CN115399738A - Quick ICU false alarm identification method - Google Patents

Abstract

Translated fromChinese

本发明涉及ICU虚假警报识别的技术领域，揭露了一种快速ICU虚假警报识别方法，包括：采集病人体征数据，构成病人体征指标数据；利用生成对抗网络构建病人指标数据样本生成模型，利用病人指标数据样本生成模型生成大量病人体征指标数据；基于概率图理论构建ICU警报识别模型；利用分裂梯度算法对所构建的ICU警报识别模型进行快速优化，得到训练优化后的ICU警报识别模型；当医院系统检测到警报时，采集与警报相关病人的病人体征指标数据，将所采集到的病人体征指标数据输入到训练优化后的ICU警报识别模型中，模型输出是否为虚假警报，若该警报不是虚假警报，则继续输出警报名称以及处理优先级，医护人员根据模型输出结果进行相应的处置措施。本发明所述方法实现大量可用样本数据的生成，实现基于概率图的虚假警报识别、警报名称识别以及处理优先级识别。

The present invention relates to the technical field of ICU false alarm identification, and discloses a fast ICU false alarm identification method, comprising: collecting patient sign data to form patient sign index data; using generative confrontation network to construct patient index data sample generation model, using patient index The data sample generation model generates a large amount of patient sign data; builds an ICU alarm recognition model based on the probability graph theory; uses the split gradient algorithm to quickly optimize the constructed ICU alarm recognition model, and obtains an optimized ICU alarm recognition model; when the hospital system When an alarm is detected, collect the patient sign data of the patient related to the alarm, and input the collected patient sign data into the optimized ICU alarm recognition model after training. Whether the model output is a false alarm, if the alarm is not a false alarm , then continue to output the alarm name and processing priority, and the medical staff will take corresponding measures according to the output results of the model. The method of the invention realizes the generation of a large amount of available sample data, realizes false alarm identification, alarm name identification and processing priority identification based on the probability graph.

Description

Translated fromChinese

一种快速ICU虚假警报识别方法A Fast ICU False Alarm Identification Method

技术领域technical field

本发明涉及ICU虚假警报识别的技术领域，尤其涉及一种快速ICU虚假警报识别方法。The invention relates to the technical field of ICU false alarm identification, in particular to a fast ICU false alarm identification method.

背景技术Background technique

在医院ICU，由于医疗资源和相应的管理方式，都是专人专管，目前现有的ICU警报都是基于阈值诊断警报，容易产生大量虚假警报现象，虚假警报发生的原因多为患者不配合以及患者活动，比如体位改变，如厕，进食，换衣服、指夹脱落、贴片脱落等。随着人工智能、机器学习等技术的不断发展，越来越多人工智能技术应用于ICU虚假警报，降低医护人员工作压力，但是现有技术存在模型时间训练时间长，数据样本量稀少的问题，针对该问题，本专利提出一种快速ICU虚拟警报识别方法。In the ICU of a hospital, because medical resources and corresponding management methods are all managed by dedicated personnel, the current existing ICU alarms are based on threshold diagnostic alarms, which are prone to a large number of false alarms. Most of the false alarms are caused by patients’ lack of cooperation and Patient activities, such as changing body position, going to the toilet, eating, changing clothes, finger clips falling off, patch falling off, etc. With the continuous development of artificial intelligence, machine learning and other technologies, more and more artificial intelligence technologies are applied to ICU false alarms to reduce the work pressure of medical staff, but the existing technology has the problems of long model training time and scarce data samples. Aiming at this problem, this patent proposes a fast ICU virtual alarm identification method.

发明内容Contents of the invention

有鉴于此，本发明提供一种快速ICU虚假警报识别方法，目的在于(1)通过利用生成对抗网络构建病人指标数据样本生成模型，利用病人指标数据样本生成模型生成大量病人体征指标数据，从而构建得到ICU警报识别模型；(2)通过将所采集到的病人体征指标数据输入到训练优化后的ICU警报识别模型中，模型输出是否为虚假警报，若该警报不是虚假警报，则继续输出警报名称以及警报名称所对应的处理优先级，其中处理优先级由高到中，医护人员根据处理优先级以及模型输出的警报名称进行相应的处置措施，从而实现虚拟警报的快速识别，警报名称的判断以及所对应的处理优先级的输出，减少警报疲劳，帮助医护人员实现更有针对性的救治。In view of this, the present invention provides a fast ICU false alarm identification method, the purpose of which is (1) to construct a patient index data sample generation model by using a generative confrontation network, and use the patient index data sample generation model to generate a large number of patient sign index data, thereby constructing Obtain the ICU alarm recognition model; (2) By inputting the collected patient sign data into the optimized ICU alarm recognition model after training, whether the model output is a false alarm, if the alarm is not a false alarm, continue to output the alarm name And the processing priority corresponding to the alarm name, where the processing priority is from high to medium, and the medical staff will take corresponding measures according to the processing priority and the alarm name output by the model, so as to realize the rapid identification of virtual alarms, the judgment of alarm names and The corresponding processing priority output reduces alarm fatigue and helps medical staff achieve more targeted treatment.

实现上述目的，本发明提供的一种快速ICU虚假警报识别方法，包括以下步骤：Realize above-mentioned purpose, a kind of fast ICU false alarm identification method provided by the present invention comprises the following steps:

S1：采集病人体征数据，其中所述体征数据包括生命体征检查时序数据和入院48小时内实验室检查指标数据集合，并利用傅里叶变换方法以及小波分解方法对采集的生命体征检查时序数据进行特征提取，将提取特征与检查指标数据集合共同构成病人体征指标数据；S1: Collect patient sign data, where the sign data includes time-series data of vital sign examination and laboratory test index data set within 48 hours of admission, and use Fourier transform method and wavelet decomposition method to analyze the collected time-series data of vital sign examination Feature extraction, which combines the extracted features and the inspection index data set to form the patient's sign index data;

S2：按照步骤S1方法采集大量病人的病人体征指标数据以及对应的警报级别构成训练集A1，并利用生成对抗网络构建病人指标数据样本生成模型，利用病人指标数据样本生成模型生成大量病人体征指标数据，将所生成的数据同训练集A1构成训练集A；S2: According to the method of step S1, collect a large number of patients' patient sign index data and corresponding alarm levels to form a training set A1, and use the generative confrontation network to construct a patient index data sample generation model, and use the patient index data sample generation model to generate a large number of patient sign index data , form the training set A with the generated data and the training set A1;

S3：基于概率图理论构建ICU警报识别模型，其中所述模型的输入为病人体征指标数据，输出是否虚假警报的识别结果，以及对应的警报名称和警报级别；S3: Construct an ICU alarm recognition model based on the probability graph theory, wherein the input of the model is patient sign index data, output the identification result of false alarm, and the corresponding alarm name and alarm level;

S4：基于训练集A，利用分裂梯度算法对所构建的ICU警报识别模型进行快速优化，得到训练优化后的ICU警报识别模型；S4: Based on the training set A, use the split gradient algorithm to quickly optimize the built ICU alarm recognition model, and obtain the optimized ICU alarm recognition model after training;

S5：当医院系统检测到警报时，采集与警报相关病人的病人体征指标数据，将所采集到的病人体征指标数据输入到训练优化后的ICU警报识别模型中，模型输出是否为虚假警报，若该警报不是虚假警报，则继续输出警报名称以及处理优先级，医护人员根据模型输出结果进行相应的处置措施。S5: When the hospital system detects an alarm, collect the patient sign data of the patient related to the alarm, and input the collected patient sign data into the optimized ICU alarm recognition model after training. Whether the model output is a false alarm, if If the alarm is not a false alarm, continue to output the alarm name and processing priority, and the medical staff will take corresponding disposal measures according to the output results of the model.

作为本发明的进一步改进方法：As a further improvement method of the present invention:

可选地，所述S1步骤中采集病人体征数据，包括：Optionally, collecting patient sign data in the S1 step includes:

在病人进入ICU后，采集病人体征数据，其中所述体征数据包括生命体征检查时序数据和病人入院48h内的实验室检查指标数据集合，所述生命体征检查时序数据为病人的心率以及呼吸频率的时序数据，所述入院48小时内的实验室检查指标包括病人尿液的酸碱度k₁、pH值k₂、尿比重k₃、尿蛋白k₄、尿糖k₅以及管型k₆，病人血液的白细胞数量k₇、白细胞中五类细胞的数量及比例k₈、红细胞数量k₉、红细胞压积k₁₀、血红蛋白浓度k₁₁、平均红细胞体积k₁₂、血小板的数量k₁₃以及血小板压积k₁₄，所采集的病人体征数据为：After the patient enters the ICU, the patient's sign data is collected, wherein the sign data includes the time series data of the vital sign examination and the laboratory test index data set within 48 hours of the patient's admission, and the time series data of the vital sign examination is the patient's heart rate and respiratory rate. Time-series data, the laboratory test indicators within 48 hours of admission include the patient’s urine pH k₁ , pH value k₂ , urine specific gravity k₃ , urine protein k₄ , urine sugar k₅ and cast k₆ , the patient’s blood The number of white blood cells k₇ , the number and proportion of the five types of cells in white blood cells k₈ , the number of red blood cells k₉ , the hematocrit k₁₀ , the concentration of hemoglobin k₁₁ , the mean red blood cell volume k₁₂ , the number of platelets k₁₃ and the hematocrit k_14. The collected patient sign data are:

{x₁(n₁),x₂(n₂),K}{x₁ (n₁ ),x₂ (n₂ ),K}

K＝{k_num|num∈[1,14]}K＝{k_num |num∈[1,14]}

其中：in:

x₁(n₁)为病人心率的时序数据，n₁＝0,1,…,N₁-1，N₁为时序数据长度；x₁ (n₁ ) is the time-series data of the patient's heart rate, n₁ =0,1,...,N₁ -1, N₁ is the length of the time-series data;

x₂(n₂)为病人呼吸频率的时序数据，n₂＝0,1,…,N₂-1，N₂为时序数据长度；x₂ (n₂ ) is the time-series data of the patient's respiratory rate, n₂ =0,1,...,N₂ -1, N₂ is the length of the time-series data;

K为病人的实验室检查指标数据集合，包括14种的实验室检查指标。K is the patient's laboratory test index data set, including 14 kinds of laboratory test indicators.

可选地，所述S1步骤中利用傅里叶变换方法以及小波分解方法对采集的生命体征检查时序数据进行特征提取，将提取特征与检查指标数据集合共同构成病人体征指标数据，包括：Optionally, in the S1 step, the Fourier transform method and the wavelet decomposition method are used to perform feature extraction on the collected vital sign examination time-series data, and the extracted features and the examination index data set together form the patient sign index data, including:

对所采集的生命体征检查时序数据进行特征提取，所述特征提取流程为：Feature extraction is performed on the collected time-series data of vital signs examination, and the process of feature extraction is as follows:

S11：利用傅里叶变换方法提取病人呼吸频率的特征f₂(p)，其中p为傅里叶变换处理的采样点数，将p设置为64，所述傅里叶变换方法的公式为：S11: Using the Fourier transform method to extract the feature f₂ (p) of the patient's respiratory frequency, where p is the number of sampling points processed by the Fourier transform, and p is set to 64. The formula of the Fourier transform method is:

其中：in:

j为虚数单位，j²＝-1；j is the imaginary number unit, j² =-1;

e为自然常数；e is a natural constant;

S12：利用小波分解方法对病人心率的特征进行提取，所述小波分解结果为：S12: Using the wavelet decomposition method to extract the characteristics of the heart rate of the patient, the result of the wavelet decomposition is:

其中：in:

q(a,x₁(n₁))为x₁(n₁)在尺度a上的小波系数，a表示小波分解过程中的最大尺度；q(a,x₁ (n₁ )) is the wavelet coefficient of x₁ (n₁ ) on scale a, and a represents the largest scale in the process of wavelet decomposition;

将小波系数作为病人心率的特征f₁(a)，若q(a，x₁(n_R))＞0.4×max{q(a，x₁(n₁))}，n₁＝0,1,…,N₁-1，则将时序位置n_R在特征f₁(a)中标记为R峰。Take the wavelet coefficient as the characteristic f₁ (a) of the patient's heart rate, if q(a, x₁ (n_R ))>0.4×max{q(a, x₁ (n₁ ))}, n₁ =0,1 ,...,N₁ -1, then the timing position n_R is marked as an R peak in the feature f₁ (a).

可选地，所述S2步骤采集大量病人的病人体征指标数据以及对应的警报级别构成训练集A1，包括：Optionally, the S2 step collects a large number of patients' patient sign data and corresponding alarm levels to form a training set A1, including:

按照步骤S1方法采集大量病人的病人体征指标数据以及对应的警报级别构成训练集A1，所述训练集A1的表示形式为：According to the method of step S1, the patient sign index data of a large number of patients and the corresponding alarm levels are collected to form a training set A1, and the expression form of the training set A1 is:

A1＝{data_u＝(f_1,u(a),f_2,u(p),judge_u,name_u,level_u)|u∈[1,U]}A1＝{data_u ＝(f_1,u (a),f_2,u (p),judge_u ,name_u ,level_u )|u∈[1,U]}

其中：in:

data_u为所采集的第u名病人的病人体征指标数据以及对应的警报级别，U表示训练集A1中病人的总数；data_u is the collected patient sign index data of the uth patient and the corresponding alarm level, and U represents the total number of patients in the training set A1;

f_1,u(a)表示所采集的第u名病人的心率特征，f_2,u(p)为所采集的第u名病人的呼吸频率特征；f_1,u (a) represents the collected heart rate feature of the uth patient, and f_2,u (p) is the collected respiratory rate feature of the uth patient;

judge_u＝{0,1}，judge_u＝0表示该病人所对应的警报为虚假警报，并将name_u以及level_u均设置为null；judge_u＝1表示该病人所对应的警报不为虚假警报；judge_u = {0,1}, judge_u = 0 indicates that the alarm corresponding to the patient is a false alarm, and set name_u and level_u to null; judge_u = 1 indicates that the alarm corresponding to the patient is not false alarm;

name_u表示该病人所对应警报的名称，name_u∈{1,2,3,…,15}，所述警报名称包括SPO2、导联脱落、血压、HR(心率)、室早、心脏停搏、PR过高、Resp窒息、室颤、RONT、室早二联律、ST(ST段)、室早三联律、房早以及室速，依次对应name_u＝1至name_u＝15；name_u indicates the name of the alarm corresponding to the patient, name_u ∈ {1,2,3,...,15}, the alarm name includes SPO2, lead off, blood pressure, HR (heart rate), premature ventricular, cardiac arrest , PR too high, Resp asphyxia, ventricular fibrillation, RONT, premature ventricular bigeminy, ST (ST segment), ventricular premature trigeminy, atrial premature and ventricular tachycardia, corresponding to name_u = 1 to name_u = 15 in turn;

level_u表示该病人所对应的警报级别，level_u＝{1,2}，level_u＝1表示高优先级的警报，level_u＝2表示中优先级的警报；level_u indicates the alarm level corresponding to the patient, level_u = {1,2}, level_u = 1 indicates a high priority alarm, level_u = 2 indicates a medium priority alarm;

在本发明一个具体实施例中，中优先级的警报警报有SPO2、房早、室早、心率、PR过高、RONT、血压、导联脱落、ST段、室早二联律、室早三联律以及Rsep窒息，高优先级的警报警报有室速、室颤以及心脏停搏。In a specific embodiment of the present invention, the alarms with medium priority include SPO2, premature ventricular, premature ventricular, heart rate, high PR, RONT, blood pressure, lead off, ST segment, premature ventricular bigeminy, premature ventricular triplet Rhythm and Rsep apnea, high priority alarms are VT, VF, and cardiac arrest.

