WO2023097785A1 - 一种基于模糊熵特征提取的人机通气异步检测模型及装置 - Google Patents

一种基于模糊熵特征提取的人机通气异步检测模型及装置 Download PDF

Info

Publication number
WO2023097785A1
WO2023097785A1 PCT/CN2021/138106 CN2021138106W WO2023097785A1 WO 2023097785 A1 WO2023097785 A1 WO 2023097785A1 CN 2021138106 W CN2021138106 W CN 2021138106W WO 2023097785 A1 WO2023097785 A1 WO 2023097785A1
Authority
WO
WIPO (PCT)
Prior art keywords
waveform
fuzzy entropy
respiratory
matrix
ventilation
Prior art date
Application number
PCT/CN2021/138106
Other languages
English (en)
French (fr)
Inventor
马良
颜延
熊富海
仲为
王磊
李慧慧
Original Assignee
中国科学院深圳先进技术研究院
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 中国科学院深圳先进技术研究院 filed Critical 中国科学院深圳先进技术研究院
Publication of WO2023097785A1 publication Critical patent/WO2023097785A1/zh

Links

Images

Classifications

    • AHUMAN NECESSITIES
    • A61MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61BDIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00Measuring for diagnostic purposes; Identification of persons
    • AHUMAN NECESSITIES
    • A61MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61BDIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00Measuring for diagnostic purposes; Identification of persons
    • A61B5/08Detecting, measuring or recording devices for evaluating the respiratory organs
    • AHUMAN NECESSITIES
    • A61MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61MDEVICES FOR INTRODUCING MEDIA INTO, OR ONTO, THE BODY; DEVICES FOR TRANSDUCING BODY MEDIA OR FOR TAKING MEDIA FROM THE BODY; DEVICES FOR PRODUCING OR ENDING SLEEP OR STUPOR
    • A61M16/00Devices for influencing the respiratory system of patients by gas treatment, e.g. mouth-to-mouth respiration; Tracheal tubes

