中低血压等危险症状,将上述的两大类方法进行了有机的融合,形成一个全新的预测围手术期并发症的模型。通过整合不同模型的输出,本方法能充分利用患者全量数据,克服了原有方法只能利用部分患者数据做出某种特定并发症预测的不足,对患者可能出现的术中术后主要并发症进行一个静态加动态的综合评估预警,并且具有良好的准确性和稳定性。
图1为本发明的部分界面示意图。
图2为心电图的阶段示意图。
图3为n个病人样本的神经网络示意图。
图4为本发明的流程示意图。
构建一种基于围术期危险预警的多模型互补增强机器学习平台,如图1所示为本发明的部分界面示意图,实现综合评估、实时调控。
一种基于围术期危险预警的多模型互补增强机器学习平台,主要包括以下步骤:
S1:对模型所采用的变量,尤其是术中生命监护仪相关信息进行处理;
S2:选取9个常用的分类算法,作为我们模型的子算法,以提高整个模型的稳定性和精确度;
S3:利用S2中所得数组中的每一层作为ANN算法的输入信号, 利用ANN算法对各层信号进行分类,判别病人是否会出现相应的症状,建立预警模型。
S1中所述变量包括患者静态数据、患者术中生命监护仪的动态数据;所述患者静态数据包括患者基本信息、患者并存疾病、入院日药物、手术类型、术前实验结果、术中药物,所述患者术中生命监护仪的动态数据包括心电图、脑电波图,具体以实际监测结果为根据。
所述的患者静态数据,若为连续型变量则经单位化后直接输入模型,若为离散型变量则转化为“0”、“1”编码后输入模型;所述的患者术中生命监护仪的动态数据,其波形予以拆分并得到收缩阶段、舒张阶段、收缩上升阶段、收缩衰减阶段以及总衰减阶段五个阶段,提取各个阶段的图形学特征作为相应的模型输入。
如心电图图2所示,我们基于一个完整的心动周期的波形予以拆分并得到收缩阶段、舒张阶段、收缩上升阶段、收缩衰减阶段以及总衰减阶段五个阶段。随后提取各个阶段的图形学特征作为相应的模型输入,具体流程如图2所示,图中数字1到5依次表示收缩阶段1、舒张阶段2、收缩上升阶段3、收缩衰减阶段4以及总衰减阶段5。
记病人的动态数据经过拆分以及特征提取后所得的数据为x
2。那么病人的全量数据即为x=(x
1,x
2)。
S2中所述9个子算法分别是:逻辑回归、XGboost、随机森林、SVM、ANN、KNN、朴素贝叶斯、GBDT、LightGBM,并经过进一步的融合组成。这些算法都是针对分类变量进行建模预测的成熟算法,在临床分类预测上已经被广泛采用。具体算法组成可以由实际数据进行适 当修改和添加以达到最优预测性能。
假设我们采用了k个子算法,分别记为f
1(x),···,f
k(x),其中x为收集到的患者全量数据。每个子算法实现一个映射,即
这个公式,即“0”代表患者没有发生相关并发症,“1”代表患者发生相关并发症。从而,针对一个样本量为n的样本,k个子算法的输出为:
其中j代表第j个要预测的并发症。由于我们的模型要同时预测多种术中及术后并发症,假设我们要预测总共p种并发症,那么基于所有并发症,k个子算法的输出结果为一个n·k·p数组,如下图所示,其中的每一层都是基于一个并发症的子算法输出。本步所采用的各种分类算法均采用grid-search以及五折cross validation进行参数调优。
记第二步中所得结果为f
ij(x),i=1,···,k;j=1,···,p,我们利用第二步中所得数组中的每一层作为ANN算法的输入信号,每一层都对 应一种术中或术后并发症,利用ANN算法对各层信号进行分类,即判别病人是否会出现相应的症状。令一个ANN模型为一个将只含有“0”和“1”的向量例如(10···1)映射为“0”或“1”二分类的映射,如下所示:
其中,g表示一个ANN映射。对于某一个并发症的全部n个病人样本,如图3所示,神经网络的输入层神经元个数等于我们在第二步中所采用的子算法个数k,对于上图矩阵中的每一行,假设我们已知真实输出值为T
ij,逐行输入g后,实际输出值为g(f
ij(x)),模型的参数调优基于预先设定损失函数为
|T
ij-g(f
ij(x))|,i=1,···,k;j=1,···,p。
如图4所示,所述的预警模型包括对术中低血压、急性肾损伤、静脉血栓、心血管并发症、神经系统并发症、术中心率失常进行监测预警。
针对围术期相关预警的重要性,本方法独创性的采用ANN算法来融合其他多种算法,用于融合的ANN模型的输入不再是传统的原始变量,而是经过一系列子算法对原始变量经过初步分类预测后的输出值,旨在取得一个全局稳健的预测评估模型,弥补了传统方法的不足。融合后的预测模型具有更加稳定的表现力,并且可以同时预测多种术中术后并发症。
Claims (8)
- 一种基于围术期危险预警的多模型互补增强机器学习平台,其特征在于,包括以下步骤:S1:对模型所采用的变量,尤其是术中生命监护仪相关信息进行处理;S2:选取9个常用的分类算法,作为我们模型的子算法,以提高整个模型的稳定性和精确度;S3:利用S2中所得数组中的每一层作为ANN算法的输入信号,利用ANN算法对各层信号进行分类,判别病人是否会出现相应的症状,建立预警模型。
- 根据权利要求1所述的一种基于围术期危险预警的多模型互补增强机器学习平台,其特征在于:S1中所述变量包括患者静态数据、患者术中生命监护仪的动态数据;所述患者静态数据包括患者基本信息、患者并存疾病、入院日药物、手术类型、术前实验结果、术中药物,所述患者术中生命监护仪的动态数据包括心电图、脑电波图,具体以实际监测结果为根据。
- 根据权利要求2所述的一种基于围术期危险预警的多模型互补增强机器学习平台,其特征在于:所述的患者静态数据,若为连续型变量则经单位化后直接输入模型,若为离散型变量则转化为“0”、“1”编码后输入模型;所述的患者术中生命监护仪的动态数据,其波形予以拆分并得到收缩阶段、舒张阶段、收缩上升阶段、收缩衰减阶段以及总衰减阶段五个阶段,提取各个阶段的图形学特征作为相应 的模型输入。