可选地，所述S2步骤中利用生成对抗网络构建病人指标数据样本生成模型，利用病人指标数据样本生成模型生成大量病人体征指标数据，将所生成的数据同训练集A1构成训练集A，包括：Optionally, in the step S2, a generative confrontation network is used to construct a patient index data sample generation model, a large amount of patient sign index data is generated by using the patient index data sample generation model, and the generated data and the training set A1 are used to form a training set A, including :

利用生成对抗网络构建病人指标数据样本生成模型，其中所述病人指标数据样本生成模型包括编码器G_e(·)和解码器G_d(·)，所述编码器的输入值为训练集A1中的病人体征指标数据，通过利用编码器G_e(·)对输入的病人体征指标数据进行编码，利用解码器G_d(·)对编码结果进行解码，得到重构后的病人体征指标数据，将解码结果作为模型的样本生成结果；Utilize the generative confrontation network to construct the patient index data sample generation model, wherein the patient index data sample generation model includes an encoder G_e (·) and a decoder G_d (·), and the input value of the encoder is in the training set A1 The patient’s sign index data, by using the encoder G_e (·) to encode the input patient’s sign index data, and using the decoder G_d (·) to decode the encoded result, the reconstructed patient sign index data is obtained, and the The decoding result is used as the sample generation result of the model;

计算训练集A1中judge＝0的病人所占的比例，若该比例小于0.6，则选取judge＝0的病人所对应的病人体征指标数据作为病人指标数据样本生成模型中编码器的输入，编码器对输入结果进行编码后，利用解码器对编码结果进行解码，得到重构后的病人体征指标数据，将重构得到的病人体征指标数据的judge参数设置为0，所述样本生成后judge＝0的病人所占的比例达到0.6，得到训练集A2；Calculate the proportion of patients with judge=0 in the training set A1, if the proportion is less than 0.6, select the patient sign index data corresponding to the patient with judge=0 as the input of the encoder in the patient index data sample generation model, the encoder After encoding the input result, use the decoder to decode the encoded result to obtain the reconstructed patient sign index data, set the judge parameter of the reconstructed patient sign index data to 0, and judge=0 after the sample is generated The proportion of patients in the group reaches 0.6, and the training set A2 is obtained;

计算训练集A2中不同警报名称的病人数，选取病人数最少的5个警报名称，将所选取的警报名称下的病人体征指标数据以及对应的警报级别分别构成五个数据集A2_min1,A2_min2，A2_min3，A2_min4以及A2_min5，其中每个数据集中病人体征指标数据所对应的警报名称相同，选取任意数据集的所有病人体征指标数据作为病人指标数据样本生成模型中编码器的输入，编码器对输入结果进行编码后，利用解码器对编码结果进行解码，得到重构后的病人体征指标数据，向该重构后的病人体征指标数据添加该数据集所对应的警报名称，以及警报名称所对应的警报级别，并将judge参数设置为1，得到生成样本，所述生成样本的生成数量为(num_A2_max-num_A2_minj′)×β，其中num_A2_max为训练集A2中出现频率最高的警报名称的病人人数，j′∈{1,2,3,4,5}，num_A2_minj′为数据集A2_minj′内的病人人数，β为样本生成参数，将其设置为0.3，将所生成的数据同训练集A2构成训练集A；Calculate the number of patients with different alarm names in the training set A2, select the five alarm names with the least number of patients, and form five data sets A2_min1 and A2_min2 respectively with the patient sign index data under the selected alarm names and the corresponding alarm levels , A2_min3 , A2_min4 and A2_min5 , the alarm names corresponding to the patient sign data in each data set are the same, select all the patient sign data in any data set as the input of the encoder in the patient index data sample generation model, encode After the input result is encoded by the decoder, the encoded result is decoded by the decoder to obtain the reconstructed patient sign data, and the alarm name corresponding to the data set and the alarm name are added to the reconstructed patient sign data The corresponding alarm level, and set the judge parameter to 1 to obtain the generated samples. The generated number of the generated samples is (num_A2_max -num_A2_minj′ )×β, where num_A2_max is the alarm with the highest frequency in the training set A2 The number of patients with the name, j′∈{1,2,3,4,5}, num_A2_minj′ is the number of patients in the data set A2_minj′ , β is the sample generation parameter, set it to 0.3, and the generated The data and the training set A2 form the training set A;

构建病人指标数据样本生成模型中编码器G_e(·)和解码器G_d(·)的训练目标函数：Construct the training objective function of encoder G_e ( ) and decoder G_d ( ) in the patient indicator data sample generation model:

其中：in:

f为病人指标数据样本生成模型中编码器G_e(·)的输入值，包括病人的心率特征以及呼吸频率特征，p(f)为f的分布；f is the input value of encoder G_e (·) in the patient index data sample generation model, including the patient's heart rate characteristics and respiratory frequency characteristics, p(f) is the distribution of f;

E(·)表示期望值计算；E( ) means expected value calculation;

为编码器G_e(·)的参数，

为解码器G_d(·)的参数；

is the parameter of the encoder G_e (·),

is the parameter of decoder G_d (·);

||·||₂为L2范数；||·||₂ is the L2 norm;

将多组病人的心率特征以及呼吸频率特征输入到训练目标函数中，使得训练目标函数达到最小的模型参数即为训练优化得到的模型参数，利用训练优化后的病人指标数据样本生成模型进行数据生成。Input the heart rate characteristics and respiratory frequency characteristics of multiple groups of patients into the training objective function, so that the model parameters that minimize the training objective function are the model parameters obtained by training optimization, and use the optimized patient index data sample generation model for data generation .

可选地，所述S3步骤中基于概率图理论构建ICU警报识别模型，包括：Optionally, in the S3 step, the ICU alarm recognition model is constructed based on probability graph theory, including:

基于概率图理论构建ICU警报识别模型，其中所述模型的输入为病人体征指标数据，输出是否虚假警报的识别结果，若识别为虚假警报则输出警报名称以及警报级别为null，否则输出对应的警报名称和警报级别；Construct an ICU alarm recognition model based on the probability graph theory, wherein the input of the model is the patient's sign index data, and output the identification result of whether it is a false alarm, if it is identified as a false alarm, output the alarm name and the alarm level is null, otherwise output the corresponding alarm name and alert level;

所述ICU警报识别模型为图结构G＝(E,V)，其中E表示节点集合，节点包括警报名称节点name、judge＝0的节点、judge＝1的节点、多种不同的病人心率特征节点以及呼吸频率特征节点，V表示边集合，所述边为有向边，有向边f_q→name_b表示在病人特征为f_q的情况下，发生警报name_b的概率P(name_b|f_q)；The ICU alarm recognition model is a graph structure G=(E, V), wherein E represents a node set, and the nodes include the alarm name node name, the node of judge=0, the node of judge=1, and a variety of different patient heart rate characteristic nodes And the respiratory frequency feature node,_V represents the edge set, the edge is a directed edge, and the directed edge f_q →name_b represents the probability P(name_b| f_q );

所述ICU警报识别模型的输入为病人体征指标数据[f_1,*(a),f_2,*(p)]，利用余弦相似度算法分别计算f_1,*(a)以及f_2,*(p)与所构建概率图中病人心率特征节点以及呼吸频率特征节点的相似度，将相似度最高的病人心率特征节点以及呼吸频率特征节点[f_1,sim(a),f_2,sim(p)]作为[f_1,*(a),f_2,*(p)]在概率图中的表示结果，分别计算该病人体征指标数据为虚假警报的概率以及非虚假警报的概率：The input of the ICU alarm recognition model is patient sign index data [f_1,* (a), f_2,* (p)], using the cosine similarity algorithm to calculate f_1,* (a) and f_2,* (p) The similarity with the patient's heart rate feature node and respiratory rate feature node in the constructed probability map, the patient's heart rate feature node and respiratory rate feature node with the highest similarity [f_1,sim (a),f_2,sim ( p)] as the expression result of [f_1,* (a),f_2,* (p)] in the probability map, respectively calculate the probability of the patient's sign index data being a false alarm and the probability of a non-false alarm:

P(judge＝0|f_1,sim(a),f_2,sim(p))＝w₁P(judge＝0|f_1,sim(a))w₂P(judge＝0|f_2,sim(p))P(judge=0|f_1,sim (a),f_2,sim (p))=w₁ P(judge=0|f_1,sim (a))w₂ P(judge=0|f_{2, sim} (p))

P(judge＝1|f_1,sim(a),f_2,sim(p))＝w₁P(judge＝1|f_1,sim(a))w₂P(judge＝1|f_2,sim(p))P(judge=1|f_1,sim (a),f_2,sim (p))=w₁ P(judge=1|f_1,sim (a))w₂ P(judge=1|f_{2, sim} (p))

其中：in:

w₁为心率特征在概率图的权重参数，w₂为呼吸频率特征在概率图中的权重参数；w₁ is the weight parameter of the heart rate feature in the probability map, and w₂ is the weight parameter of the respiratory frequency feature in the probability map;

P(judge＝0|f_1,sim(a),f_2,sim(p))为该病人体征指标数据为虚假警报的概率；P(judge＝0|f_{1, sim} (a), f_{2, sim} (p)) is the probability that the patient's sign index data is a false alarm;

P(judge＝1|f_1,sim(a),f_2,sim(p))为该病人体征指标数据为非虚假警报的概率；P(judge=1|f_{1, sim} (a), f_{2, sim} (p)) is the probability that the patient's sign index data is not a false alarm;

若P(judge＝0|f_1,sim(a),f_2,sim(p))≥P(judge＝1|f_1,sim(a),f_2,sim(p))则说明该病人所对应的警报为虚假警报，则直接输出警报名称以及警报级别为null；If P(judge=0|f_{1, sim} (a), f_{2, sim} (p))≥P(judge=1|f_{1, sim} (a), f_{2, sim} (p)), then the patient If the corresponding alarm is a false alarm, then directly output the alarm name and alarm level as null;

否则计算该病人体征指标数据会导致不同警报发生的概率：Otherwise, calculating the patient's sign data will result in the probability of different alarm occurrences:

P(name＝m|f_1,sim(a),f_2,sim(p))＝P(name＝m|f_1,sim(a))P(name＝m|f_2,sim(p))P(name=m|f_1,sim (a),f_2,sim (p))=P(name=m|f_1,sim (a))P(name=m|f_2,sim (p) )

其中：in:

m∈{1,2,3,…,15}，对应不同的警报名称；m∈{1,2,3,…,15}, corresponding to different alarm names;

选取发生概率最大的警报名称以及警报对应的警报级别作为模型的输出。Select the alarm name with the highest probability of occurrence and the alarm level corresponding to the alarm as the output of the model.

可选地，所述S4步骤中利用分裂梯度算法对所构建的ICU警报识别模型进行快速优化，包括：Optionally, in the S4 step, the split gradient algorithm is used to quickly optimize the built ICU alarm recognition model, including:

利用训练集A中的样本数据对所构建的ICU警报识别模型进行快速优化，能够快速得到可用的模型参数，所述ICU警报识别模型的优化流程为：Using the sample data in the training set A to quickly optimize the built ICU alarm recognition model, the available model parameters can be obtained quickly, and the optimization process of the ICU alarm recognition model is:

S41：利用K-means算法将训练集A所有病人体征指标数据进行聚类处理，将每类的聚类中心作为概率图中病人心率特征节点以及呼吸频率特征节点，病人心率特征节点以及呼吸频率特征节点与警报名称节点之间的条件概率，即为聚类中心邻近病人体征指标数据与警报名称节点条件概率的累乘，并忽略条件概率为0的点；S41: Use the K-means algorithm to cluster all the patient sign data in the training set A, and use the cluster center of each class as the patient's heart rate feature node and respiratory rate feature node, patient heart rate feature node and respiratory rate feature in the probability map The conditional probability between the node and the alarm name node is the cumulative multiplication of the patient’s sign index data adjacent to the cluster center and the conditional probability of the alarm name node, and the point with a conditional probability of 0 is ignored;

S42：构建ICU警报识别模型训练的目标函数：S42: Construct the objective function of ICU alarm recognition model training:

其中：in:

f_z为训练集A中第z个病人的病人体征指标数据，Z为训练集A中病人的总数；f_z is the patient sign index data of the zth patient in the training set A, and Z is the total number of patients in the training set A;

w₁,w₂为待训练优化的权重参数，令w＝[w₁,w₂]；w₁ , w₂ are weight parameters to be optimized for training, let w=[w₁ ,w₂ ];

S43：设置当前分裂梯度算法迭代次数为n，分裂梯度算法参数ε>0，σ∈(0,1)，设置

S43: Set the number of iterations of the current split gradient algorithm to n, the split gradient algorithm parameters ε>0, σ∈(0,1), set

S44：若

终止算法迭代流程，输出当前的权重参数wⁿ，其中wⁿ表示第n次迭代时的权重参数，将训练得到的权重参数作为ICU警报识别模型的权重参数，得到训练优化后的ICU警报识别模型；否则转向下一步；S44: if

Terminate the algorithm iteration process, output the current weight parameter wⁿ , where wⁿ represents the weight parameter at the nth iteration, and use the weight parameter obtained from training as the weight parameter of the ICU alarm recognition model to obtain the optimized ICU alarm recognition model after training ; Otherwise go to the next step;

S45：若

则令d_n＝g_n，否则令d_n＝-g_n；S45: if

Then let d_n =g_n , otherwise let d_n =-g_n ;

S46：计算第n+1次迭代的权重参数：S46: Calculate the weight parameter of the n+1th iteration:

wⁿ⁺¹＝wⁿ+λ_nd_nwⁿ⁺¹ ＝wⁿ +λ_n d_n

令n＝n+1，返回步骤S44。If n=n+1, return to step S44.