Definitions

  • the invention relates to the field of medical detection, in particular to a human-computer ventilation asynchronous detection model and device based on fuzzy entropy feature extraction.
  • the human-machine dyssynchrony In the process of mechanical ventilation in the intensive care unit (ICU), the phenomenon of uncoordinated interaction between the ventilator and the patient is called human-machine dyssynchrony.
  • the human-machine asynchronous types found at present roughly fall into the following four categories and eight types: (1) trigger types: invalid trigger, repeated trigger, automatic trigger and reverse trigger.
  • Expiratory phase endogenous positive end-expiratory pressure (PEEPi).
  • the existing research on human-computer asynchronous waveform detection mainly focuses on the frequent invalid inspiratory effort and double triggering during mechanical ventilation, such as the identification method based on wavelet features proposed for invalid inspiratory effort.
  • the machine learning method gradually matures, its application tentacles have also extended to this field, such as the human-machine asynchronous detection method for mechanical ventilation based on the recurrent neural network.
  • the process of classifying human-computer dyssynchrony based on machine learning or deep learning can be summarized as follows: (1) Obtain the respiratory mechanics waveform data derived from the ventilator; (2) Label the human-computer asynchronous waveform through professional physicians; (3) Data preprocessing and dividing the data set into training set, verification set and test set; (4) Input the preprocessed data into the defined model for model training; (5) Save the trained model for application .
  • Embodiments of the present invention provide a human-computer ventilation asynchronous detection model and device based on fuzzy entropy feature extraction, so as to at least solve the phenomenon that the existing technology cannot be used to detect multiple human-computer ventilation asynchronous extractions at the same time.
  • a human-computer ventilation asynchronous detection model based on fuzzy entropy feature extraction including the following steps:
  • S1 Collect the respiratory waveform of the current respiratory cycle, select the appropriate channel data in the respiratory waveform, and the respiratory waveform includes tidal volume waveform, airway pressure time waveform and flow velocity time waveform;
  • S6 Extract the fuzzy entropy from the airway pressure time waveform and the flow velocity time waveform to obtain a feature matrix, and obtain a fuzzy entropy feature vector by flattening the feature matrix, which is a training sample of a respiratory cycle;
  • S7 Repeat the process of S1-S6 for all respiratory cycle waveforms to obtain a number of training samples of N, expressed as a matrix of size (N, 6), and correspondingly obtain the labels of N samples, and store them in (N, 1)
  • N, 6 a matrix of size
  • N, 1 In the column vector of size, double triggering, invalid inspiratory effort and normal ventilation are marked as [1, 2, 3] respectively, and a human-computer ventilation asynchronous detection model is generated.
  • the respiratory waveform of each respiratory cycle is marked with the asynchronous type of human-machine ventilation, including double triggering, invalid inspiratory effort, and normal ventilation.
  • tidal volume waveform, the airway pressure time waveform and the flow velocity time waveform are sequentially compared to form two window subsequences in the respiratory cycle, specifically:
  • the window size is 40 and the stride is 40.
  • the embedding matrix is constructed sequentially for the two window subsequences as follows:
  • the matrix calculation formula is:
  • 1 ⁇ i ⁇ N-m+1, m is the dimension embedded in the calculation of entropy features, and N is the length of the time series under the sub-window.
  • the infinite norm calculation formula is:
  • the fuzzy entropy is calculated according to the fuzzy entropy formula, and the fuzzy entropy formula is:
  • FuzzyEn(m,n,r) ln( ⁇ m (n,r))-ln( ⁇ m +1 (n,r)).
  • the fuzzy entropy extraction is performed on the airway pressure time waveform and the flow velocity time waveform to obtain the feature matrix.
  • the fuzzy entropy feature vector is obtained as follows:
  • the respiratory waveform of each channel of each respiratory cycle is extracted to obtain two fuzzy entropy features, and the shape is a (2, 1) feature matrix;
  • a human-computer ventilation asynchronous detection device based on fuzzy entropy feature extraction comprising:
  • the waveform selection module is used to collect the respiratory waveform of the current respiratory cycle, and select the appropriate channel data in the respiratory waveform.
  • the respiratory waveform includes tidal volume waveform, airway pressure time waveform and flow velocity time waveform;
  • Waveform labeling module used for labeling the selected respiratory waveform
  • the waveform windowing module is used to sequentially perform the tidal volume waveform, the airway pressure time waveform and the flow velocity time waveform to form two window subsequences in the respiratory cycle;
  • a matrix construction module is used to construct an embedding matrix for two window subsequences in turn;
  • a fuzzy entropy calculation module used to calculate fuzzy entropy based on two embedded matrices constructed
  • the sample training module is used to extract the fuzzy entropy of the airway pressure time waveform and the flow velocity time waveform to obtain the feature matrix. By flattening the feature matrix, the fuzzy entropy feature vector is obtained.
  • the fuzzy entropy feature vector is the training of a breathing cycle sample
  • the model training module is used to repeatedly execute the process from the waveform selection module 100 to the sample training module 600 for all respiratory cycle waveforms to obtain a number of training samples of N, expressed as a matrix of (N, 6) size, and correspondingly obtain N
  • the label of the sample is stored in a column vector of size (N, 1), where double triggering, invalid inspiratory effort and normal ventilation are marked as [1, 2, 3] respectively.
  • the waveform labeling module includes:
  • the type marking unit is used to mark the type of man-machine ventilation asynchronously on the respiratory waveform of each breathing cycle, and the types include double triggering, invalid inspiratory effort and normal ventilation.
  • the device also includes:
  • the window setting module is used to set the window size to 40 and the step size to 40.
  • a computer-readable storage medium stores one or more programs, and the one or more programs can be executed by one or more processors, so as to realize the fuzzy entropy-based feature extraction of any one of the above Steps in the machine-ventilation asynchronous detection model.
  • the human-computer ventilation asynchronous detection model and device based on fuzzy entropy feature extraction in the embodiment of the present invention through the collected respiratory waveforms with various human-computer phenomena, the feature extraction method based on fuzzy entropy can detect multi-channel respiratory waveforms Perform fuzzy entropy feature extraction, calculate fuzzy entropy features with certain parameters, and then flatten the three-dimensional fuzzy entropy features to one-dimensional feature samples to form a respiratory cycle, and repeatedly perform feature extraction on all respiratory cycle waveforms Finally, a model that can classify or detect a variety of human-computer ventilation asynchronous phenomena is finally obtained.
  • Fig. 1 is the flowchart of the man-machine ventilation asynchronous detection model based on fuzzy entropy feature extraction of the present invention
  • Fig. 2 is the figure of confusion moments of each model of the present invention on the test set
  • Fig. 3 is a schematic diagram of the human-computer ventilation asynchronous detection device based on fuzzy entropy feature extraction according to the present invention.
  • a human-computer ventilation asynchronous detection model based on fuzzy entropy feature extraction including the following steps:
  • S1 Collect the respiratory waveform of the current respiratory cycle, select the appropriate channel data in the respiratory waveform, and the respiratory waveform includes tidal volume waveform, airway pressure time waveform and flow velocity time waveform;
  • S6 Extract the fuzzy entropy from the airway pressure time waveform and the flow velocity time waveform to obtain a feature matrix, and obtain a fuzzy entropy feature vector by flattening the feature matrix, which is a training sample of a respiratory cycle;
  • S7 Repeat the process of S1-S6 for all respiratory cycle waveforms to obtain a number of training samples of N, expressed as a matrix of size (N, 6), and correspondingly obtain the labels of N samples, and store them in (N, 1)
  • N, 6 a matrix of size
  • N, 1 In the column vector of size, double triggering, invalid inspiratory effort and normal ventilation are marked as [1, 2, 3] respectively, and a human-computer ventilation asynchronous detection model is generated.
  • the invention proposes a new detection and classification method aiming at the human-machine asynchronous phenomenon common in the mechanical ventilation process of the intensive care unit, and the method can be used to classify various human-machine asynchronous phenomena at the same time.
  • Its basic content is to calculate the fuzzy entropy feature of the original respiratory waveform marked by the physician, including the flow velocity waveform, airway pressure time waveform, and tidal volume time waveform, respectively with certain parameters, and then flatten the three-dimensional fuzzy entropy features
  • the feature samples transformed into one dimension to form a respiratory cycle are used for the learning and training of the classification model later.
  • the feature extraction method based on fuzzy entropy is used to perform fuzzy entropy feature extraction on multi-channel respiratory waveforms, and then put them into the constructed machine learning model for training.