- 根据权利要求1所述的一种基于围术期危险预警的多模型互补增强机器学习平台,其特征在于:S2中所述9个子算法分别是:逻辑回归、XGboost、随机森林、SVM、ANN、KNN、朴素贝叶斯、GBDT、LightGBM,并经过进一步的融合组成。
- 根据权利要求4所述的一种基于围术期危险预警的多模型互补增强机器学习平台,其特征在于:模型要同时预测多种术中及术后并发症,针对一个样本量为n的样本,若预测总共p种并发症,k个子算法的输出结果为一个n·k·p数组,所采用的各种分类算法均采用grid-search以及五折cross validation进行参数调优。
- 根据权利要求1所述的一种基于围术期危险预警的多模型互补增强机器学习平台,其特征在于:一个ANN模型为一个将只含有“0”和“1”的向量,映射为“0”或“1”二分类的映射。
- 根据权利要求6所述的一种基于围术期危险预警的多模型互补增强机器学习平台,其特征在于:ANN算法中神经网络的输入层神经元个数等于我们在S2中所采用的子算法个数k。
- 根据权利要求1所述的一种基于围术期危险预警的多模型互补增强机器学习平台,其特征在于:所述的预警模型包括对术中低血压、急性肾损伤、静脉血栓、心血管并发症、神经系统并发症、术中心率失常进行监测预警。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
PCT/CN2019/097415 WO2021012203A1 (zh) | 2019-07-24 | 2019-07-24 | 基于围术期危险预警的多模型互补增强机器学习平台 |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
PCT/CN2019/097415 WO2021012203A1 (zh) | 2019-07-24 | 2019-07-24 | 基于围术期危险预警的多模型互补增强机器学习平台 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2021012203A1 true WO2021012203A1 (zh) | 2021-01-28 |
Family
ID=74192917
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/CN2019/097415 WO2021012203A1 (zh) | 2019-07-24 | 2019-07-24 | 基于围术期危险预警的多模型互补增强机器学习平台 |
Country Status (1)
Country | Link |
---|---|
WO (1) | WO2021012203A1 (zh) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2023126832A1 (en) * | 2021-12-30 | 2023-07-06 | Pfizer Inc. | Digital medicine companion for cdk inhibitor medications for cancer patients |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106295180A (zh) * | 2016-08-10 | 2017-01-04 | 中国人民解放军第三军医大学第三附属医院 | 老年骨科手术风险预测方法 |
CN107180155A (zh) * | 2017-04-17 | 2017-09-19 | 中国科学院计算技术研究所 | 一种基于异构集成模型的疾病预测方法及系统 |
CN109785973A (zh) * | 2019-01-25 | 2019-05-21 | 中国医学科学院阜外医院 | 一种手术患者在围手术期发生并发症的定量预测方法 |
CN109934415A (zh) * | 2019-03-22 | 2019-06-25 | 中国科学院重庆绿色智能技术研究院 | 一种基于跨模态深度学习的围术期危重事件预测方法 |
-
2019
- 2019-07-24 WO PCT/CN2019/097415 patent/WO2021012203A1/zh active Application Filing
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106295180A (zh) * | 2016-08-10 | 2017-01-04 | 中国人民解放军第三军医大学第三附属医院 | 老年骨科手术风险预测方法 |
CN107180155A (zh) * | 2017-04-17 | 2017-09-19 | 中国科学院计算技术研究所 | 一种基于异构集成模型的疾病预测方法及系统 |
CN109785973A (zh) * | 2019-01-25 | 2019-05-21 | 中国医学科学院阜外医院 | 一种手术患者在围手术期发生并发症的定量预测方法 |
CN109934415A (zh) * | 2019-03-22 | 2019-06-25 | 中国科学院重庆绿色智能技术研究院 | 一种基于跨模态深度学习的围术期危重事件预测方法 |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2023126832A1 (en) * | 2021-12-30 | 2023-07-06 | Pfizer Inc. | Digital medicine companion for cdk inhibitor medications for cancer patients |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN112365978B (zh) | 心动过速事件早期风险评估的模型的建立方法及其装置 | |