可选地，所述S5步骤中当医院系统检测到警报时，采集与警报相关病人的病人体征指标数据，将所采集到的病人体征指标数据输入到训练优化后的ICU警报识别模型中，模型输出是否为虚假警报，若该警报不是虚假警报，则继续输出警报名称以及处理优先级，包括：Optionally, in the S5 step, when the hospital system detects an alarm, collect the patient sign index data of the patient related to the alarm, and input the collected patient sign index data into the optimized ICU alarm recognition model after training, the model Whether the output is a false alarm, if the alarm is not a false alarm, continue to output the alarm name and processing priority, including:

当医院系统检测到警报时，按照步骤S1方法采集与警报相关病人的病人体征指标数据，将所采集到的病人体征指标数据输入到训练优化后的ICU警报识别模型中，模型输出是否为虚假警报，若该警报不是虚假警报，则继续输出警报名称以及警报名称所对应的处理优先级，其中处理优先级由高到中，医护人员根据处理优先级以及模型输出的警报名称进行相应的处置措施，减少警报疲劳。When the hospital system detects an alarm, collect the patient sign data of the patient related to the alarm according to the method of step S1, and input the collected patient sign data into the optimized ICU alarm recognition model after training, whether the model output is a false alarm , if the alarm is not a false alarm, then continue to output the alarm name and the processing priority corresponding to the alarm name, where the processing priority is from high to medium, and the medical staff will take corresponding measures according to the processing priority and the alarm name output by the model, Reduced alert fatigue.

为了解决上述问题，本发明还提供一种快速ICU虚假警报识别装置，其特征在于，所述装置包括：In order to solve the above problems, the present invention also provides a fast ICU false alarm identification device, characterized in that the device includes:

特征提取模块，用于采集病人体征数据，利用傅里叶变换方法以及小波分解方法对采集的生命体征检查时序数据进行特征提取，将提取特征与检查指标数据集合共同构成病人体征指标数据；The feature extraction module is used to collect the patient's vital signs data, and uses the Fourier transform method and the wavelet decomposition method to perform feature extraction on the collected vital sign examination time-series data, and the extracted features and the examination index data set together form the patient's sign index data;

训练集获取装置，用于采集大量病人的病人体征指标数据以及对应的警报级别构成训练集A1，并利用生成对抗网络构建病人指标数据样本生成模型，利用病人指标数据样本生成模型生成大量病人体征指标数据，将所生成的数据同训练集A1构成训练集A；The training set acquisition device is used to collect patient sign index data of a large number of patients and corresponding alarm levels to form a training set A1, and use the generative confrontation network to construct a patient index data sample generation model, and use the patient index data sample generation model to generate a large number of patient sign indexes Data, the generated data and the training set A1 form the training set A;

ICU警报识别装置，用于基于概率图理论构建ICU警报识别模型，基于训练集A，利用分裂梯度算法对所构建的ICU警报识别模型进行快速优化，将所采集到的病人体征指标数据输入到训练优化后的ICU警报识别模型中，模型输出是否为虚假警报，若该警报不是虚假警报，则继续输出警报名称以及处理优先级。The ICU alarm recognition device is used to build an ICU alarm recognition model based on the probability graph theory. Based on the training set A, the split gradient algorithm is used to quickly optimize the built ICU alarm recognition model, and the collected patient sign data is input into the training In the optimized ICU alarm recognition model, whether the model output is a false alarm, if the alarm is not a false alarm, continue to output the alarm name and processing priority.

为了解决上述问题，本发明还提供一种电子设备，所述电子设备包括：In order to solve the above problems, the present invention also provides an electronic device, which includes:

存储器，存储至少一个指令；及a memory storing at least one instruction; and

处理器，执行所述存储器中存储的指令以实现上述所述的快速ICU虚假警报识别方法。A processor, executing the instructions stored in the memory to implement the above-mentioned rapid ICU false alarm identification method.

为了解决上述问题，本发明还提供一种计算机可读存储介质，所述计算机可读存储介质中存储有至少一个指令，所述至少一个指令被电子设备中的处理器执行以实现上述所述的快速ICU虚假警报识别方法。In order to solve the above problems, the present invention also provides a computer-readable storage medium, at least one instruction is stored in the computer-readable storage medium, and the at least one instruction is executed by a processor in the electronic device to realize the above-mentioned Rapid ICU false alarm identification method.

相对于现有技术，本发明提出一种快速ICU虚假警报识别方法，该技术具有以下优势：Compared with prior art, the present invention proposes a kind of fast ICU false alarm identification method, and this technology has following advantage:

首先，本方案提出一种数据样本生成方法，由于现有数据样本大多为虚假警报样本，较难获取带有警报名称的数据样本，因此本方案利用生成对抗网络构建病人指标数据样本生成模型，从而生成得到大量可用的数据样本用于构建ICU警报识别模型，其中所述病人指标数据样本生成模型包括编码器G_e(·)和解码器G_d(·)，所述编码器的输入值为训练集A1中的病人体征指标数据，通过利用编码器G_e(·)对输入的病人体征指标数据进行编码，利用解码器G_d(·)对编码结果进行解码，得到重构后的病人体征指标数据，将解码结果作为模型的样本生成结果；计算训练集A1中judge＝0的病人所占的比例，若该比例小于0.6，则选取judge＝0的病人所对应的病人体征指标数据作为病人指标数据样本生成模型中编码器的输入，编码器对输入结果进行编码后，利用解码器对编码结果进行解码，得到重构后的病人体征指标数据，将重构得到的病人体征指标数据的judge参数设置为0，所述样本生成后judge＝0的病人所占的比例达到0.6，得到训练集A2；计算训练集A2中不同警报名称的病人数，选取病人数最少的5个警报名称，将所选取的警报名称下的病人体征指标数据以及对应的警报级别分别构成五个数据集A2_min1,A2_min2，A2_min3，A2_min4以及A2_min5，其中每个数据集中病人体征指标数据所对应的警报名称相同，选取任意数据集的所有病人体征指标数据作为病人指标数据样本生成模型中编码器的输入，编码器对输入结果进行编码后，利用解码器对编码结果进行解码，得到重构后的病人体征指标数据，向该重构后的病人体征指标数据添加该数据集所对应的警报名称，以及警报名称所对应的警报级别，并将judge参数设置为1，得到生成样本，所述生成样本的生成数量为(num_A2_max-num_A2_minj′)×β，其中num_A2_max为训练集A2中出现频率最高的警报名称的病人人数，j′∈{1,2,3,4,5}，num_A2_minj′为数据集A2_minj′内的病人人数，β为样本生成参数，将其设置为0.3，将所生成的数据同训练集A2构成训练集A；构建病人指标数据样本生成模型中编码器G_e(·)和解码器G_d(·)的训练目标函数：First of all, this scheme proposes a data sample generation method. Since most of the existing data samples are false alarm samples, it is difficult to obtain data samples with alarm names. Therefore, this scheme uses generative adversarial networks to build a patient index data sample generation model, so that A large number of available data samples are generated to be used to build an ICU alarm recognition model, wherein the patient indicator data sample generation model includes an encoder G_e (·) and a decoder G_d (·), and the input value of the encoder is training For the patient’s sign data in set A1, use the encoder G_e (·) to encode the input patient’s sign data, and use the decoder G_d (·) to decode the encoding result to obtain the reconstructed patient’s sign index Data, use the decoding result as the sample generation result of the model; calculate the proportion of patients with judge=0 in the training set A1, if the proportion is less than 0.6, select the corresponding patient sign index data of the patient with judge=0 as the patient index The input of the encoder in the data sample generation model. After the encoder encodes the input result, the decoder decodes the encoded result to obtain the reconstructed patient sign index data. The judge parameter of the reconstructed patient sign index data is Set to 0, the proportion of patients with judge=0 reaches 0.6 after the sample is generated, and the training set A2 is obtained; the number of patients with different alarm names in the training set A2 is calculated, and the 5 alarm names with the least number of patients are selected, and all The patient sign index data under the selected alarm name and the corresponding alarm level constitute five data sets A2_min1 , A2_min2 , A2_min3 , A2_min4 and A2_min5 , and the alarm name corresponding to the patient sign index data in each data set Similarly, select all patient sign data in any data set as the input of the encoder in the patient index data sample generation model. After the encoder encodes the input result, use the decoder to decode the encoded result to obtain the reconstructed patient sign Index data, add the alarm name corresponding to the data set to the reconstructed patient sign index data, and the alarm level corresponding to the alarm name, and set the judge parameter to 1 to obtain a generated sample, the generation of the generated sample The number is (num_A2_max -num_A2_minj′ )×β, where num_A2_max is the number of patients with the most frequent alarm name in the training set A2, j′∈{1,2,3,4,5}, num_A2_minj′ is The number of patients in the data set A2_minj′ , β is the sample generation parameter, which is set to 0.3, and the generated data is combined with the training set A2 to form the training set A; the encoder G_e (· ) and the training objective function of the decoder G_d (·):

其中：f为病人指标数据样本生成模型中编码器G_e(·)的输入值，包括病人的心率特征以及呼吸频率特征，p(f)为f的分布；E(·)表示期望值计算；

为编码器G_e(·)的参数，

为解码器G_d(·)的参数；||·||₂为L2范数；将多组病人的心率特征以及呼吸频率特征输入到训练目标函数中，使得训练目标函数达到最小的模型参数即为训练优化得到的模型参数，利用训练优化后的病人指标数据样本生成模型进行数据生成。Among them: f is the input value of the encoder_Ge (·) in the patient index data sample generation model, including the patient's heart rate characteristics and respiratory frequency characteristics, p(f) is the distribution of f; E(·) represents the calculation of expected value;

is the parameter of the encoder G_e (·),

is the parameter of the decoder G_d ( ); ||·||₂ is the L2 norm; input the heart rate characteristics and respiratory frequency characteristics of multiple groups of patients into the training objective function, so that the training objective function can achieve the minimum model parameters, namely The optimized model parameters are obtained for training, and the optimized patient indicator data sample generation model is used for data generation.

同时，本方案基于概率图理论构建ICU警报识别模型，其中所述模型的输入为病人体征指标数据，输出是否虚假警报的识别结果，若识别为虚假警报则输出警报名称以及警报级别为null，否则输出对应的警报名称和警报级别；所述ICU警报识别模型为图结构G＝(E,V)，其中E表示节点集合，节点包括警报名称节点name、judge＝0的节点、judge＝1的节点、多种不同的病人心率特征节点以及呼吸频率特征节点，V表示边集合，所述边为有向边，有向边f_q→name_b表示在病人特征为f_q的情况下，发生警报name_b的概率P(name_b|f_q)；所述ICU警报识别模型的输入为病人体征指标数据[f_1,*(a),f_2,*(p)]，利用余弦相似度算法分别计算f_1,*(a)以及f_2,*(p)与所构建概率图中病人心率特征节点以及呼吸频率特征节点的相似度，将相似度最高的病人心率特征节点以及呼吸频率特征节点[f_1,sim(a),f_2,sim(p)]作为[f_1,*(a),f_2,*(p)]在概率图中的表示结果，分别计算该病人体征指标数据为虚假警报的概率以及非虚假警报的概率：At the same time, this program builds an ICU alarm recognition model based on the probability graph theory, where the input of the model is the patient's sign index data, and outputs the identification result of whether the alarm is false. If it is identified as a false alarm, the alarm name and alarm level are output as null, otherwise Output the corresponding alarm name and alarm level; the ICU alarm recognition model is a graph structure G=(E, V), where E represents a node set, and the nodes include the node name of the alarm name node, judge=0, and the node of judge=1 , a variety of different patient heart rate feature nodes and respiratory rate feature nodes, V represents a set of edges, and the edges are directed edges, and the directed edge f_q → name_b indicates that an alarm occurs when the patient characteristic is f_q name The probability P(name_b |f_q ) of_b ; the input of the ICU alarm recognition model is the patient's sign index data [f_1,* (a),f_2,* (p)], which are respectively calculated using the cosine similarity algorithm The similarity between f_1,* (a) and f_2,* (p) and the patient's heart rate feature node and respiratory rate feature node in the constructed probability map, the patient's heart rate feature node and respiratory rate feature node with the highest similarity [f_{1, sim} (a), f_{2, sim} (p)] as the representation result of [f_{1, *} (a), f_{2, *} (p)] in the probability map, the data of the patient's signs and indicators are calculated as false The probability of an alarm and the probability of a non-false alarm:

P(judge＝0|f_1,sim(a),f_2,sim(p))＝w₁P(judge＝0|f_1,sim(a))w₂P(judge＝0|f_2,sim(p))P(judge=0|f_1,sim (a),f_2,sim (p))=w₁ P(judge=0|f_1,sim (a))w₂ P(judge=0|f_{2, sim} (p))

P(judge＝1|f_1,sim(a),f_2,sim(p))＝w₁P(judge＝1|f_1,sim(a))w₂P(judge＝1|f_2,sim(p))P(judge=1|f_1,sim (a),f_2,sim (p))=w₁ P(judge=1|f_1,sim (a))w₂ P(judge=1|f_{2, sim} (p))

其中：w₁为心率特征在概率图的权重参数，w₂为呼吸频率特征在概率图中的权重参数；P(judge＝0|f_1,sim(a),f_2,sim(p))为该病人体征指标数据为虚假警报的概率；Among them: w₁ is the weight parameter of the heart rate feature in the probability map, w₂ is the weight parameter of the respiratory frequency feature in the probability map; P(judge=0|f_{1, sim} (a), f_{2, sim} (p)) is the probability that the patient's sign index data is a false alarm;

P(judge＝1|f_1,sim(a),f_2,sim(p))为该病人体征指标数据为非虚假警报的概率；若P(judge＝1|f_{1, sim} (a), f_{2, sim} (p)) is the probability that the patient’s sign index data is not a false alarm; if

P(judge＝0|f_1,sim(a),f_2,sim(p))≥P(judge＝1|f_1,sim(a),f_2,sim(p))则说明该病人所对应的警报为虚假警报，则直接输出警报名称以及警报级别为null；否则计算该病人体征指标数据会导致不同警报发生的概率：P(judge＝0|f_{1, sim} (a), f_{2, sim} (p))≥P(judge＝1|f_{1, sim} (a), f_{2, sim} (p)) means that the patient If the corresponding alarm is a false alarm, then directly output the alarm name and the alarm level as null; otherwise, calculating the patient’s sign data will result in different alarm occurrence probabilities:

P(name＝m|f_1,sim(a),f_2,sim(p))＝P(name＝m|f_1,sim(a))P(name＝m|f_2,sim(p))P(name=m|f_1,sim (a),f_2,sim (p))=P(name=m|f_1,sim (a))P(name=m|f_2,sim (p) )

其中：m∈{1,2,3,…,15}，对应不同的警报名称；选取发生概率最大的警报名称以及警报对应的警报级别作为模型的输出。当医院系统检测到警报时，采集与警报相关病人的病人体征指标数据，将所采集到的病人体征指标数据输入到训练优化后的ICU警报识别模型中，模型输出是否为虚假警报，若该警报不是虚假警报，则继续输出警报名称以及警报名称所对应的处理优先级，其中处理优先级由高到中，医护人员根据处理优先级以及模型输出的警报名称进行相应的处置措施，从而实现虚拟警报的快速识别，警报名称的判断以及所对应的处理优先级的输出，减少警报疲劳，帮助医护人员实现更有针对性的救治。Among them: m∈{1,2,3,…,15}, corresponding to different alarm names; select the alarm name with the highest probability of occurrence and the corresponding alarm level as the output of the model. When the hospital system detects an alarm, collect the patient sign data of the patient related to the alarm, and input the collected patient sign data into the optimized ICU alarm recognition model after training. Whether the model output is a false alarm, if the alarm If it is not a false alarm, continue to output the alarm name and the corresponding processing priority of the alarm name, where the processing priority is from high to medium, and the medical staff will take corresponding measures according to the processing priority and the alarm name output by the model, so as to realize the virtual alarm The rapid identification of alarms, the judgment of alarm names and the output of corresponding processing priorities can reduce alarm fatigue and help medical staff to achieve more targeted treatment.