  • Step 1 Select the appropriate channel data; considering factors such as the correlation between waveforms and whether the effectiveness of feature extraction and classification under different ventilation modes can be guaranteed, select three-dimensional respiratory waveforms including tidal waveforms for the original data of respiratory waveforms Volume waveform, airway pressure time waveform and flow velocity time waveform, and entropy features were extracted from these three waveforms respectively.
  • Step 2 The above respiratory waveform needs to be marked by the physician.
  • the marking process is to mark the waveform of each respiratory cycle.
  • the specific mark is the type of man-machine asynchrony, including DT (double trigger), IEE (ineffective inspiration effort) and Normal (normal ventilation).
  • Step 3 Since the acquisition frequency is 30Hz, there are about 80-90 time points in one respiratory cycle; for the tidal volume waveform, airway pressure time waveform and flow velocity time waveform in turn, the window size is 40, and the step size is also is 40, the tidal volume waveform data is firstly windowed, so there are two window subsequences in one respiratory cycle.
  • Step 4 In constructing the embedding matrix, construct the embedding matrix through the matrix calculation formula, and construct the embedding matrix for the two windows in step 3 in turn.
  • the mathematical expression of the matrix calculation formula is as follows:
  • 1 ⁇ i ⁇ N-m+1, m is the dimension embedded in the calculation of entropy features, and N is the length of the time series under the sub-window.
  • Step 5 Calculate the fuzzy entropy based on the constructed two embedding matrices; wherein, the following steps are included:
  • Step 1 Calculate the infinite norm between two embedded dimension vectors; use the infinite norm calculation formula to calculate the infinite norm between the embedded dimension vectors for the two embedded matrices constructed in step 4, the infinite norm calculation formula as follows:
  • the second step calculate the similarity. Define it as the following expression:
  • Step 3 Calculate the fuzzy function ⁇ m (n,r):
  • ⁇ m (n, r) is a factor in the fuzzy entropy formula.
  • Step 5 Calculate the fuzzy entropy FuzzyEn(m,n,r) according to the fuzzy entropy, the fuzzy entropy formula is:
  • FuzzyEn(m,n,r) ln( ⁇ m (n,r))-ln( ⁇ m+1 (n,r))
  • Step 6 According to the process from step 3 to step 5, two fuzzy entropy features can be obtained for the waveform of each channel in each respiratory cycle, the shape is (2, 1) matrix, and the airway pressure time waveform and flow velocity time are repeated. Waveform undergoes the same fuzzy entropy extraction process, and finally the feature matrix of (2, 3) can be obtained. After flattening it, a fuzzy entropy feature vector of the shape (1, 6) can be obtained, which is the training of a breathing cycle sample.
  • Step 7 Repeat the above process, input all respiratory cycle waveforms into the feature extraction algorithm, the number of samples that can be obtained finally is N, expressed as a (N, 6) size matrix, and correspondingly get the labels of N samples, store into a column vector of size (N, 1).
  • double trigger (DT), invalid inspiratory effort (IEE) and normal ventilation (Normal) are respectively marked as [1,2,3] to generate an asynchronous detection model of human-machine ventilation.
  • LR logistic regression
  • SVM support vector machine
  • DT decision Tree
  • MLP multilayer perceptron
  • the scheme of the present invention has the following characteristics:
  • This program selectively selects the tidal volume time waveform, airway pressure time waveform and flow velocity time waveform from the collected respiratory waveform data.
  • the selection method of this scheme comprehensively considers the correlation between waveforms and whether the effectiveness of feature extraction and classification under different ventilation modes can be guaranteed.
  • the selection of the present invention is not limited to only selecting the waveforms of these three dimensions as the elements of feature extraction, and the waveforms of other three or more dimensions can also be selected as the elements of feature extraction according to the above method.
  • This program uses the fuzzy entropy algorithm to obtain the characteristics of the respiratory waveform from step 3 to step 6, as follows:
  • the method of windowing the signal of each respiratory cycle into different sub-window sequences which includes two parameters of window size and step size.
  • the parameters can be properly selected according to the data sampling frequency and data length. Values limited to this scenario.
  • This program is to calculate the fuzzy entropy separately for one-dimensional time series, and the sequential extraction process used in the implementation process, that is to say, the fuzzy time waveform of tidal volume time waveform, airway pressure time waveform and flow velocity time waveform are sequentially extracted Entropy, but does not affect the algorithm can simultaneously calculate fuzzy entropy for three channel time series.
  • the calculation of the existing technical solution is relatively complicated.
  • the proposed solution of the present invention directly calculates the fuzzy entropy feature for the original signal, and then uses it as the input sample of the machine learning algorithm.
  • the accuracy rate and F1-score of the prior art are lower than that of the present invention 96% of the program.
  • the present scheme is for multi-classification tasks while the former one only classifies ineffective inspiratory effort phenomena.
  • the method of using the fuzzy entropy algorithm to extract respiratory waveform features is relatively simple and has strong model adaptability. As shown in Table 1, relatively high classification accuracy can be easily achieved for most simple and common machine learning models .
  • the method used in the scheme of the present invention directly learns and classifies a variety of human-machine asynchronous phenomena in mechanical ventilation. Compared with the current two-category scheme, it is obviously more clinically practical.
  • the scheme of the present invention has adopted multiple machine learning algorithm construction models, and carried out experiment, specifically as follows:
  • the data set used in the experiment is collected from the preset patient data in the simulated lung.
  • the case selection is ARDS patient, the patient's spontaneous breathing rate is 21, the ventilation mode is CPAP/PSV, and the data sampling rate is 50Hz.
  • the annotated data set contains 1530 cycles of the double trigger (DT) type, 1447 cycles of the invalid inspiratory effort (IEE), and 1360 cycles of the normal waveform (Normal). According to the commonly used division ratio of training set and test set, this scheme selects the test set to account for 20% of the total number of samples.
  • DT double trigger
  • IEE invalid inspiratory effort
  • Normal normal waveform
  • a human-computer ventilation asynchronous detection device based on fuzzy entropy feature extraction including:
  • the waveform selection module 100 is used to collect the respiratory waveform of the current respiratory cycle, select the appropriate channel data in the respiratory waveform, and the respiratory waveform includes tidal volume waveform, airway pressure time waveform and flow velocity time waveform;
  • Waveform labeling module 200 for labeling the selected respiratory waveform
  • the waveform windowing module 300 is used to sequentially perform the tidal volume waveform, the airway pressure time waveform and the flow velocity time waveform to form two window subsequences in the respiratory cycle;
  • a matrix construction module 400 configured to construct an embedding matrix for two window subsequences in turn;
  • the fuzzy entropy calculation module 500 is used to calculate the fuzzy entropy based on two embedded matrices constructed
  • the sample training module 600 is used to perform fuzzy entropy extraction on the airway pressure time waveform and flow velocity time waveform to obtain a feature matrix, and obtain a fuzzy entropy feature vector by flattening the feature matrix, which is a breathing cycle Training samples;
  • the model training module 700 is used to repeat the process of S1-S6 for all respiratory cycle waveforms to obtain a number of training samples of N, expressed as a matrix of (N, 6) size, and correspondingly obtain labels of N samples, and store them in In a column vector of size (N, 1), double triggers, invalid inspiratory effort, and normal ventilation are respectively marked as [1, 2, 3] to generate a human-machine ventilation asynchronous detection model.
  • the human-computer ventilation asynchronous detection device based on fuzzy entropy feature extraction in the embodiment of the present invention fuzzy multi-channel respiratory waveforms through the collected respiratory waveforms with various human-machine different phenomena, based on the feature extraction method of fuzzy entropy Entropy feature extraction is to perform fuzzy entropy feature calculation with certain parameters, and then flatten the three-dimensional fuzzy entropy features to one-dimensional to form a feature sample of a respiratory cycle, and repeatedly perform feature extraction and calculation on all respiratory cycle waveforms, The result is a model that can classify multiple human-machine out-of-sync phenomena simultaneously.
  • the waveform labeling module includes:
  • the type marking unit is used to mark the type of man-machine ventilation asynchronously on the respiratory waveform of each breathing cycle, and the types include double triggering, invalid inspiratory effort and normal ventilation.
  • the device also includes:
  • the window setting module is used to set the window size to 40 and the step size to 40.
  • this embodiment Based on the asynchronous detection model of man-machine ventilation based on fuzzy entropy feature extraction, this embodiment provides a computer-readable storage medium, which stores one or more programs, and one or more programs can be used by one or more processors to implement the steps in the human-computer ventilation asynchronous detection model based on fuzzy entropy feature extraction as in the above embodiment.