Shaji | Predictionand diagnosis of heart disease patients using data mining technique | |
Krishnani et al. | Prediction of coronary heart disease using supervised machine learning algorithms | |
ȚĂRANU | Data mining in healthcare: decision making and precision | |
US8554580B2 (en) | Automated management of medical data using expert knowledge and applied complexity science for risk assessment and diagnoses | |
CN110491511B (zh) | 一种基于围术期危险预警的多模型互补增强机器学习方法 | |
WO2021227511A1 (zh) | 一种基于电子病历大数据的并发症发病风险预测方法、系统、终端以及存储介质 | |
Ghosh et al. | Hypotension risk prediction via sequential contrast patterns of ICU blood pressure | |
Bhattacharya et al. | A novel classification method for predicting acute hypotensive episodes in critical care | |
CN112967803A (zh) | 基于集成模型的急诊患者早期死亡率预测方法及系统 | |
TWI796228B (zh) | 急性腎損傷預測系統及其方法 | |
KR102421172B1 (ko) | 앙상블 딥러닝과 형상 융합 기반 심장병 예측을 위한 스마트 헬스케어 모니터링 방법 및 시스템 | |
CN115083604A (zh) | 一种术后不良事件预测模型的构建方法、系统和预测装置 | |
Feng et al. | Dcmn: Double core memory network for patient outcome prediction with multimodal data | |
WO2021012203A1 (zh) | 基于围术期危险预警的多模型互补增强机器学习平台 | |
Vilas-Boas et al. | Hourly prediction of organ failure and outcome in intensive care based on data mining techniques | |
CN110895969A (zh) | 房颤预测决策树及其剪枝的方法 | |
Anooj | Implementing decision tree fuzzy rules in clinical decision support system after comparing with fuzzy based and neural network based systems | |
TWM631259U (zh) | 急性腎損傷預測系統 | |
Iscra et al. | Interpretable machine learning models to support differential diagnosis between Ischemic Heart Disease and Dilated Cardiomyopathy | |
Arora | Mining rules for head injury patients using fuzzy taxonomic structures | |
Senthilkumar et al. | Diabetes disease diagnosis using multivariate adaptive regression splines | |
Bhattacharya et al. | Performance evaluation of predictive machine learning models for diabetic disease using python | |
Ajay et al. | An Aquila-optimized SVM classifier for Diabetes prediction | |
Jelinek et al. | A survey of data mining methods for automated diagnosis of cardiac autonomic neuropathy progression |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 19938984 Country of ref document: EP Kind code of ref document: A1 |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 19938984 Country of ref document: EP Kind code of ref document: A1 |
|
32PN | Ep: public notification in the ep bulletin as address of the adressee cannot be established |
Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 13/09/2023) |