附图说明Description of drawings

图1为本发明一实施例提供的一种快速ICU虚假警报识别方法的流程示意图；Fig. 1 is a schematic flow chart of a fast ICU false alarm identification method provided by an embodiment of the present invention;

图2为本发明一实施例提供的病人ICU警报识别模型的模型结构示意图；Fig. 2 is a schematic model structure diagram of a patient ICU alarm recognition model provided by an embodiment of the present invention;

图3为本发明一实施例提供的快速ICU虚假警报识别装置的功能模块图；Fig. 3 is the functional block diagram of the fast ICU false alarm identification device provided by an embodiment of the present invention;

图4为本发明一实施例提供的实现快速ICU虚假警报识别方法的电子设备的结构示意图。FIG. 4 is a schematic structural diagram of an electronic device implementing a method for quickly identifying false ICU alarms provided by an embodiment of the present invention.

本发明目的的实现、功能特点及优点将结合实施例，参照附图做进一步说明。The realization of the purpose of the present invention, functional characteristics and advantages will be further described in conjunction with the embodiments and with reference to the accompanying drawings.

具体实施方式Detailed ways

应当理解，此处所描述的具体实施例仅仅用以解释本发明，并不用于限定本发明。It should be understood that the specific embodiments described here are only used to explain the present invention, not to limit the present invention.

本申请实施例提供一种快速ICU虚假警报识别方法。所述快速ICU虚假警报识别方法的执行主体包括但不限于服务端、终端等能够被配置为执行本申请实施例提供的该方法的电子设备中的至少一种。换言之，所述快速ICU虚假警报识别方法可以由安装在终端设备或服务端设备的软件或硬件来执行，所述软件可以是区块链平台。所述服务端包括但不限于：单台服务器、服务器集群、云端服务器或云端服务器集群等。An embodiment of the present application provides a fast ICU false alarm identification method. The subject of execution of the fast ICU false alarm identification method includes but is not limited to at least one of electronic devices such as a server and a terminal that can be configured to execute the method provided by the embodiment of the present application. In other words, the fast ICU false alarm identification method can be executed by software or hardware installed on the terminal device or server device, and the software can be a block chain platform. The server includes, but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, and the like.

实施例1：Example 1:

S1：采集病人体征数据，其中所述体征数据包括生命体征检查时序数据和入院48小时内实验室检查指标数据集合，并利用傅里叶变换方法以及小波分解方法对采集的生命体征检查时序数据进行特征提取，将提取特征与检查指标数据集合共同构成病人体征指标数据。S1: Collect patient sign data, where the sign data includes time-series data of vital sign examination and laboratory test index data set within 48 hours of admission, and use Fourier transform method and wavelet decomposition method to analyze the collected time-series data of vital sign examination Feature extraction, which combines the extracted features and the inspection index data set to form the patient's sign index data.

所述S1步骤中采集病人体征数据，包括：Collect patient sign data in the S1 step, including:

在病人进入ICU后，采集病人体征数据，其中所述体征数据包括生命体征检查时序数据和病人入院48h内的实验室检查指标数据集合，所述生命体征检查时序数据为病人的心率以及呼吸频率的时序数据，所述入院48小时内的实验室检查指标包括病人尿液的酸碱度k₁、pH值k₂、尿比重k₃、尿蛋白k₄、尿糖k₅以及管型k₆，病人血液的白细胞数量k₇、白细胞中五类细胞的数量及比例k₈、红细胞数量k₉、红细胞压积k₁₀、血红蛋白浓度k₁₁、平均红细胞体积k₁₂、血小板的数量k₁₃以及血小板压积k₁₄，所采集的病人体征数据为：After the patient enters the ICU, the patient's sign data is collected, wherein the sign data includes the time series data of the vital sign examination and the laboratory test index data set within 48 hours of the patient's admission, and the time series data of the vital sign examination is the patient's heart rate and respiratory rate. Time-series data, the laboratory test indicators within 48 hours of admission include the patient’s urine pH k₁ , pH value k₂ , urine specific gravity k₃ , urine protein k₄ , urine sugar k₅ and cast k₆ , the patient’s blood The number of white blood cells k₇ , the number and proportion of the five types of cells in white blood cells k₈ , the number of red blood cells k₉ , the hematocrit k₁₀ , the concentration of hemoglobin k₁₁ , the mean red blood cell volume k₁₂ , the number of platelets k₁₃ and the hematocrit k_14. The collected patient sign data are:

{x₁(n₁),x₂(n₂),K}{x₁ (n₁ ),x₂ (n₂ ),K}

K＝{k_num|num∈[1,14]}K＝{k_num |num∈[1,14]}

其中：in:

x₁(n₁)为病人心率的时序数据，n₁＝0,1,…,N₁-1，N₁为时序数据长度；x₁ (n₁ ) is the time-series data of the patient's heart rate, n₁ =0,1,...,N₁ -1, N₁ is the length of the time-series data;

x₂(n₂)为病人呼吸频率的时序数据，n₂＝0,1,…,N₂-1，N₂为时序数据长度；x₂ (n₂ ) is the time-series data of the patient's respiratory rate, n₂ =0,1,...,N₂ -1, N₂ is the length of the time-series data;

K为病人的实验室检查指标数据集合，包括14种的实验室检查指标。K is the patient's laboratory test index data set, including 14 kinds of laboratory test indicators.

所述S1步骤中利用傅里叶变换方法以及小波分解方法对采集的生命体征检查时序数据进行特征提取，将提取特征与检查指标数据集合共同构成病人体征指标数据，包括：In the S1 step, the Fourier transform method and the wavelet decomposition method are used to extract the features of the collected vital signs inspection time series data, and the extracted features and the inspection index data set jointly form the patient sign index data, including:

对所采集的生命体征检查时序数据进行特征提取，所述特征提取流程为：Feature extraction is performed on the collected time-series data of vital signs examination, and the process of feature extraction is as follows:

S11：利用傅里叶变换方法提取病人呼吸频率的特征f₂(p)，其中p为傅里叶变换处理的采样点数，将p设置为64，所述傅里叶变换方法的公式为：S11: Using the Fourier transform method to extract the feature f₂ (p) of the patient's respiratory frequency, where p is the number of sampling points processed by the Fourier transform, and p is set to 64. The formula of the Fourier transform method is:

其中：in:

j为虚数单位，j²＝-1；j is the imaginary number unit, j² =-1;

e为自然常数；e is a natural constant;

S12：利用小波分解方法对病人心率的特征进行提取，所述小波分解结果为：S12: Using the wavelet decomposition method to extract the characteristics of the heart rate of the patient, the result of the wavelet decomposition is:

其中：in:

q(a,x₁(n₁))为x₁(n₁)在尺度a上的小波系数，a表示小波分解过程中的最大尺度；q(a,x₁ (n₁ )) is the wavelet coefficient of x₁ (n₁ ) on scale a, and a represents the largest scale in the process of wavelet decomposition;

将小波系数作为病人心率的特征f₁(a)，若q(a，x₁(n_R))＞0.4×max{q(a，x₁(n₁))}，n₁＝0,1,…,N₁-1，则将时序位置n_R在特征f₁(a)中标记为R峰。Take the wavelet coefficient as the characteristic f₁ (a) of the patient's heart rate, if q(a, x₁ (n_R ))>0.4×max{q(a, x₁ (n₁ ))}, n₁ =0,1 ,...,N₁ -1, then the timing position n_R is marked as an R peak in the feature f₁ (a).

S2：按照步骤S1方法采集大量病人的病人体征指标数据以及对应的警报级别构成训练集A1，并利用生成对抗网络构建病人指标数据样本生成模型，利用病人指标数据样本生成模型生成大量病人体征指标数据，将所生成的数据同训练集A1构成训练集A。S2: According to the method of step S1, collect a large number of patients' patient sign index data and corresponding alarm levels to form a training set A1, and use the generative confrontation network to construct a patient index data sample generation model, and use the patient index data sample generation model to generate a large number of patient sign index data , combine the generated data with the training set A1 to form the training set A.

所述S2步骤采集大量病人的病人体征指标数据以及对应的警报级别构成训练集A1，包括：The S2 step collects patient sign index data of a large number of patients and corresponding alarm levels to form a training set A1, including:

按照步骤S1方法采集大量病人的病人体征指标数据以及对应的警报级别构成训练集A1，所述训练集A1的表示形式为：According to the method of step S1, the patient sign index data of a large number of patients and the corresponding alarm levels are collected to form a training set A1, and the expression form of the training set A1 is:

A1＝{data_u＝(f_1,u(a),f_2,u(p),judge_u,name_u,level_u)|u∈[1,U]}A1＝{data_u ＝(f_1,u (a),f_2,u (p),judge_u ,name_u ,level_u )|u∈[1,U]}

其中：in:

data_u为所采集的第u名病人的病人体征指标数据以及对应的警报级别，U表示训练集A1中病人的总数；data_u is the collected patient sign index data of the uth patient and the corresponding alarm level, and U represents the total number of patients in the training set A1;

f_1,u(a)表示所采集的第u名病人的心率特征，f_2,u(p)为所采集的第u名病人的呼吸频率特征；f_1,u (a) represents the collected heart rate feature of the uth patient, and f_2,u (p) is the collected respiratory rate feature of the uth patient;

judge_u＝{0,1}，judge_u＝0表示该病人所对应的警报为虚假警报，并将name_u以及level_u均设置为null；judge_u＝1表示该病人所对应的警报不为虚假警报；judge_u = {0,1}, judge_u = 0 indicates that the alarm corresponding to the patient is a false alarm, and set name_u and level_u to null; judge_u = 1 indicates that the alarm corresponding to the patient is not false alarm;

name_u表示该病人所对应警报的名称，name_u∈{1,2,3,…,15}，所述警报名称包括SPO2、、血压、HR(心率)、室早、心脏停搏、PR过高、Resp窒息、室颤、RONT、室早二联律、ST(ST段)、室早三联律、房早以及室速，依次对应name_u＝1至name_u＝15；name_u indicates the name of the alarm corresponding to the patient, name_u ∈ {1,2,3,...,15}, the alarm name includes SPO2, blood pressure, HR (heart rate), premature ventricular, cardiac arrest, PR excessive High, Resp apnea, ventricular fibrillation, RONT, premature ventricular bigeminy, ST (ST segment), ventricular trigeminy, atrial premature and ventricular tachycardia, corresponding to name_u = 1 to name_u = 15 in turn;

level_u表示该病人所对应的警报级别，level_u＝{1,2}，level_u＝1表示高优先级的警报，level_u＝2表示中优先级的警报；level_u indicates the alarm level corresponding to the patient, level_u = {1,2}, level_u = 1 indicates a high priority alarm, level_u = 2 indicates a medium priority alarm;

在本发明一个具体实施例中，如血压、HR(心率)、心脏停搏及Resp窒息其中某一项指标突然异常，而其他指标又是正常的情况下，则可以快速判定为虚假警报。导联脱落是较多造成虚假警报的原因，根据多种体征参数的综合判断，如一个患者血压正常、心率正常，突然出现血压为0的情况，一定是出现了虚假警报；可以通过体征的综合指标判断，体征本身是相互关联的，不会全部体征都正常突然出现没有呼吸的情况。In a specific embodiment of the present invention, if one of the indicators such as blood pressure, HR (heart rate), cardiac arrest and Resp asphyxia is suddenly abnormal, while other indicators are normal, it can be quickly judged as a false alarm. Lead off is the cause of many false alarms. According to the comprehensive judgment of various physical signs and parameters, such as a patient with normal blood pressure and normal heart rate, and suddenly the blood pressure is 0, there must be false alarms; Judging by the indicators, the signs themselves are interrelated, and not all the signs are normal and suddenly there is no breathing.

在本发明一个具体实施例中，中优先级的警报有SPO2、房早、室早、心率、PR过高、RONT、血压、导联脱落、ST段、室早二联律、室早三联律以及Rsep窒息，高优先级的警报警报有室速、室颤以及心脏停搏。In a specific embodiment of the present invention, the alarms with medium priority include SPO2, premature ventricular, premature ventricular, heart rate, high PR, RONT, blood pressure, lead off, ST segment, premature ventricular bigeminy, premature ventricular triplet As well as Rsep apnea, the high priority alarms are ventricular tachycardia, ventricular fibrillation, and cardiac arrest.