Landscapes

  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Animal Behavior & Ethology (AREA)
  • Engineering & Computer Science (AREA)
  • Biomedical Technology (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Veterinary Medicine (AREA)
  • Public Health (AREA)
  • Biophysics (AREA)
  • Pulmonology (AREA)
  • Physics & Mathematics (AREA)
  • Pathology (AREA)
  • Medical Informatics (AREA)
  • Molecular Biology (AREA)
  • Surgery (AREA)
  • Emergency Medicine (AREA)
  • Hematology (AREA)
  • Physiology (AREA)
  • Anesthesiology (AREA)
  • Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)

Abstract

一种基于模糊熵特征提取的人机通气异步检测模型及装置,涉及医疗检测领域,通过对采集到的具有多种人机不同现象的呼吸波形,基于模糊熵的特征提取方法,对多通道呼吸波形进行模糊熵特征提取,分别以一定的参数进行模糊熵特征计算,然后将三个维度的模糊熵特征扁平化到一维构成一个呼吸周期的特征样本,重复的将所有的呼吸周期波形进行特征提取计算,最后得到一个可同时分类多种人机不同步的现象的模型。

Description

一种基于模糊熵特征提取的人机通气异步检测模型及装置 技术领域
本发明涉及医疗检测领域,具体而言,涉及一种基于模糊熵特征提取的人机通气异步检测模型及装置。
背景技术
在重症监护室(ICU)机械通气过程中,呼吸机与患者之间的交互出现不协调的现象称为人机不同步。经过医学界和学术界长时间的研究报道,目前发现的人机不同步类型大致有以下四类八种:(1)触发类型:无效触发、重复触发、自动触发和反向触发。(2)吸气阶段:流速不匹配。(3)切换阶段:提前切换和延迟切换。(4)呼气阶段:内源性呼气末正压(PEEPi)。
当床旁医生及时发现以上任一现象发生时,可以根据自己的经验知识进行判断分类,而后更改呼吸机呼吸力学参数的设置。但是要求医生24小时注意留心患者呼吸机的波形会占用极大的人力资源且也是不现实的。因此,提高呼吸机自动识别与检测人机不同步的能力便是呼吸机功能的一项重要内容,随之而来的就是对机械通气过程中人机不同步波形检测的研究日益增多。
经过调查,现有的关于人机不同步波形检测的研究主要集中在机械通气过程中多发的无效吸气努力和双重触发,如针对无效吸气努力提出的基于小波特征的识别方法。随着机器学习的方法逐渐成熟其应用的触角也延伸到了该领域,如基于循环神经网络的机械通气人机不同步检测方法。
目前基于机器学习或深度学习的分类人机不同步现象的过程可以概括为:(1)获取从呼吸机导出的呼吸力学波形数据;(2)经过专业医师对人机不同步的波形进行标注;(3)数据预处理以及划分数据集为训练集、验证集和测试集;(4)将预处理的数据输入到已经定义好的模型中进行模 型训练;(5)保存训练好的模型加以应用。
现有方案在解决机械通气过程中人机不同步方法时往往只针对一种不同步现象(常见的是无效吸气努力)进行分类,而现实情况正如前文所述,患者在一次机械通气过程中可能发生的人机不同步现象有多种,而非一成不变的如无效吸气努力。此外,医生在实施通气治疗的过程中会根据患者病情发展变化而会呼吸机通气模式作出调整,这个时候又有可能出现其它种类的不同步现象,而目前的解决方案无法适应实际的临床需要,因此,需要可以同时分类多种人机通气不同步现象的检测方法。
发明内容
本发明实施例提供了一种基于模糊熵特征提取的人机通气异步检测模型及装置,以至少解决现有技术无法同时用于检测多种人机提取不同步现象。
根据本发明的一实施例,提供了一种基于模糊熵特征提取的人机通气异步检测模型,包括以下步骤:
S1:采集当前呼吸周期的呼吸波形,选取呼吸波形中合适的通道数据,呼吸波形包括潮气量波形、气道压力时间波形及流速时间波形;
S2:对选取的呼吸波形进行标注;
S3:依次对所述潮气量波形、气道压力时间波形及流速时间波形,以形成呼吸周期内两个窗口子序列;
S4:依次对两个窗口子序列构造嵌入矩阵;
S5:基于构造的两个嵌入矩阵计算模糊熵;
S6:对气道压力时间波形及流速时间波形进行模糊熵提取,得到特征矩阵,通过对特征矩阵进行扁平化处理,得到模糊熵特征向量,模糊熵特征向量为一个呼吸周期的训练样本;
S7:将所有的呼吸周期波形重复S1-S6的过程,得到数量为N的训练 样本,表示为(N,6)大小的矩阵,同时对应得到N个样本的标签,存入(N,1)大小的列向量中,其中,将双重触发、无效吸气努力及正常通气分别标记为[1,2,3],生成人机通气异步检测模型。
进一步地,对选取的呼吸波形进行标注具体为:
对每个呼吸周期的呼吸波形均进行人机通气异步的类型标记,类型包括双重触发、无效吸气努力及正常通气。
进一步地,所述依次对所述潮气量波形、气道压力时间波形及流速时间波形,以形成呼吸周期内两个窗口子序列具体为:
窗口大小为40,步长为40。
进一步地,依次对两个窗口子序列构造嵌入矩阵具体为:
通过矩阵计算公式构造嵌入矩阵,矩阵计算公式为:
Figure PCTCN2021138106-appb-000001
Figure PCTCN2021138106-appb-000002
其中1<i<N-m+1,m为计算熵特征而嵌入的维度,N是子窗口下时间序列的长度。
进一步地,基于构造的两个嵌入矩阵计算模糊熵中,计算模糊熵的具体过程为:
通过无穷范数计算公式分别对构造的两个嵌入矩阵计算嵌入维度向量之间的无穷范数,无穷范数计算公式为:
Figure PCTCN2021138106-appb-000003
其中k=1,2,…,m;
计算相似度,定义其为如下表达式:
Figure PCTCN2021138106-appb-000004
其中r为阈值,通常为r=0.25*std,std为该输入波形片段的标准差,n为超参数;
计算模糊函数Φ m(n,r):
Figure PCTCN2021138106-appb-000005
其中,取m:=m+1;
根据模糊熵公式计算模糊熵,模糊熵公式为:
FuzzyEn(m,n,r)=ln(Φ m(n,r))-ln(Φ m+1(n,r))。
进一步地,对气道压力时间波形及流速时间波形进行模糊熵提取,得到特征矩阵,通过对特征矩阵进行扁平化处理,得到模糊熵特征向量具体为:
对每个呼吸周期的每个通道的呼吸波形进行提取,得到两个模糊熵特征,形状为(2,1)特征矩阵;
重复对气道压力时间波形及流速时间波形进行相同的模糊熵提取,得到(2,3)的特征矩阵;
将所得到的特征矩阵进行扁平化后得到形如(1,6)的模糊熵特征向量。
一种基于模糊熵特征提取的人机通气异步检测装置,包括:
波形选取模块,用于采集当前呼吸周期的呼吸波形,选取呼吸波形中合适的通道数据,呼吸波形包括潮气量波形、气道压力时间波形及流速时间波形;
波形标注模块,用于对选取的呼吸波形进行标注;
波形加窗模块,用于依次对所述潮气量波形、气道压力时间波形及流速时间波形,以形成呼吸周期内两个窗口子序列;
矩阵构造模块,用于依次对两个窗口子序列构造嵌入矩阵;
模糊熵计算模块,用于基于构造的两个嵌入矩阵计算模糊熵;
样本训练模块,用于对气道压力时间波形及流速时间波形进行模糊熵提取,得到特征矩阵,通过对特征矩阵进行扁平化处理,得到模糊熵特征向量,模糊熵特征向量为一个呼吸周期的训练样本;
模型训练模块,用于将所有的呼吸周期波形重复执行波形选取模块100至样本训练模块600的过程,得到数量为N的训练样本,表示为(N,6)大小的矩阵,同时对应得到N个样本的标签,存入(N,1)大小的列向量中,其中,将双重触发、无效吸气努力及正常通气分别标记为[1,2,3]。
进一步地,波形标注模块包括:
类型标记单元,用于对每个呼吸周期的呼吸波形均进行人机通气异步的类型标记,类型包括双重触发、无效吸气努力及正常通气。
进一步地,装置还包括:
窗口设置模块,用于设置窗口大小为40,步长为40。
一种计算机可读存储介质,计算机可读存储介质存储一个或多个程序,一个或多个程序可被一个或多个处理器执行,以实现如上述任意一项的基于模糊熵特征提取的人机通气异步检测模型中的步骤。
本发明实施例中的基于模糊熵特征提取的人机通气异步检测模型及装置,通过对采集到的具有多种人机不同现象的呼吸波形,基于模糊熵的特征提取方法,对多通道呼吸波形进行模糊熵特征提取,分别以一定的参数进行模糊熵特征计算,然后将三个维度的模糊熵特征扁平化到一维构成一个呼吸周期的特征样本,重复的将所有的呼吸周期波形进行特征提取计算,最后得到一个可同时分类或检测多种人机通气不同步现象的模型。
附图说明
此处所说明的附图用来提供对本发明的进一步理解,构成本申请的一部分,本发明的示意性实施例及其说明用于解释本发明,并不构成对本发明的不当限定。在附图中:
图1为本发明基于模糊熵特征提取的人机通气异步检测模型的流程图;
图2为本发明各模型在测试集上的混淆矩图;
图3为本发明基于模糊熵特征提取的人机通气异步检测装置原理图。
具体实施方式
为了使本技术领域的人员更好地理解本发明方案,下面将结合本发明实施例中的附图,对本发明实施例中的技术方案进行清楚、完整地描述,显然,所描述的实施例仅仅是本发明一部分的实施例,而不是全部的实施例。基于本发明中的实施例,本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例,都应当属于本发明保护的范围。
需要说明的是,本发明的说明书和权利要求书及上述附图中的术语“第一”、“第二”等是用于区别类似的对象,而不必用于描述特定的顺序或先后次序。应该理解这样使用的数据在适当情况下可以互换,以便这里描述的本发明的实施例能够以除了在这里图示或描述的那些以外的顺序实施。此外,术语“包括”和“具有”以及他们的任何变形,意图在于覆盖不排他的包含,例如,包含了一系列步骤或单元的过程、方法、系统、产品或设备不必限于清楚地列出的那些步骤或单元,而是可包括没有清楚地列出的或对于这些过程、方法、产品或设备固有的其它步骤或单元。
参见图1,根据本发明一实施例,提供了一种基于模糊熵特征提取的人机通气异步检测模型,包括以下步骤:
S1:采集当前呼吸周期的呼吸波形,选取呼吸波形中合适的通道数据,呼吸波形包括潮气量波形、气道压力时间波形及流速时间波形;
S2:对选取的呼吸波形进行标注;
S3:依次对所述潮气量波形、气道压力时间波形及流速时间波形,以形成呼吸周期内两个窗口子序列;
S4:依次对两个窗口子序列构造嵌入矩阵;
S5:基于构造的两个嵌入矩阵计算模糊熵;
S6:对气道压力时间波形及流速时间波形进行模糊熵提取,得到特征矩阵,通过对特征矩阵进行扁平化处理,得到模糊熵特征向量,模糊熵特征向量为一个呼吸周期的训练样本;
S7:将所有的呼吸周期波形重复S1-S6的过程,得到数量为N的训练样本,表示为(N,6)大小的矩阵,同时对应得到N个样本的标签,存入(N,1)大小的列向量中,其中,将双重触发、无效吸气努力及正常通气分别标记为[1,2,3],生成人机通气异步检测模型。
本发明针对常见于重病监护室机械通气过程中人机不同步现象,提出了一种新的检测分类方法,该方法可同时用于分类多种人机不同步现象。其基本内容是对由医师标注后的原始呼吸波形,包括流速波形、气道压力时间波形、潮气量时间波形,分别以一定的参数进行模糊熵特征计算,然后将三个维度的模糊熵特征扁平化到一维构成一个呼吸周期的特征样本用于后面分类模型的学习训练。
通过对采集到的具有多种人机不同现象的呼吸波形,基于模糊熵的特征提取方法,对多通道呼吸波形进行模糊熵特征提取,然后将其放入所构建的机器学习模型中进行训练,分别以一定的参数进行模糊熵特征计算,然后将三个维度的模糊熵特征扁平化到一维构成一个呼吸周期的特征样本,重复的将所有的呼吸周期波形进行特征提取计算,最后得到一个可同时分类多种人机不同步的现象的模型。
下面以具体实施例,对本发明的基于模糊熵特征提取的人机通气异步检测模型进行详细说明:
步骤一:选取合适的通道数据;考虑了波形之间的相关性、以及是否 可以保证不同通气模式下特征提取分类的有效性等因素,对呼吸波形的原始数据选取三个维度的呼吸波形包括潮气量波形、气道压力时间波形和流速时间波形,分别对此三种波形提取熵特征。
步骤二:上述呼吸波形需要经过医师对其标注,标注的过程为对每个呼吸周期的波形均有一个标记,具体标记为人机不同步的类型,包括DT(双重触发)、IEE(无效吸气努力)和Normal(正常通气)。
步骤三:由于采集频率为30Hz,即在一个呼吸周期内大约有80-90个时间点;依次对所述潮气量波形、气道压力时间波形及流速时间波形,窗口大小为40,步长也为40,首先对潮气量波形的数据进行加窗,因此一个呼吸周期内就有两个窗口子序列。
步骤四:在构造嵌入矩阵中,通过矩阵计算公式构造嵌入矩阵,依次对步骤三中的两个窗口构造嵌入矩阵,矩阵计算公式的数学表达式如下:
Figure PCTCN2021138106-appb-000006
Figure PCTCN2021138106-appb-000007
其中1<i<N-m+1,m为计算熵特征而嵌入的维度,N是子窗口下时间序列的长度。
步骤五:基于构造的两个嵌入矩阵计算模糊熵;其中,包括以下步骤:
第一步:计算嵌入维度向量两两之间的无穷范数;通过无穷范数计算公式分别对步骤四中构造的两个嵌入矩阵计算嵌入维度向量之间的无穷范数,无穷范数计算公式如下:
Figure PCTCN2021138106-appb-000008
其中k=1,2,…,m。
第二步:计算相似度。定义其为如下表达式:
Figure PCTCN2021138106-appb-000009
其中r为阈值,通常为r=0.25*std,std为该输入波形片段的标准差,在本方案中取r=std,n为超参数,本方案中取2。
第三步:计算模糊函数Φ m(n,r):
Figure PCTCN2021138106-appb-000010
Φ m(n,r)是模糊熵公式中的一个因子。