所述S2步骤中利用生成对抗网络构建病人指标数据样本生成模型，利用病人指标数据样本生成模型生成大量病人体征指标数据，将所生成的数据同训练集A1构成训练集A，包括：In the S2 step, use the generative confrontation network to construct the patient index data sample generation model, utilize the patient index data sample generation model to generate a large number of patient sign index data, and form the training set A with the generated data with the training set A1, including:

利用生成对抗网络构建病人指标数据样本生成模型，其中所述病人指标数据样本生成模型包括编码器G_e(·)和解码器G_d(·)，所述编码器的输入值为训练集A1中的病人体征指标数据，通过利用编码器G_e(·)对输入的病人体征指标数据进行编码，利用解码器G_d(·)对编码结果进行解码，得到重构后的病人体征指标数据，将解码结果作为模型的样本生成结果；Utilize the generative confrontation network to construct the patient index data sample generation model, wherein the patient index data sample generation model includes an encoder G_e (·) and a decoder G_d (·), and the input value of the encoder is in the training set A1 The patient’s sign index data, by using the encoder G_e (·) to encode the input patient’s sign index data, and using the decoder G_d (·) to decode the encoded result, the reconstructed patient sign index data is obtained, and the The decoding result is used as the sample generation result of the model;

计算训练集A1中judge＝0的病人所占的比例，若该比例小于0.6，则选取judge＝0的病人所对应的病人体征指标数据作为病人指标数据样本生成模型中编码器的输入，编码器对输入结果进行编码后，利用解码器对编码结果进行解码，得到重构后的病人体征指标数据，将重构得到的病人体征指标数据的judge参数设置为0，所述样本生成后judge＝0的病人所占的比例达到0.6，得到训练集A2；Calculate the proportion of patients with judge=0 in the training set A1, if the proportion is less than 0.6, select the patient sign index data corresponding to the patient with judge=0 as the input of the encoder in the patient index data sample generation model, the encoder After encoding the input result, use the decoder to decode the encoded result to obtain the reconstructed patient sign index data, set the judge parameter of the reconstructed patient sign index data to 0, and judge=0 after the sample is generated The proportion of patients in the group reaches 0.6, and the training set A2 is obtained;

计算训练集A2中不同警报名称的病人数，选取病人数最少的5个警报名称，将所选取的警报名称下的病人体征指标数据以及对应的警报级别分别构成五个数据集A2_min1,A2_min2，A2_min3，A2_min4以及A2_min5，其中每个数据集中病人体征指标数据所对应的警报名称相同，选取任意数据集的所有病人体征指标数据作为病人指标数据样本生成模型中编码器的输入，编码器对输入结果进行编码后，利用解码器对编码结果进行解码，得到重构后的病人体征指标数据，向该重构后的病人体征指标数据添加该数据集所对应的警报名称，以及警报名称所对应的警报级别，并将judge参数设置为1，得到生成样本，所述生成样本的生成数量为(num_A2_max-num_A2_minj′)×β，其中num_A2_max为训练集A2中出现频率最高的警报名称的病人人数，j′∈{1,2,3,4,5}，num_A2_minj′为数据集A2_minj′内的病人人数，β为样本生成参数，将其设置为0.3，将所生成的数据同训练集A2构成训练集A；Calculate the number of patients with different alarm names in the training set A2, select the five alarm names with the least number of patients, and form five data sets A2_min1 and A2_min2 respectively with the patient sign index data under the selected alarm names and the corresponding alarm levels , A2_min3 , A2_min4 and A2_min5 , the alarm names corresponding to the patient sign data in each data set are the same, select all the patient sign data in any data set as the input of the encoder in the patient index data sample generation model, encode After the input result is encoded by the decoder, the encoded result is decoded by the decoder to obtain the reconstructed patient sign data, and the alarm name corresponding to the data set and the alarm name are added to the reconstructed patient sign data The corresponding alarm level, and set the judge parameter to 1 to obtain the generated samples. The generated number of the generated samples is (num_A2_max -num_A2_minj′ )×β, where num_A2_max is the alarm with the highest frequency in the training set A2 The number of patients with the name, j′∈{1,2,3,4,5}, num_A2_minj′ is the number of patients in the data set A2_minj′ , β is the sample generation parameter, set it to 0.3, and the generated The data and the training set A2 form the training set A;

构建病人指标数据样本生成模型中编码器G_e(·)和解码器G_d(·)的训练目标函数：Construct the training objective function of encoder G_e ( ) and decoder G_d ( ) in the patient indicator data sample generation model:

其中：in:

f为病人指标数据样本生成模型中编码器G_e(·)的输入值，包括病人的心率特征以及呼吸频率特征，p(f)为f的分布；f is the input value of encoder G_e (·) in the patient index data sample generation model, including the patient's heart rate characteristics and respiratory frequency characteristics, p(f) is the distribution of f;

E(·)表示期望值计算；E( ) means expected value calculation;

为编码器G_e(·)的参数，

为解码器G_d(·)的参数；

is the parameter of the encoder G_e (·),

is the parameter of decoder G_d (·);

||·||₂为L2范数；||·||₂ is the L2 norm;

将多组病人的心率特征以及呼吸频率特征输入到训练目标函数中，使得训练目标函数达到最小的模型参数即为训练优化得到的模型参数，利用训练优化后的病人指标数据样本生成模型进行数据生成。Input the heart rate characteristics and respiratory frequency characteristics of multiple groups of patients into the training objective function, so that the model parameters that minimize the training objective function are the model parameters obtained by training optimization, and use the optimized patient index data sample generation model for data generation .

S3：基于概率图理论构建ICU警报识别模型，其中所述模型的输入为病人体征指标数据，输出是否虚假警报的识别结果，以及对应的警报名称和警报级别。S3: Construct an ICU alarm recognition model based on the probability graph theory, wherein the input of the model is the patient's sign index data, and output the identification result of whether there is a false alarm, as well as the corresponding alarm name and alarm level.

所述S3步骤中基于概率图理论构建ICU警报识别模型，包括：In the S3 step, the ICU alarm recognition model is constructed based on the probability map theory, including:

基于概率图理论构建ICU警报识别模型，参见图2所示，为本发明一实施例提供的病人ICU警报识别模型的模型结构示意图，其中f1、f2分别为心率特征节点以及呼吸频率特征节点，name为警报名称节点，judge＝0or1为虚假警报节点，其中所述模型的输入为病人体征指标数据，输出是否虚假警报的识别结果，若识别为虚假警报则输出警报名称以及警报级别为null，否则输出对应的警报名称和警报级别；Construct the ICU alarm recognition model based on the probability graph theory, as shown in Figure 2, which is a schematic diagram of the model structure of the patient ICU alarm recognition model provided by an embodiment of the present invention, wherein f1 and f2 are respectively the heart rate feature node and the respiratory rate feature node, name It is an alarm name node, and judge=0or1 is a false alarm node, wherein the input of the model is the patient’s sign index data, and outputs the identification result of whether it is a false alarm, if it is identified as a false alarm, then output the alarm name and the alarm level is null, otherwise output Corresponding alert name and alert level;

所述ICU警报识别模型为图结构G＝(E,V)，其中E表示节点集合，节点包括警报名称节点name、judge＝0的节点、judge＝1的节点、多种不同的病人心率特征节点以及呼吸频率特征节点，V表示边集合，所述边为有向边，有向边f_q→name_b表示在病人特征为f_q的情况下，发生警报name_b的概率P(name_b|f_q)；The ICU alarm recognition model is a graph structure G=(E, V), wherein E represents a node set, and the nodes include the alarm name node name, the node of judge=0, the node of judge=1, and a variety of different patient heart rate characteristic nodes And the respiratory frequency feature node,_V represents the edge set, the edge is a directed edge, and the directed edge f_q →name_b represents the probability P(name_b| f_q );

所述ICU警报识别模型的输入为病人体征指标数据[f_1,*(a),f_2,*(p)]，利用余弦相似度算法分别计算f_1,*(a)以及f_2,*(p)与所构建概率图中病人心率特征节点以及呼吸频率特征节点的相似度，将相似度最高的病人心率特征节点以及呼吸频率特征节点[f_1,sim(a),f_2,sim(p)]作为[f_1,*(a),f_2,*(p)]在概率图中的表示结果，分别计算该病人体征指标数据为虚假警报的概率以及非虚假警报的概率：The input of the ICU alarm recognition model is patient sign index data [f_1,* (a), f_2,* (p)], using the cosine similarity algorithm to calculate f_1,* (a) and f_2,* (p) The similarity with the patient's heart rate feature node and respiratory rate feature node in the constructed probability map, the patient's heart rate feature node and respiratory rate feature node with the highest similarity [f_1,sim (a),f_2,sim ( p)] as the expression result of [f_1,* (a),f_2,* (p)] in the probability map, respectively calculate the probability of the patient's sign index data being a false alarm and the probability of a non-false alarm:

P(judge＝0|f_1,sim(a),f_2,sim(p))＝w₁P(judge＝0|f_1,sim(a))w₂P(judge＝0|f_2,sim(p))P(judge=0|f_1,sim (a),f_2,sim (p))=w₁ P(judge=0|f_1,sim (a))w₂ P(judge=0|f_{2, sim} (p))

P(judge＝1|f_1,sim(a),f_2,sim(p))＝w₁P(judge＝1|f_1,sim(a))w₂P(judge＝1|f_2,sim(p))P(judge=1|f_1,sim (a),f_2,sim (p))=w₁ P(judge=1|f_1,sim (a))w₂ P(judge=1|f_{2, sim} (p))

其中：in:

w₁为心率特征在概率图的权重参数，w₂为呼吸频率特征在概率图中的权重参数；w₁ is the weight parameter of the heart rate feature in the probability map, and w₂ is the weight parameter of the respiratory frequency feature in the probability map;

P(judge＝0|f_1,sim(a),f_2,sim(p))为该病人体征指标数据为虚假警报的概率；P(judge＝0|f_{1, sim} (a), f_{2, sim} (p)) is the probability that the patient's sign index data is a false alarm;

P(judge＝1|f_1,sim(a),f_2,sim(p))为该病人体征指标数据为非虚假警报的概率；P(judge=1|f_{1, sim} (a), f_{2, sim} (p)) is the probability that the patient's sign index data is not a false alarm;

若P(judge＝0|f_1,sim(a),f_2,sim(p))≥P(judge＝1|f_1,sim(a),f_2,sim(p))则说明该病人所对应的警报为虚假警报，则直接输出警报名称以及警报级别为null；If P(judge=0|f_{1, sim} (a), f_{2, sim} (p))≥P(judge=1|f_{1, sim} (a), f_{2, sim} (p)), then the patient If the corresponding alarm is a false alarm, then directly output the alarm name and alarm level as null;

否则计算该病人体征指标数据会导致不同警报发生的概率：Otherwise, calculating the patient's sign data will result in the probability of different alarm occurrences:

P(name＝m|f_1,sim(a),f_2,sim(p))＝P(name＝m|f_1,sim(a))P(name＝m|f_2,sim(p))P(name=m|f_1,sim (a),f_2,sim (p))=P(name=m|f_1,sim (a))P(name=m|f_2,sim (p) )

其中：in:

m∈{1,2,3,…,15}，对应不同的警报名称；m∈{1,2,3,…,15}, corresponding to different alarm names;

选取发生概率最大的警报名称以及警报对应的警报级别作为模型的输出。Select the alarm name with the highest probability of occurrence and the alarm level corresponding to the alarm as the output of the model.

S4：基于训练集A，利用分裂梯度算法对所构建的ICU警报识别模型进行快速优化，得到训练优化后的ICU警报识别模型。S4: Based on the training set A, use the split gradient algorithm to quickly optimize the constructed ICU alarm recognition model, and obtain the optimized ICU alarm recognition model after training.

所述S4步骤中利用分裂梯度算法对所构建的ICU警报识别模型进行快速优化，包括：In the S4 step, the split gradient algorithm is used to quickly optimize the built ICU alarm recognition model, including:

利用训练集A中的样本数据对所构建的ICU警报识别模型进行快速优化，所述ICU警报识别模型的优化流程为：Using the sample data in the training set A to quickly optimize the built ICU alarm recognition model, the optimization process of the ICU alarm recognition model is:

S41：利用K-means算法将训练集A所有病人体征指标数据进行聚类处理，将每类的聚类中心作为概率图中病人心率特征节点以及呼吸频率特征节点，病人心率特征节点以及呼吸频率特征节点与警报名称节点之间的条件概率，即为聚类中心邻近病人体征指标数据与警报名称节点条件概率的累乘，并忽略条件概率为0的点；S41: Use the K-means algorithm to cluster all the patient sign data in the training set A, and use the cluster center of each class as the patient's heart rate feature node and respiratory rate feature node, patient heart rate feature node and respiratory rate feature in the probability map The conditional probability between the node and the alarm name node is the cumulative multiplication of the patient’s sign index data adjacent to the cluster center and the conditional probability of the alarm name node, and the point with a conditional probability of 0 is ignored;

S42：构建ICU警报识别模型训练的目标函数：S42: Construct the objective function of ICU alarm recognition model training:

其中：in:

f_z为训练集A中第z个病人的病人体征指标数据，Z为训练集A中病人的总数；f_z is the patient sign index data of the zth patient in the training set A, and Z is the total number of patients in the training set A;

w₁,w₂为待训练优化的权重参数，令w＝[w₁,w₂]；w₁ , w₂ are weight parameters to be optimized for training, let w=[w₁ ,w₂ ];

S43：设置当前分裂梯度算法迭代次数为n，分裂梯度算法参数ε>0，σ∈(0,1)，设置

S43: Set the number of iterations of the current split gradient algorithm to n, the split gradient algorithm parameters ε>0, σ∈(0,1), set

S44：若

终止算法迭代流程，输出当前的权重参数wⁿ，其中wⁿ表示第n次迭代时的权重参数，将训练得到的权重参数作为ICU警报识别模型的权重参数，得到训练优化后的ICU警报识别模型；否则转向下一步；S44: if

Terminate the algorithm iteration process, output the current weight parameter wⁿ , where wⁿ represents the weight parameter at the nth iteration, and use the weight parameter obtained from training as the weight parameter of the ICU alarm recognition model to obtain the optimized ICU alarm recognition model after training ; Otherwise go to the next step;

S45：若

则令d_n＝g_n，否则令d_n＝-g_n；S45: if

Then let d_n =g_n , otherwise let d_n =-g_n ;

S46：计算第n+1次迭代的权重参数：S46: Calculate the weight parameter of the n+1th iteration:

wⁿ⁺¹＝wⁿ+λ_nd_nwⁿ⁺¹ ＝wⁿ +λ_n d_n

令n＝n+1，返回步骤S44。If n=n+1, return to step S44.

S5：当医院系统检测到警报时，采集与警报相关病人的病人体征指标数据，将所采集到的病人体征指标数据输入到训练优化后的ICU警报识别模型中，模型输出是否为虚假警报，若该警报不是虚假警报，则继续输出警报名称以及处理优先级，医护人员根据模型输出结果进行相应的处置措施。S5: When the hospital system detects an alarm, collect the patient sign data of the patient related to the alarm, and input the collected patient sign data into the optimized ICU alarm recognition model after training. Whether the model output is a false alarm, if If the alarm is not a false alarm, continue to output the alarm name and processing priority, and the medical staff will take corresponding disposal measures according to the output results of the model.