第四步:取m:=m+1,重复步骤三至步骤五的第三步的过程。
m:=m+1,表示将m更新为m+1,然后重复上述所有步骤。
第五步:根据模糊熵计算模糊熵FuzzyEn(m,n,r),模糊熵公式为:
FuzzyEn(m,n,r)=ln(Φ m(n,r))-ln(Φ m+1(n,r))
步骤六:按照步骤三至步骤五的过程,对每个呼吸周期的每个通道的波形可以得到两个模糊熵特征,形状为(2,1)矩阵,重复对气道压力时间波形及流速时间波形进行相同的模糊熵提取过程,最后可以得到(2,3)的特征矩阵,将其扁平化后可以得到形如(1,6)的模糊熵特征向量,该向量即为一个呼吸周期的训练样本。
步骤七:重复上述过程,将所有的呼吸周期波形输入到该特征提取算法中,最后可以得到的样本数为N,表示为(N,6)大小矩阵,同时对应得到N个样本的标签,存入(N,1)大小的列向量中。此处将双重触发(DT)、无效吸气努力(IEE)和正常通气(Normal)分别标记为[1,2,3],生成人机通气异步检测模型。
此外,模型训练的过程使用现有的几种机器学习算法分别对该特征进行了学习。其中包括逻辑回归(logistic regression,LR)、支持向量机(SVM)、决策树(Decision Tree,DT)和多层感知机(MLP)。
训练后各模型的结果分别如表1所示。可以看出除了决策树的评分相对较低外,其它三个模型的结果均相差无几,这也体现出了该特征具体较 好的模型适应性。
表1各机器学习算法评分表
Figure PCTCN2021138106-appb-000011
相较于现有的方案而言,本发明方案有以下特点:
1.以往的方案多是针对一种不同步现象进行分类,属于二分类任务,本发明的关键点在于首次将模糊熵应用在机械通气人机不同步现象多分类任务中,但不局限于本方案所实验的三分类任务,也可将分类任务数提高到四类、五类等。
2.本方案对从采集到的呼吸波形数据有选择地选取了潮气量时间波形、气道压力时间波形和流速时间波形,该选择不同于以往的二分类任务中仅以气道压力时间波形作为选项进行分类,本方案的选取方法综合考虑了波形之间的相关性、以及是否可以保证不同通气模式下特征提取分类的有效性等因素。
需要说明的是,本发明选择并不局限于仅仅选取这三个维度的波形作 为特征提取的要素,也可以根据上述方法选择其它三个或多个维度的波形作为特征提取的要素。
3.本方案利用模糊熵算法对呼吸波形的求取特征的过程步骤三至步骤六的步骤,具体如下:
3-1.首先是对每个呼吸周期的信号进行加窗分成不同的子窗口序列的方法,其中包括窗口大小和步长两个参数,参数可以根据数据采样频率和数据长度进行适当选取,不局限于本方案中的数值。
3-2.本方案是针对一维时间序列分别求取模糊熵,在实现的过程中使用的顺序提取过程,也就是说依次提取潮气量时间波形、气道压力时间波形和流速时间波形的模糊熵,但是不影响算法可以同时对三个通道时间序列求取模糊熵。
3-3.对于多个通道的时间序列,使用模糊熵提取到的特征采用扁平化的方法使其降为一维,从而可以方便的作为机器学习算法的样本输入。
相较于现有的方案而言,本方案有以下明显的优点:
1.现有技术方案的计算较为复杂,本发明方案所提出的直接针对原始信号求模糊熵特征,然后作为机器学习算法的输入样本,现有技术的准确率与F1-score均低于本发明方案的96%。除此之外,本方案是针对多分类的任务而言的而前者则仅仅对无效吸气努力现象作分类。
2.使用模糊熵算法提取呼吸波形特征的方法相对简单,且具有较强的模型适应能力,正如表1所示,对大多数简单常见的机器学习模型便可轻易达到相对较高的分类准确率。
3.本发明方案使用的方法直接针对多种机械通气人机不同步现象进行学习分类,对比于目前的二分类方案,显然更具有临床实际意义。
参考图2,本发明方案采用了多种机器学习算法构造模型,并进行了实验,具体如下:
实验采用的数据集采自模拟肺中预置的患者数据,病例选择的是ARDS患者,患者自发呼吸率为21,通气模式为CPAP/PSV,数据采样率为50Hz。经过标注后的数据集中包含双重触发(DT)类型的小型有1530个周期,无效吸气努力(IEE)的波形包含1447个周期,正常波形(Normal)波形包含1360个周期。按照训练集与测试集常用的划分比例,本方案选取测试集占样本总数的20%。
实验结果表明,对于大多数的机器学习算法,本发明方案提出的特征提取方法均可作为训练样本,且有不俗的效果,证明了模糊熵提取的样本用于机械通气过程人机不同步现象多分类任务是可行的。各模型在测试集上预测后其混淆矩阵如图1所示;其中,多层感知机在训练过程中,有多个参数可以选择,因此对该算法同时测试了多个参数,最终选取了效果最好的一个参数并使其体现在表1中的结果,图2展示了各种参数下的训练过程随着迭代次数增加损失函数的变化图。
图2中,依次是四个模型测试结果的混淆矩阵,从左至右从上至下,四个模型依次是决策树模型、逻辑回归模型、多层感知机模型和支持向量机模型。表1中的结果即可以从这四个混淆矩阵中计算得出。
参考图3,根据本发明一实施例,提供了一种基于模糊熵特征提取的人机通气异步检测装置,包括:
波形选取模块100,用于采集当前呼吸周期的呼吸波形,选取呼吸波形中合适的通道数据,呼吸波形包括潮气量波形、气道压力时间波形及流速时间波形;
波形标注模块200,用于对选取的呼吸波形进行标注;
波形加窗模块300,用于依次对所述潮气量波形、气道压力时间波形及流速时间波形,以形成呼吸周期内两个窗口子序列;
矩阵构造模块400,用于依次对两个窗口子序列构造嵌入矩阵;
模糊熵计算模块500,用于基于构造的两个嵌入矩阵计算模糊熵;
样本训练模块600,用于对气道压力时间波形及流速时间波形进行模糊熵提取,得到特征矩阵,通过对特征矩阵进行扁平化处理,得到模糊熵特征向量,模糊熵特征向量为一个呼吸周期的训练样本;
模型训练模块700,用于将所有的呼吸周期波形重复S1-S6的过程,得到数量为N的训练样本,表示为(N,6)大小的矩阵,同时对应得到N个样本的标签,存入(N,1)大小的列向量中,其中,将双重触发、无效吸气努力及正常通气分别标记为[1,2,3],生成人机通气异步检测模型。
本发明实施例中的基于模糊熵特征提取的人机通气异步检测装置,通过对采集到的具有多种人机不同现象的呼吸波形,基于模糊熵的特征提取方法,对多通道呼吸波形进行模糊熵特征提取,分别以一定的参数进行模糊熵特征计算,然后将三个维度的模糊熵特征扁平化到一维构成一个呼吸周期的特征样本,重复的将所有的呼吸周期波形进行特征提取计算,最后得到一个可同时分类多种人机不同步的现象的模型。
实施例中,波形标注模块包括:
类型标记单元,用于对每个呼吸周期的呼吸波形均进行人机通气异步的类型标记,类型包括双重触发、无效吸气努力及正常通气。
实施例中,装置还包括:
窗口设置模块,用于设置窗口大小为40,步长为40。
基于上述基于模糊熵特征提取的人机通气异步检测模型,本实施例提供了一种计算机可读存储介质,计算机可读存储介质存储一个或多个程序,一个或多个程序可被一个或多个处理器执行,以实现如上述实施例的基于模糊熵特征提取的人机通气异步检测模型中的步骤。
以上所述仅是本发明的优选实施方式,应当指出,对于本技术领域的普通技术人员来说,在不脱离本发明原理的前提下,还可以做出若干改进和润饰,这些改进和润饰也应视为本发明的保护范围。

Claims (10)

  1. 一种基于模糊熵特征提取的人机通气异步检测模型,其特征在于,包括以下步骤:
    S1:采集当前呼吸周期的呼吸波形,选取所述呼吸波形中合适的通道数据,所述呼吸波形包括潮气量波形、气道压力时间波形及流速时间波形;
    S2:对选取的所述呼吸波形进行标注;
    S3:依次对所述潮气量波形、气道压力时间波形及流速时间波形进行加窗处理,以形成呼吸周期内两个窗口子序列;
    S4:依次对两个所述窗口子序列构造嵌入矩阵;
    S5:基于构造的两个所述嵌入矩阵计算模糊熵;
    S6:对所述气道压力时间波形及流速时间波形进行模糊熵提取,得到特征矩阵,通过对所述特征矩阵进行扁平化处理,得到模糊熵特征向量,所述模糊熵特征向量为一个呼吸周期的训练样本;
    S7:将所有的呼吸周期波形重复S1-S6的过程,得到数量为N的训练样本,表示为(N,6)大小的矩阵,同时对应得到N个样本的标签,存入(N,1)大小的列向量中,其中,将双重触发、无效吸气努力及正常通气分别标记为[1,2,3],生成人机通气异步检测模型。
  2. 根据权利要求1所述的基于模糊熵特征提取的人机通气异步检测模型,其特征在于,所述对选取的所述呼吸波形进行标注具体为:
    对每个呼吸周期的呼吸波形均进行人机通气异步的类型标记,类型包括双重触发、无效吸气努力及正常通气。
  3. 根据权利要求1所述的基于模糊熵特征提取的人机通气异步检测模型,其特征在于,所述依次对所述潮气量波形、气道压力时间波形及流速时间波形进行加窗处理,以形成呼吸周期内两个窗口子序列具体为:
    所述窗口大小为40,步长为40。
  4. 根据权利要求1所述的基于模糊熵特征提取的人机通气异步检测模型,其特征在于,所述依次对两个所述窗口子序列构造嵌入矩阵具体为:
    通过矩阵计算公式构造所述嵌入矩阵,所述矩阵计算公式为:
    Figure PCTCN2021138106-appb-100001
    Figure PCTCN2021138106-appb-100002
    其中1<i<N-m+1,m为计算熵特征而嵌入的维度,N是子窗口下时间序列的长度。
  5. 根据权利要求1所述的基于模糊熵特征提取的人机通气异步检测模型,其特征在于,所述基于构造的两个所述嵌入矩阵计算模糊熵中,计算模糊熵的具体过程为:
    通过无穷范数计算公式分别对构造的两个所述嵌入矩阵计算嵌入维度向量之间的无穷范数,所述无穷范数计算公式为:
    Figure PCTCN2021138106-appb-100003
    其中k=1,2,…,m;
    计算相似度,定义其为如下表达式:
    Figure PCTCN2021138106-appb-100004
    其中r为阈值,通常为r=0.25*std,std为该输入波形片段的标准差,n为超参数;
    计算模糊函数Φ m(n,r):
    Figure PCTCN2021138106-appb-100005
    其中,取m:=m+1;
    根据模糊熵公式计算模糊熵,所述模糊熵公式为:
    FuzzyEn(m,n,r)=ln(Φ m(n,r))-ln(Φ m+1(n,r))。
  6. 根据权利要求1所述的基于模糊熵特征提取的人机通气异步检测模型,其特征在于,所述对所述气道压力时间波形及流速时间波形进行模糊熵提取,得到特征矩阵,通过对所述特征矩阵进行扁平化处理,得到模糊熵特征向量具体为:
    对每个呼吸周期的每个通道的呼吸波形进行提取,得到两个模糊熵特征,形状为(2,1)特征矩阵;
    重复对所述气道压力时间波形及流速时间波形进行相同的模糊熵提取,得到(2,3)的特征矩阵;
    将所得到的所述特征矩阵进行扁平化后得到形如(1,6)的模糊熵特征向量。
  7. 一种基于模糊熵特征提取的人机通气异步检测装置,其特征在于,包括:
    波形选取模块,用于采集当前呼吸周期的呼吸波形,选取所述呼吸波形中合适的通道数据,所述呼吸波形包括潮气量波形、气道压力时间波形及流速时间波形;
    波形标注模块,用于对选取的所述呼吸波形进行标注;
    波形加窗模块,用于依次对所述潮气量波形、气道压力时间波形及流速时间波形,以形成呼吸周期内两个窗口子序列;
    矩阵构造模块,用于依次对两个所述窗口子序列构造嵌入矩阵;
    模糊熵计算模块,用于基于构造的两个所述嵌入矩阵计算模糊熵;
    样本训练模块,用于对所述气道压力时间波形及流速时间波形进行模糊熵提取,得到特征矩阵,通过对所述特征矩阵进行扁平化处理,得到模糊熵特征向量,所述模糊熵特征向量为一个呼吸周期的训练样本;
    模型训练模块,用于将所有的呼吸周期波形重复执行波形选取模块 100至样本训练模块600的过程,得到数量为N的训练样本,表示为(N,6)大小的矩阵,同时对应得到N个样本的标签,存入(N,1)大小的列向量中,其中,将双重触发、无效吸气努力及正常通气分别标记为[1,2,3]。
  8. 根据权利要求7所述的基于模糊熵特征提取的人机通气异步检测装置,其特征在于,所述波形标注模块包括:
    类型标记单元,用于对每个呼吸周期的呼吸波形均进行人机通气异步的类型标记,类型包括双重触发、无效吸气努力及正常通气。
  9. 根据权利要求7所述的基于模糊熵特征提取的人机通气异步检测装置,其特征在于,所述装置还包括:
    窗口设置模块,用于设置所述窗口大小为40,步长为40。
  10. 一种计算机可读存储介质,其特征在于,所述计算机可读存储介质存储一个或多个程序,所述一个或多个程序可被一个或多个处理器执行,以实现如权利要求1-6任意一项所述的基于模糊熵特征提取的人机通气异步检测模型中的步骤。
PCT/CN2021/138106 2021-12-02 2021-12-14 一种基于模糊熵特征提取的人机通气异步检测模型及装置 WO2023097785A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202111461918.2 2021-12-02
CN202111461918.2A CN114288500B (zh) 2021-12-02 2021-12-02 一种基于模糊熵特征提取的人机通气异步检测模型及装置