所述S5步骤中当医院系统检测到警报时，采集与警报相关病人的病人体征指标数据，将所采集到的病人体征指标数据输入到训练优化后的ICU警报识别模型中，模型输出是否为虚假警报，若该警报不是虚假警报，则继续输出警报名称以及处理优先级，包括：In the S5 step, when the hospital system detects an alarm, collect the patient sign index data of the patient related to the alarm, and input the collected patient sign index data into the optimized ICU alarm recognition model after training, whether the model output is false Alarm, if the alarm is not a false alarm, continue to output the alarm name and processing priority, including:

当医院系统检测到警报时，按照步骤S1方法采集与警报相关病人的病人体征指标数据，将所采集到的病人体征指标数据输入到训练优化后的ICU警报识别模型中，模型输出是否为虚假警报，若该警报不是虚假警报，则继续输出警报名称以及警报名称所对应的处理优先级，其中处理优先级由高到中，医护人员根据处理优先级以及模型输出的警报名称进行相应的处置措施，减少警报疲劳。When the hospital system detects an alarm, collect the patient sign data of the patient related to the alarm according to the method of step S1, and input the collected patient sign data into the optimized ICU alarm recognition model after training, whether the model output is a false alarm , if the alarm is not a false alarm, then continue to output the alarm name and the processing priority corresponding to the alarm name, where the processing priority is from high to medium, and the medical staff will take corresponding measures according to the processing priority and the alarm name output by the model, Reduced alert fatigue.

实施例2：Example 2:

如图3所示，是本发明一实施例提供的快速ICU虚假警报识别装置的功能模块图，其可以实现实施例1中的快速ICU虚假警报识别方法。As shown in FIG. 3 , it is a functional block diagram of a rapid ICU false alarm identification device provided by an embodiment of the present invention, which can realize the rapid ICU false alarm identification method inEmbodiment 1.

本发明所述快速ICU虚假警报识别装置100可以安装于电子设备中。根据实现的功能，所述快速ICU虚假警报识别装置可以包括特征提取模块101、训练集获取装置102及ICU警报识别装置103。本发明所述模块也可以称之为单元，是指一种能够被电子设备处理器所执行，并且能够完成固定功能的一系列计算机程序段，其存储在电子设备的存储器中。The rapid ICU falsealarm identification device 100 of the present invention can be installed in electronic equipment. According to the realized functions, the fast ICU false alarm recognition device may include afeature extraction module 101 , a training setacquisition device 102 and an ICUalarm recognition device 103 . The module in the present invention can also be called a unit, which refers to a series of computer program segments that can be executed by the processor of the electronic device and can complete fixed functions, and are stored in the memory of the electronic device.

特征提取模块101，用于采集病人体征数据，利用傅里叶变换方法以及小波分解方法对采集的生命体征检查时序数据进行特征提取，将提取特征与检查指标数据集合共同构成病人体征指标数据；Thefeature extraction module 101 is used to collect the patient's vital sign data, and use the Fourier transform method and the wavelet decomposition method to perform feature extraction on the collected vital sign examination time-series data, and combine the extracted features and the examination index data set to form the patient's physical sign index data;

训练集获取装置102，用于采集大量病人的病人体征指标数据以及对应的警报级别构成训练集A1，并利用生成对抗网络构建病人指标数据样本生成模型，利用病人指标数据样本生成模型生成大量病人体征指标数据，将所生成的数据同训练集A1构成训练集A；The training setacquisition device 102 is used to collect patient sign index data and corresponding alarm levels of a large number of patients to form a training set A1, and use the generative confrontation network to construct a patient index data sample generation model, and use the patient index data sample generation model to generate a large number of patient signs Index data, the generated data and the training set A1 form the training set A;

ICU警报识别装置103，用于基于概率图理论构建ICU警报识别模型，基于训练集A，利用分裂梯度算法对所构建的ICU警报识别模型进行快速优化，将所采集到的病人体征指标数据输入到训练优化后的ICU警报识别模型中，模型输出是否为虚假警报，若该警报不是虚假警报，则继续输出警报名称以及处理优先级。The ICUalarm identification device 103 is used to construct an ICU alarm identification model based on the probability map theory, based on the training set A, use the split gradient algorithm to quickly optimize the constructed ICU alarm identification model, and input the collected patient sign index data into In the optimized ICU alarm recognition model after training, whether the model output is a false alarm, if the alarm is not a false alarm, continue to output the alarm name and processing priority.

详细地，本发明实施例中所述快速ICU虚假警报识别装置100中的所述各模块在使用时采用与上述的图1中所述的快速ICU虚假警报识别方法一样的技术手段，并能够产生相同的技术效果，这里不再赘述。In detail, the modules in the fast ICU falsealarm identification device 100 in the embodiment of the present invention use the same technical means as the above-mentioned fast ICU false alarm identification method described in FIG. 1 , and can generate The same technical effect will not be repeated here.

实施例3：Example 3:

如图4所示，是本发明一实施例提供的实现快速ICU虚假警报识别方法的电子设备的结构示意图。As shown in FIG. 4 , it is a schematic structural diagram of an electronic device for implementing a rapid ICU false alarm identification method provided by an embodiment of the present invention.

所述电子设备1可以包括处理器10、存储器11和总线，还可以包括存储在所述存储器11中并可在所述处理器10上运行的计算机程序，如快速ICU虚假警报识别程序12。Theelectronic device 1 may include aprocessor 10 , a memory 11 and a bus, and may also include a computer program stored in the memory 11 and operable on theprocessor 10 , such as a fast ICU false alarm recognition program 12 .

其中，所述存储器11至少包括一种类型的可读存储介质，所述可读存储介质包括闪存、移动硬盘、多媒体卡、卡型存储器(例如：SD或DX存储器等)、磁性存储器、磁盘、光盘等。所述存储器11在一些实施例中可以是电子设备1的内部存储单元，例如该电子设备1的移动硬盘。所述存储器11在另一些实施例中也可以是电子设备1的外部存储设备，例如电子设备1上配备的插接式移动硬盘、智能存储卡(Smart Media Card，SMC)、安全数字(SecureDigital，SD)卡、闪存卡(Flash Card)等。进一步地，所述存储器11还可以既包括电子设备1的内部存储单元也包括外部存储设备。所述存储器11不仅可以用于存储安装于电子设备1的应用软件及各类数据，例如快速ICU虚假警报识别程序12的代码等，还可以用于暂时地存储已经输出或者将要输出的数据。Wherein, the memory 11 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, mobile hard disk, multimedia card, card type memory (for example: SD or DX memory, etc.), magnetic memory, magnetic disk, CD etc. The storage 11 may be an internal storage unit of theelectronic device 1 in some embodiments, such as a mobile hard disk of theelectronic device 1 . The memory 11 can also be an external storage device of theelectronic device 1 in other embodiments, such as a plug-in mobile hard disk equipped on theelectronic device 1, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, flash memory card (Flash Card), etc. Further, the memory 11 may also include both an internal storage unit of theelectronic device 1 and an external storage device. The memory 11 can not only be used to store application software and various data installed in theelectronic device 1, such as the code of the fast ICU false alarm recognition program 12, but also can be used to temporarily store data that has been output or will be output.

所述处理器10在一些实施例中可以由集成电路组成，例如可以由单个封装的集成电路所组成，也可以是由多个相同功能或不同功能封装的集成电路所组成，包括一个或者多个中央处理器(Central Processing unit，CPU)、微处理器、数字处理芯片、图形处理器及各种控制芯片的组合等。所述处理器10是所述电子设备的控制核心(Control Unit)，利用各种接口和线路连接整个电子设备的各个部件，通过运行或执行存储在所述存储器11内的程序或者模块(快速ICU虚假警报识别程序等)，以及调用存储在所述存储器11内的数据，以执行电子设备1的各种功能和处理数据。In some embodiments, theprocessor 10 may be composed of integrated circuits, for example, may be composed of a single packaged integrated circuit, or may be composed of multiple integrated circuits with the same function or different functions, including one or more Combination of central processing unit (Central Processing unit, CPU), microprocessor, digital processing chip, graphics processor and various control chips, etc. Describedprocessor 10 is the control core (Control Unit) of described electronic equipment, utilizes various interfaces and lines to connect each part of whole electronic equipment, by running or executing the program or module (fast ICU) stored in the memory 11 False alarm recognition program, etc.), and call the data stored in the memory 11 to execute various functions of theelectronic device 1 and process data.

所述总线可以是外设部件互连标准(peripheral component interconnect，简称PCI)总线或扩展工业标准结构(extended industry standard architecture，简称EISA)总线等。该总线可以分为地址总线、数据总线、控制总线等。所述总线被设置为实现所述存储器11以及至少一个处理器10等之间的连接通信。The bus may be a peripheral component interconnect (PCI for short) bus or an extended industry standard architecture (EISA for short) bus or the like. The bus can be divided into address bus, data bus, control bus and so on. The bus is configured to realize connection and communication between the memory 11 and at least oneprocessor 10 and the like.

图4仅示出了具有部件的电子设备，本领域技术人员可以理解的是，图4示出的结构并不构成对所述电子设备1的限定，可以包括比图示更少或者更多的部件，或者组合某些部件，或者不同的部件布置。FIG. 4 only shows an electronic device with components. Those skilled in the art can understand that the structure shown in FIG. 4 does not constitute a limitation to theelectronic device 1, and may include fewer or more components, or combinations of certain components, or different arrangements of components.

例如，尽管未示出，所述电子设备1还可以包括给各个部件供电的电源(比如电池)，优选地，电源可以通过电源管理装置与所述至少一个处理器10逻辑相连，从而通过电源管理装置实现充电管理、放电管理、以及功耗管理等功能。电源还可以包括一个或一个以上的直流或交流电源、再充电装置、电源故障检测电路、电源转换器或者逆变器、电源状态指示器等任意组件。所述电子设备1还可以包括多种传感器、蓝牙模块、Wi-Fi模块等，在此不再赘述。For example, although not shown, theelectronic device 1 can also include a power supply (such as a battery) for supplying power to various components. Preferably, the power supply can be logically connected to the at least oneprocessor 10 through a power management device, so that the power supply can be controlled by power management. The device implements functions such as charge management, discharge management, and power consumption management. The power supply may also include one or more DC or AC power supplies, recharging devices, power failure detection circuits, power converters or inverters, power status indicators and other arbitrary components. Theelectronic device 1 may also include various sensors, bluetooth modules, Wi-Fi modules, etc., which will not be repeated here.

进一步地，所述电子设备1还可以包括网络接口，可选地，所述网络接口可以包括有线接口和/或无线接口(如WI-FI接口、蓝牙接口等)，通常用于在该电子设备1与其他电子设备之间建立通信连接。Further, theelectronic device 1 may also include a network interface, optionally, the network interface may include a wired interface and/or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), which are usually used in theelectronic device 1 Establish a communication connection with other electronic devices.

可选地，该电子设备1还可以包括用户接口，用户接口可以是显示器(Display)、输入单元(比如键盘(Keyboard))，可选地，用户接口还可以是标准的有线接口、无线接口。可选地，在一些实施例中，显示器可以是LED显示器、液晶显示器、触控式液晶显示器以及OLED(Organic Light-Emitting Diode，有机发光二极管)触摸器等。其中，显示器也可以适当的称为显示屏或显示单元，用于显示在电子设备1中处理的信息以及用于显示可视化的用户界面。Optionally, theelectronic device 1 may further include a user interface, which may be a display (Display) or an input unit (such as a keyboard (Keyboard)). Optionally, the user interface may also be a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode, Organic Light-Emitting Diode) touch panel, and the like. Wherein, the display may also be appropriately called a display screen or a display unit, and is used for displaying information processed in theelectronic device 1 and for displaying a visualized user interface.

应该了解，所述实施例仅为说明之用，在专利申请范围上并不受此结构的限制。It should be understood that the embodiments are only for illustration, and are not limited by the structure in terms of the scope of the patent application.

所述电子设备1中的所述存储器11存储的快速ICU虚假警报识别程序12是多个指令的组合，在所述处理器10中运行时，可以实现：The fast ICU false alarm recognition program 12 stored in the memory 11 in theelectronic device 1 is a combination of multiple instructions. When running in theprocessor 10, it can realize:

采集病人体征数据，其中所述体征数据包括生命体征检查时序数据和入院48小时内实验室检查指标数据集合，并利用傅里叶变换方法以及小波分解方法对采集的生命体征检查时序数据进行特征提取，将提取特征与检查指标数据集合共同构成病人体征指标数据；Collect patient sign data, where the sign data includes vital sign examination time series data and laboratory test index data set within 48 hours of admission, and use Fourier transform method and wavelet decomposition method to perform feature extraction on the collected vital sign examination time series data , combining the extracted features and the inspection index data set to form the patient sign index data;

采集大量病人的病人体征指标数据以及对应的警报级别构成训练集A1，并利用生成对抗网络构建病人指标数据样本生成模型，利用病人指标数据样本生成模型生成大量病人体征指标数据，将所生成的数据同训练集A1构成训练集A；Collect patient sign index data of a large number of patients and the corresponding alarm levels to form the training set A1, and use the generative confrontation network to construct a patient index data sample generation model, use the patient index data sample generation model to generate a large number of patient sign index data, and convert the generated data to Constitute the training set A with the training set A1;

基于概率图理论构建ICU警报识别模型，其中所述模型的输入为病人体征指标数据，输出是否虚假警报的识别结果，以及对应的警报名称和警报级别；Constructing an ICU alarm recognition model based on the probability graph theory, wherein the input of the model is patient sign index data, and the output is a false alarm identification result, and the corresponding alarm name and alarm level;

基于训练集A，利用分裂梯度算法对所构建的ICU警报识别模型进行快速优化，得到训练优化后的ICU警报识别模型；Based on the training set A, use the split gradient algorithm to quickly optimize the constructed ICU alarm recognition model, and obtain the optimized ICU alarm recognition model after training;

当医院系统检测到警报时，采集与警报相关病人的病人体征指标数据，将所采集到的病人体征指标数据输入到训练优化后的ICU警报识别模型中，模型输出是否为虚假警报，若该警报不是虚假警报，则继续输出警报名称以及处理优先级，医护人员根据模型输出结果进行相应的处置措施。When the hospital system detects an alarm, collect the patient sign data of the patient related to the alarm, and input the collected patient sign data into the optimized ICU alarm recognition model after training. Whether the model output is a false alarm, if the alarm If it is not a false alarm, continue to output the alarm name and processing priority, and the medical staff will take corresponding measures according to the output results of the model.

具体地，所述处理器10对上述指令的具体实现方法可参考图1至图4对应实施例中相关步骤的描述，在此不赘述。Specifically, for the specific implementation method of the above instructions by theprocessor 10, reference may be made to the description of relevant steps in the embodiments corresponding to FIG. 1 to FIG. 4 , and details are not repeated here.

需要说明的是，上述本发明实施例序号仅仅为了描述，不代表实施例的优劣。并且本文中的术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含，从而使得包括一系列要素的过程、装置、物品或者方法不仅包括那些要素，而且还包括没有明确列出的其他要素，或者是还包括为这种过程、装置、物品或者方法所固有的要素。在没有更多限制的情况下，由语句“包括一个……”限定的要素，并不排除在包括该要素的过程、装置、物品或者方法中还存在另外的相同要素。It should be noted that the serial numbers of the above embodiments of the present invention are only for description, and do not represent the advantages and disadvantages of the embodiments. And herein the term "comprises", "comprises" or any other variation thereof is intended to cover a non-exclusive inclusion such that a process, apparatus, article or method comprising a set of elements includes not only those elements, but also includes the elements not expressly included. other elements listed, or also include elements inherent in the process, apparatus, article, or method. Without further limitations, an element defined by the phrase "comprising a..." does not preclude the presence of additional identical elements in the process, apparatus, article or method comprising that element.