Publications (1)

Publication Number Publication Date
WO2023097785A1 true WO2023097785A1 (zh) 2023-06-08

Family

ID=80966560

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/138106 WO2023097785A1 (zh) 2021-12-02 2021-12-14 一种基于模糊熵特征提取的人机通气异步检测模型及装置

Country Status (2)

Country Link
CN (1) CN114288500B (zh)
WO (1) WO2023097785A1 (zh)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109893732A (zh) * 2019-02-28 2019-06-18 杭州智瑞思科技有限公司 一种基于循环神经网络的机械通气人机不同步检测方法
CN111563451A (zh) * 2020-05-06 2020-08-21 浙江工业大学 基于多尺度小波特征的机械通气无效吸气努力识别方法
WO2021122011A1 (en) * 2019-12-16 2021-06-24 Consorci Corporació Sanitària Parc A device and method for respiratory monitoring in mechanically ventilated patients
CN113539501A (zh) * 2021-06-25 2021-10-22 中国科学院深圳先进技术研究院 一种呼吸机人机异步分类方法、系统、终端以及存储介质
CN113599647A (zh) * 2021-08-18 2021-11-05 深圳先进技术研究院 呼吸机机械通气的通气模式匹配方法、装置和相关设备

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101201980B (zh) * 2007-12-19 2010-06-02 北京交通大学 一种基于语音情感识别的远程汉语教学系统
CN102779510B (zh) * 2012-07-19 2013-12-18 东南大学 基于特征空间自适应投影的语音情感识别方法
WO2018090467A1 (zh) * 2016-11-17 2018-05-24 丹阳华神电器有限公司 基于模糊熵的含噪信号处理方法及迭代奇异谱软阈值去噪方法
AU2018366291B2 (en) * 2017-11-09 2020-09-03 Autonomous Healthcare, Inc. Clinical decision support system for patient-ventilator asynchrony detection and management
CN111143586B (zh) * 2019-08-09 2023-09-08 华为技术有限公司 一种图片处理方法及相关装置
CN110991364A (zh) * 2019-12-09 2020-04-10 四川大学 脑电信号分类方法及系统
CN111476323B (zh) * 2020-06-01 2023-05-12 合肥工业大学 一种轴承故障分类方法及系统
CN113642512B (zh) * 2021-08-30 2023-10-24 深圳先进技术研究院 呼吸机人机异步检测方法、装置、设备及存储介质

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109893732A (zh) * 2019-02-28 2019-06-18 杭州智瑞思科技有限公司 一种基于循环神经网络的机械通气人机不同步检测方法
WO2021122011A1 (en) * 2019-12-16 2021-06-24 Consorci Corporació Sanitària Parc A device and method for respiratory monitoring in mechanically ventilated patients
CN111563451A (zh) * 2020-05-06 2020-08-21 浙江工业大学 基于多尺度小波特征的机械通气无效吸气努力识别方法
CN113539501A (zh) * 2021-06-25 2021-10-22 中国科学院深圳先进技术研究院 一种呼吸机人机异步分类方法、系统、终端以及存储介质
CN113599647A (zh) * 2021-08-18 2021-11-05 深圳先进技术研究院 呼吸机机械通气的通气模式匹配方法、装置和相关设备

Also Published As

Publication number Publication date
CN114288500A (zh) 2022-04-08
CN114288500B (zh) 2024-03-19

Similar Documents

Publication Publication Date Title
Gao et al. An effective LSTM recurrent network to detect arrhythmia on imbalanced ECG dataset
JP2022523741A (ja) 描写および分類のための心電図処理システム
Alghamdi et al. A novel blood pressure estimation method based on the classification of oscillometric waveforms using machine-learning methods
CN107106028B (zh) 用于心肺睡眠阶段分类的系统和方法
CN110459328B (zh) 临床监护设备
WO2022267381A1 (zh) 一种呼吸机人机异步分类方法、系统、终端以及存储介质
CN111488793A (zh) 心电信号分类方法、装置、电子设备和存储介质
WO2022267382A1 (zh) 一种呼吸机人机异步分类方法、系统、终端以及存储介质
Xia et al. Generative adversarial network with transformer generator for boosting ECG classification
Ye et al. FENet: a frequency extraction network for obstructive sleep apnea detection
Feng et al. Unsupervised semantic-aware adaptive feature fusion network for arrhythmia detection
CN112957054B (zh) 一种基于通道注意力分组残差网络的12导联心电信号分类方法
CN111685992A (zh) 一种心肺复苏反馈系统
WO2023097780A1 (zh) 机械通气过程中人机异步现象的分类方法和分类装置
Khodabakhshi et al. The attractor recurrent neural network based on fuzzy functions: An effective model for the classification of lung abnormalities
CN115778403B (zh) 心电图分析方法、装置、电子设备和存储介质
Pettas et al. Recognition of breathing activity and medication adherence using LSTM neural networks
WO2023097785A1 (zh) 一种基于模糊熵特征提取的人机通气异步检测模型及装置
CN115814222B (zh) 混合机械通气模式下人机不同步波形识别方法及相关设备
Zeng et al. A multimodal parallel method for left ventricular dysfunction identification based on phonocardiogram and electrocardiogram signals synchronous analysis
Kofman et al. Discovery of multiple level heart-sound morphological variability resulting from changes in physiological states
Xie et al. Prediction of chronic obstructive pulmonary disease exacerbation using physiological time series patterns
Melms et al. Training one model to detect heart and lung sound events from single point auscultations
CN108182974A (zh) 病症评估方法、终端设备及计算机可读介质
CN112022149A (zh) 一种基于心电信号的房颤检测方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21966209

Country of ref document: EP

Kind code of ref document: A1