通过以上的实施方式的描述，本领域的技术人员可以清楚地了解到上述实施例方法可借助软件加必需的通用硬件平台的方式来实现，当然也可以通过硬件，但很多情况下前者是更佳的实施方式。基于这样的理解，本发明的技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来，该计算机软件产品存储在如上所述的一个存储介质(如ROM/RAM、磁碟、光盘)中，包括若干指令用以使得一台终端设备(可以是手机，计算机，服务器，或者网络设备等)执行本发明各个实施例所述的方法。Through the description of the above embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus a necessary general-purpose hardware platform, and of course also by hardware, but in many cases the former is better implementation. Based on such an understanding, the technical solution of the present invention can be embodied in the form of a software product in essence or in other words, the part that contributes to the prior art, and the computer software product is stored in a storage medium (such as ROM/RAM) as described above. , magnetic disk, optical disk), including several instructions to enable a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in various embodiments of the present invention.

以上仅为本发明的优选实施例，并非因此限制本发明的专利范围，凡是利用本发明说明书及附图内容所作的等效结构或等效流程变换，或直接或间接运用在其他相关的技术领域，均同理包括在本发明的专利保护范围内。The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process conversion made by using the description of the present invention and the contents of the accompanying drawings, or directly or indirectly used in other related technical fields , are all included in the scope of patent protection of the present invention in the same way.

Claims (8)

Translated fromChinese

1.一种快速ICU虚假警报识别方法，其特征在于，所述方法包括：1. A fast ICU false alarm identification method, is characterized in that, described method comprises:S1：采集病人体征数据，其中所述体征数据包括生命体征检查时序数据和入院48小时内实验室检查指标数据集合，并利用傅里叶变换方法以及小波分解方法对采集的生命体征检查时序数据进行特征提取，将提取特征与检查指标数据集合共同构成病人体征指标数据；S1: Collect patient sign data, where the sign data includes time-series data of vital sign examination and laboratory test index data set within 48 hours of admission, and use Fourier transform method and wavelet decomposition method to analyze the collected time-series data of vital sign examination Feature extraction, which combines the extracted features and the inspection index data set to form the patient's sign index data;S2：按照步骤S1方法采集大量病人的病人体征指标数据以及对应的警报级别构成训练集A1，并利用生成对抗网络构建病人指标数据样本生成模型，利用病人指标数据样本生成模型生成大量病人体征指标数据，将所生成的数据同训练集A1构成训练集A；S2: According to the method of step S1, collect a large number of patients' patient sign index data and corresponding alarm levels to form a training set A1, and use the generative confrontation network to construct a patient index data sample generation model, and use the patient index data sample generation model to generate a large number of patient sign index data , form the training set A with the generated data and the training set A1;S3：基于概率图理论构建ICU警报识别模型，其中所述模型的输入为病人体征指标数据，输出是否虚假警报的识别结果，以及对应的警报名称和警报级别；S3: Construct an ICU alarm recognition model based on the probability graph theory, wherein the input of the model is patient sign index data, output the identification result of false alarm, and the corresponding alarm name and alarm level;S4：基于训练集A，利用分裂梯度算法对所构建的ICU警报识别模型进行快速优化，得到训练优化后的ICU警报识别模型，其中所述分裂梯度算法，包括：S4: Based on the training set A, use the split gradient algorithm to quickly optimize the built ICU alarm recognition model, and obtain the optimized ICU alarm recognition model after training, wherein the split gradient algorithm includes:利用训练集A中的样本数据对所构建的ICU警报识别模型进行快速优化，所述ICU警报识别模型的优化流程为：Using the sample data in the training set A to quickly optimize the built ICU alarm recognition model, the optimization process of the ICU alarm recognition model is:S41：利用K-means算法将训练集A所有病人体征指标数据进行聚类处理，将每类的聚类中心作为概率图中病人心率特征节点以及呼吸频率特征节点，病人心率特征节点以及呼吸频率特征节点与警报名称节点之间的条件概率，即为聚类中心邻近病人体征指标数据与警报名称节点条件概率的累乘，并忽略条件概率为0的点；S41: Use the K-means algorithm to cluster all the patient sign data in the training set A, and use the cluster center of each class as the patient's heart rate feature node and respiratory rate feature node, patient heart rate feature node and respiratory rate feature in the probability map The conditional probability between the node and the alarm name node is the cumulative multiplication of the patient’s sign index data adjacent to the cluster center and the conditional probability of the alarm name node, and the point with a conditional probability of 0 is ignored;S42：构建ICU警报识别模型训练的目标函数：S42: Construct the objective function of ICU alarm recognition model training:

其中：in:f_z为训练集A中第z个病人的病人体征指标数据，Z为训练集A中病人的总数；f_z is the patient sign index data of the zth patient in the training set A, and Z is the total number of patients in the training set A;w₁,w₂为待训练优化的权重参数，令w＝[w₁,w₂]；w₁ , w₂ are weight parameters to be optimized for training, let w=[w₁ ,w₂ ];S43：设置当前分裂梯度算法迭代次数为n，分裂梯度算法参数ε>0，σ∈(0,1)，设置

S43: Set the number of iterations of the current split gradient algorithm to n, the split gradient algorithm parameters ε>0, σ∈(0,1), set

S44：若

终止算法迭代流程，输出当前的权重参数wⁿ，其中wⁿ表示第n次迭代时的权重参数，将训练得到的权重参数作为ICU警报识别模型的权重参数，得到训练优化后的ICU警报识别模型；否则转向下一步；S44: if

Terminate the algorithm iteration process, output the current weight parameter wⁿ , where wⁿ represents the weight parameter at the nth iteration, and use the weight parameter obtained from training as the weight parameter of the ICU alarm recognition model to obtain the optimized ICU alarm recognition model after training ; Otherwise go to the next step;S45：若

则令d_n＝g_n，否则令d_n＝-g_n；S45: if

Then let d_n =g_n , otherwise let d_n =-g_n ;S46：计算第n+1次迭代的权重参数：S46: Calculate the weight parameter of the n+1th iteration:wⁿ⁺¹＝wⁿ+λ_nd_nwⁿ⁺¹ ＝wⁿ +λ_n d_n

令n＝n+1，返回步骤S44；Make n=n+1, return to step S44;S5：当医院系统检测到警报时，采集与警报相关病人的病人体征指标数据，将所采集到的病人体征指标数据输入到训练优化后的ICU警报识别模型中，模型输出是否为虚假警报，若该警报不是虚假警报，则继续输出警报名称以及处理优先级，医护人员根据模型输出结果进行相应的处置措施。S5: When the hospital system detects an alarm, collect the patient sign data of the patient related to the alarm, and input the collected patient sign data into the optimized ICU alarm recognition model after training. Whether the model output is a false alarm, if If the alarm is not a false alarm, continue to output the alarm name and processing priority, and the medical staff will take corresponding disposal measures according to the output results of the model.2.如权利要求1所述的一种快速ICU虚假警报识别方法，其特征在于，所述S1步骤中采集病人体征数据，包括：2. a kind of fast ICU false alarm identification method as claimed in claim 1, is characterized in that, in the described S1 step, gathers patient sign data, comprises:在病人进入ICU后，采集病人体征数据，其中所述体征数据包括生命体征检查时序数据和病人入院48h内的实验室检查指标数据集合，所述生命体征检查时序数据为病人的心率以及呼吸频率的时序数据，所述入院48小时内的实验室检查指标包括病人尿液的酸碱度k₁、pH值k₂、尿比重k₃、尿蛋白k₄、尿糖k₅以及管型k₆，病人血液的白细胞数量k₇、白细胞中五类细胞的数量及比例k₈、红细胞数量k₉、红细胞压积k₁₀、血红蛋白浓度k₁₁、平均红细胞体积k₁₂、血小板的数量k₁₃以及血小板压积k₁₄，所采集的病人体征数据为：After the patient enters the ICU, the patient's sign data is collected, wherein the sign data includes the time series data of the vital sign examination and the laboratory test index data set within 48 hours of the patient's admission, and the time series data of the vital sign examination is the patient's heart rate and respiratory rate. Time-series data, the laboratory test indicators within 48 hours of admission include the patient’s urine pH k₁ , pH value k₂ , urine specific gravity k₃ , urine protein k₄ , urine sugar k₅ and cast k₆ , the patient’s blood The number of white blood cells k₇ , the number and proportion of the five types of cells in white blood cells k₈ , the number of red blood cells k₉ , the hematocrit k₁₀ , the concentration of hemoglobin k₁₁ , the mean red blood cell volume k₁₂ , the number of platelets k₁₃ and the hematocrit k_14. The collected patient sign data are:{x₁(n₁),x₂(n₂),K}{x₁ (n₁ ),x₂ (n₂ ),K}K＝{k_num|num∈[1,14]}K＝{k_num |num∈[1,14]}其中：in:x₁(n₁)为病人心率的时序数据，n₁＝0,1,…,N₁-1，N₁为时序数据长度；x₁ (n₁ ) is the time-series data of the patient's heart rate, n₁ =0,1,...,N₁ -1, N₁ is the length of the time-series data;x₂(n₂)为病人呼吸频率的时序数据，n₂＝0,1,…,N₂-1，N₂为时序数据长度；x₂ (n₂ ) is the time-series data of the patient's respiratory rate, n₂ =0,1,...,N₂ -1, N₂ is the length of the time-series data;K为病人的实验室检查指标数据集合，包括14种的实验室检查指标。K is the patient's laboratory test index data set, including 14 kinds of laboratory test indicators.3.如权利要求2所述的一种快速ICU虚假警报识别方法，其特征在于，所述S1步骤中利用傅里叶变换方法以及小波分解方法对采集的生命体征检查时序数据进行特征提取，将提取特征与检查指标数据集合共同构成病人体征指标数据，包括：3. a kind of fast ICU false alarm identification method as claimed in claim 2, it is characterized in that, utilize Fourier transform method and wavelet decomposition method to carry out feature extraction to the vital sign inspection time-series data of collection in the described S1 step, will The extracted features and the inspection index data set together constitute the patient's sign index data, including:对所采集的生命体征检查时序数据进行特征提取，所述特征提取流程为：Feature extraction is performed on the collected time-series data of vital signs examination, and the process of feature extraction is as follows:S11：利用傅里叶变换方法提取病人呼吸频率的特征f₂(p)，其中p为傅里叶变换处理的采样点数，将p设置为64，所述傅里叶变换方法的公式为：S11: Using the Fourier transform method to extract the feature f₂ (p) of the patient's respiratory frequency, where p is the number of sampling points processed by the Fourier transform, and p is set to 64. The formula of the Fourier transform method is:

其中：in:j为虚数单位，j²＝-1；j is the imaginary number unit, j² =-1;e为自然常数；e is a natural constant;S12：利用小波分解方法对病人心率的特征进行提取，所述小波分解结果为：S12: Using the wavelet decomposition method to extract the characteristics of the heart rate of the patient, the result of the wavelet decomposition is:

其中：in:q(a,x₁(n₁))为x₁(n₁)在尺度a上的小波系数，a表示小波分解过程中的最大尺度；q(a,x₁ (n₁ )) is the wavelet coefficient of x₁ (n₁ ) on scale a, and a represents the largest scale in the process of wavelet decomposition;将小波系数作为病人心率的特征f₁(a)，若q(a，x₁(n_R))＞0.4×max{q(a，x₁(n₁))}，n₁＝0,1,…,N₁-1，则将时序位置n_R在特征f₁(a)中标记为R峰。Take the wavelet coefficient as the characteristic f₁ (a) of the patient's heart rate, if q(a, x₁ (n_R ))>0.4×max{q(a, x₁ (n₁ ))}, n₁ =0,1 ,...,N₁ -1, then the timing position n_R is marked as an R peak in the feature f₁ (a).4.如权利要求2-3所述的一种快速ICU虚假警报识别方法，其特征在于，所述S2步骤采集大量病人的病人体征指标数据以及对应的警报级别构成训练集A1，包括：4. A kind of fast ICU false alarm identification method as claimed in claim 2-3, is characterized in that, described S2 step gathers the patient's sign index data of a large number of patients and corresponding alarm level constitutes training set A1, comprises:按照步骤S1方法采集大量病人的病人体征指标数据以及对应的警报级别构成训练集A1，所述训练集A1的表示形式为：According to the method of step S1, the patient sign index data of a large number of patients and the corresponding alarm levels are collected to form a training set A1, and the expression form of the training set A1 is:A1＝{data_u＝(f_1,u(a),f_2,u(p),judge_u,name_u,level_u)|u∈[1,U]}A1＝{data_u ＝(f_1,u (a),f_2,u (p),judge_u ,name_u ,level_u )|u∈[1,U]}其中：in:data_u为所采集的第u名病人的病人体征指标数据以及对应的警报级别，U表示训练集A1中病人的总数；data_u is the collected patient sign index data of the uth patient and the corresponding alarm level, and U represents the total number of patients in the training set A1;f_1,u(a)表示所采集的第u名病人的心率特征，f_2,u(p)为所采集的第u名病人的呼吸频率特征；f_1,u (a) represents the collected heart rate feature of the uth patient, and f_2,u (p) is the collected respiratory rate feature of the uth patient;judge_u＝{0,1}，judge_u＝0表示该病人所对应的警报为虚假警报，并将name_u以及level_u均设置为null；judge_u＝1表示该病人所对应的警报不为虚假警报；judge_u = {0,1}, judge_u = 0 indicates that the alarm corresponding to the patient is a false alarm, and set name_u and level_u to null; judge_u = 1 indicates that the alarm corresponding to the patient is not false alarm;name_u表示该病人所对应警报的名称，name_u∈{1,2,3,…,15}，所述警报名称包括SPO2、导联脱落、血压、心率、室早、心脏停搏、PR过高、Resp窒息、室颤、RONT、室早二联律、ST段、室早三联律、房早以及室速，依次对应name_u＝1至name_u＝15；name_u indicates the name of the alarm corresponding to the patient, name_u ∈ {1,2,3,...,15}, the alarm name includes SPO2, lead off, blood pressure, heart rate, ventricular premature, cardiac arrest, PR excessive High, Resp apnea, ventricular fibrillation, RONT, premature ventricular bigeminy, ST segment, ventricular premature trigeminy, atrial premature and ventricular tachycardia, corresponding to name_u = 1 to name_u = 15 in turn;level_u表示该病人所对应的警报级别，level_u＝{1,2}，level_u＝1表示高优先级的警报，level_u＝2表示中优先级的警报。level_u indicates the alarm level corresponding to the patient, level_u ={1,2}, level_u =1 indicates a high priority alarm, and level_u =2 indicates a medium priority alarm.5.如权利要求4所述的一种快速ICU虚假警报识别方法，其特征在于，所述S2步骤中利用生成对抗网络构建病人指标数据样本生成模型，利用病人指标数据样本生成模型生成大量病人体征指标数据，将所生成的数据同训练集A1构成训练集A，包括：5. A kind of fast ICU false alarm identification method as claimed in claim 4, is characterized in that, utilizes generation confrontation network to construct patient index data sample generation model in the described S2 step, utilizes patient index data sample generation model to generate a large amount of patient signs Index data, the generated data and the training set A1 form the training set A, including:利用生成对抗网络构建病人指标数据样本生成模型，其中所述病人指标数据样本生成模型包括编码器G_e(·)和解码器G_d(·)，所述编码器的输入值为训练集A1中的病人体征指标数据，通过利用编码器G_e(·)对输入的病人体征指标数据进行编码，利用解码器G_d(·)对编码结果进行解码，得到重构后的病人体征指标数据，将解码结果作为模型的样本生成结果；Utilize the generative confrontation network to construct the patient index data sample generation model, wherein the patient index data sample generation model includes an encoder G_e (·) and a decoder G_d (·), and the input value of the encoder is in the training set A1 The patient's sign index data, by using the encoder G_e (·) to encode the input patient sign index data, and using the decoder G_d (·) to decode the encoding result, the reconstructed patient sign index data is obtained, and the The decoding result is used as the sample generation result of the model;计算训练集A1中judge＝0的病人所占的比例，若该比例小于0.6，则选取judge＝0的病人所对应的病人体征指标数据作为病人指标数据样本生成模型中编码器的输入，编码器对输入结果进行编码后，利用解码器对编码结果进行解码，得到重构后的病人体征指标数据，将重构得到的病人体征指标数据的judge参数设置为0，所述样本生成后judge＝0的病人所占的比例达到0.6，得到训练集A2；Calculate the proportion of patients with judge=0 in the training set A1, if the proportion is less than 0.6, select the patient sign index data corresponding to the patient with judge=0 as the input of the encoder in the patient index data sample generation model, the encoder After encoding the input result, use the decoder to decode the encoded result to obtain the reconstructed patient sign index data, set the judge parameter of the reconstructed patient sign index data to 0, and judge=0 after the sample is generated The proportion of the patients reached 0.6, and the training set A2 was obtained;计算训练集A2中不同警报名称的病人数，选取病人数最少的5个警报名称，将所选取的警报名称下的病人体征指标数据以及对应的警报级别分别构成五个数据集A2_min1,A2_min2，A2_min3，A2_min4以及A2_min5，其中每个数据集中病人体征指标数据所对应的警报名称相同，选取任意数据集的所有病人体征指标数据作为病人指标数据样本生成模型中编码器的输入，编码器对输入结果进行编码后，利用解码器对编码结果进行解码，得到重构后的病人体征指标数据，向该重构后的病人体征指标数据添加该数据集所对应的警报名称，以及警报名称所对应的警报级别，并将judge参数设置为1，得到生成样本，所述生成样本的生成数量为(num_A2_max-num_A2_minj′)×β，其中num_A2_max为训练集A2中出现频率最高的警报名称的病人人数，j′∈{1,2,3,4,5}，num_A2_minj′为数据集A2_minj′内的病人人数，β为样本生成参数，将其设置为0.3，将所生成的数据同训练集A2构成训练集A；Calculate the number of patients with different alarm names in the training set A2, select the five alarm names with the least number of patients, and form five data sets A2_min1 and A2_min2 respectively with the patient sign index data under the selected alarm names and the corresponding alarm levels , A2_min3 , A2_min4 and A2_min5 , the alarm names corresponding to the patient sign data in each data set are the same, select all the patient sign data in any data set as the input of the encoder in the patient index data sample generation model, encode After the input result is encoded by the decoder, the encoded result is decoded by the decoder to obtain the reconstructed patient sign data, and the alarm name corresponding to the data set and the alarm name are added to the reconstructed patient sign data The corresponding alarm level, and set the judge parameter to 1 to obtain the generated samples. The generated number of the generated samples is (num_A2_max -num_A2_minj′ )×β, where num_A2_max is the alarm with the highest frequency in the training set A2 The number of patients with the name, j′∈{1,2,3,4,5}, num_A2_minj′ is the number of patients in the data set A2_minj′ , β is the sample generation parameter, set it to 0.3, and the generated The data and the training set A2 form the training set A;构建病人指标数据样本生成模型中编码器G_e(·)和解码器G_d(·)的训练目标函数：Construct the training objective function of encoder G_e ( ) and decoder G_d ( ) in the patient indicator data sample generation model:

其中：in:f为病人指标数据样本生成模型中编码器G_e(·)的输入值，包括病人的心率特征以及呼吸频率特征，p(f)为f的分布；f is the input value of encoder G_e (·) in the patient index data sample generation model, including the patient's heart rate characteristics and respiratory frequency characteristics, p(f) is the distribution of f;E(·)表示期望值计算；E( ) means expected value calculation;

为编码器G_e(·)的参数，

为解码器G_d(·)的参数；

is the parameter of the encoder G_e (·),

is the parameter of decoder G_d (·);||·||₂为L2范数；||·||₂ is the L2 norm;将多组病人的心率特征以及呼吸频率特征输入到训练目标函数中，使得训练目标函数达到最小的模型参数即为训练优化得到的模型参数，利用训练优化后的病人指标数据样本生成模型进行数据生成。Input the heart rate characteristics and respiratory frequency characteristics of multiple groups of patients into the training objective function, so that the model parameters that minimize the training objective function are the model parameters obtained by training optimization, and use the optimized patient index data sample generation model for data generation .6.如权利要求1所述的一种快速ICU虚假警报识别方法，其特征在于，所述S3步骤中基于概率图理论构建ICU警报识别模型，包括：6. a kind of fast ICU false alarm identification method as claimed in claim 1, is characterized in that, in the described S3 step, builds ICU alarm identification model based on probability map theory, comprising:基于概率图理论构建ICU警报识别模型，其中所述模型的输入为病人体征指标数据，输出是否虚假警报的识别结果，若识别为虚假警报则输出警报名称以及警报级别为null，否则输出对应的警报名称和警报级别；Construct an ICU alarm recognition model based on the probability graph theory, wherein the input of the model is the patient's sign index data, and output the identification result of whether it is a false alarm, if it is identified as a false alarm, output the alarm name and the alarm level is null, otherwise output the corresponding alarm name and alert level;所述ICU警报识别模型为图结构G＝(E,V)，其中E表示节点集合，节点包括警报名称节点name、judge＝0的节点、judge＝1的节点、多种不同的病人心率特征节点以及呼吸频率特征节点，V表示边集合，所述边为有向边，有向边f_q→name_b表示在病人特征为f_q的情况下，发生警报name_b的概率P(name_b|f_q)；The ICU alarm recognition model is a graph structure G=(E, V), wherein E represents a node set, and the nodes include the alarm name node name, the node of judge=0, the node of judge=1, and a variety of different patient heart rate characteristic nodes And the respiratory frequency feature node,_V represents the edge set, the edge is a directed edge, and the directed edge f_q →name_b represents the probability P(name_b| f_q );所述ICU警报识别模型的输入为病人体征指标数据[f_1,*(a),f_2,*(p)]，利用余弦相似度算法分别计算f_1,*(a)以及f_2,*(p)与所构建概率图中病人心率特征节点以及呼吸频率特征节点的相似度，将相似度最高的病人心率特征节点以及呼吸频率特征节点[f_1,sim(a),f_2,sim(p)]作为[f_1,*(a),f_2,*(p)]在概率图中的表示结果，分别计算该病人体征指标数据为虚假警报的概率以及非虚假警报的概率：The input of the ICU alarm recognition model is patient sign index data [f_1,* (a), f_2,* (p)], using the cosine similarity algorithm to calculate f_1,* (a) and f_2,* (p) The similarity with the patient's heart rate feature node and respiratory rate feature node in the constructed probability map, the patient's heart rate feature node and respiratory rate feature node with the highest similarity [f_1,sim (a),f_2,sim ( p)] as the expression result of [f_1,* (a),f_2,* (p)] in the probability map, respectively calculate the probability of the patient's sign index data being a false alarm and the probability of a non-false alarm:P(judge＝0|f_1,sim(a),f_2,sim(p))＝w₁P(judge＝0|f_1,sim(a))w₂P(judge＝0|f_2,sim(p))P(judge=0|f_1,sim (a),f_2,sim (p))=w₁ P(judge=0|f_1,sim (a))w₂ P(judge=0|f_{2, sim} (p))P(judge＝1|f_1,sim(a),f_2,sim(p))＝w₁P(judge＝1|f_1,sim(a))w₂P(judge＝1|f_2,sim(p))P(judge=1|f_1,sim (a),f_2,sim (p))=w₁ P(judge=1|f_1,sim (a))w₂ P(judge=1|f_{2, sim} (p))其中：in:w₁为心率特征在概率图的权重参数，w₂为呼吸频率特征在概率图中的权重参数；w₁ is the weight parameter of the heart rate feature in the probability map, and w₂ is the weight parameter of the respiratory frequency feature in the probability map;P(judge＝0|f_1,sim(a),f_2,sim(p))为该病人体征指标数据为虚假警报的概率；P(judge＝0|f_{1, sim} (a), f_{2, sim} (p)) is the probability that the patient's sign index data is a false alarm;P(judge＝1|f_1,sim(a),f_2,sim(p))为该病人体征指标数据为非虚假警报的概率；P(judge=1|f_{1, sim} (a), f_{2, sim} (p)) is the probability that the patient's sign index data is not a false alarm;若P(judge＝0|f_1,sim(a),f_2,sim(p))≥P(judge＝1|f_1,sim(a),f_2,sim(p))则说明该病人所对应的警报为虚假警报，则直接输出警报名称以及警报级别为null；If P(judge=0|f_{1, sim} (a), f_{2, sim} (p))≥P(judge=1|f_{1, sim} (a), f_{2, sim} (p)), then the patient If the corresponding alarm is a false alarm, then directly output the alarm name and alarm level as null;否则计算该病人体征指标数据会导致不同警报发生的概率：Otherwise, calculating the patient's sign data will result in the probability of different alarm occurrences:P(name＝m|f_1,sim(a),f_2,sim(p))＝P(name＝m|f_1,sim(a))P(name＝m|f_2,sim(p))P(name=m|f_1,sim (a),f_2,sim (p))=P(name=m|f_1,sim (a))P(name=m|f_2,sim (p) )其中：in:m∈{1,2,3,…,15}，对应不同的警报名称；m∈{1,2,3,…,15}, corresponding to different alarm names;选取发生概率最大的警报名称以及警报对应的警报级别作为模型的输出。Select the alarm name with the highest probability of occurrence and the alarm level corresponding to the alarm as the output of the model.7.如权利要求1所述的一种快速ICU虚假警报识别方法，其特征在于，所述S5步骤中当医院系统检测到警报时，采集与警报相关病人的病人体征指标数据，将所采集到的病人体征指标数据输入到训练优化后的ICU警报识别模型中，模型输出是否为虚假警报，若该警报不是虚假警报，则继续输出警报名称以及处理优先级，包括：7. A kind of fast ICU false alarm identification method as claimed in claim 1, is characterized in that, in described S5 step, when hospital system detects alarm, gathers the patient's sign index data of patient relevant to alarm, collects Input the data of the patient's signs and indicators into the optimized ICU alarm recognition model after training. Whether the model output is a false alarm, if the alarm is not a false alarm, continue to output the alarm name and processing priority, including:当医院系统检测到警报时，按照步骤S1方法采集与警报相关病人的病人体征指标数据，将所采集到的病人体征指标数据输入到训练优化后的ICU警报识别模型中，模型输出是否为虚假警报，若该警报不是虚假警报，则继续输出警报名称以及警报名称所对应的处理优先级，其中处理优先级由高到中，医护人员根据处理优先级以及模型输出的警报名称进行相应的处置措施。When the hospital system detects an alarm, collect the patient sign data of the patient related to the alarm according to the method of step S1, and input the collected patient sign data into the optimized ICU alarm recognition model after training, whether the model output is a false alarm , if the alarm is not a false alarm, continue to output the alarm name and the processing priority corresponding to the alarm name, where the processing priority is from high to medium, and the medical staff will take corresponding measures according to the processing priority and the alarm name output by the model.8.一种快速ICU虚假警报识别装置，其特征在于，所述装置包括：8. A fast ICU false alarm identification device, is characterized in that, said device comprises:特征提取模块，用于采集病人体征数据，利用傅里叶变换方法以及小波分解方法对采集的生命体征检查时序数据进行特征提取，将提取特征与检查指标数据集合共同构成病人体征指标数据；The feature extraction module is used to collect the patient's vital signs data, and uses the Fourier transform method and the wavelet decomposition method to perform feature extraction on the collected vital sign examination time-series data, and the extracted features and the examination index data set together form the patient's sign index data;训练集获取装置，用于采集大量病人的病人体征指标数据以及对应的警报级别构成训练集A1，并利用生成对抗网络构建病人指标数据样本生成模型，利用病人指标数据样本生成模型生成大量病人体征指标数据，将所生成的数据同训练集A1构成训练集A；The training set acquisition device is used to collect patient sign index data of a large number of patients and corresponding alarm levels to form a training set A1, and use the generative confrontation network to construct a patient index data sample generation model, and use the patient index data sample generation model to generate a large number of patient sign indexes Data, the generated data and the training set A1 form the training set A;ICU警报识别装置，用于基于概率图理论构建ICU警报识别模型，基于训练集A，利用分裂梯度算法对所构建的ICU警报识别模型进行快速优化，将所采集到的病人体征指标数据输入到训练优化后的ICU警报识别模型中，模型输出是否为虚假警报，若该警报不是虚假警报，则继续输出警报名称以及处理优先级，以实现一种如权利要求1-7任一一项所述的快速ICU虚假警报识别方法。The ICU alarm recognition device is used to build an ICU alarm recognition model based on the probability graph theory. Based on the training set A, the split gradient algorithm is used to quickly optimize the built ICU alarm recognition model, and the collected patient sign data is input into the training In the optimized ICU alarm identification model, whether the model output is a false alarm, if the alarm is not a false alarm, then continue to output the alarm name and processing priority, to achieve a kind of as described in any one of claims 1-7 Rapid ICU false alarm identification method.

Images