WO2019105105A1 - 脉搏波血压测量装置的分类预测数据处理方法 - Google Patents
脉搏波血压测量装置的分类预测数据处理方法 Download PDFInfo
- Publication number
- WO2019105105A1 WO2019105105A1 PCT/CN2018/106183 CN2018106183W WO2019105105A1 WO 2019105105 A1 WO2019105105 A1 WO 2019105105A1 CN 2018106183 W CN2018106183 W CN 2018106183W WO 2019105105 A1 WO2019105105 A1 WO 2019105105A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- blood pressure
- data
- classification
- pressure value
- training model
- Prior art date
Links
Images
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/02—Detecting, measuring or recording pulse, heart rate, blood pressure or blood flow; Combined pulse/heart-rate/blood pressure determination; Evaluating a cardiovascular condition not otherwise provided for, e.g. using combinations of techniques provided for in this group with electrocardiography or electroauscultation; Heart catheters for measuring blood pressure
- A61B5/021—Measuring pressure in heart or blood vessels
Definitions
- the present invention relates to data processing, and in particular to a method for processing classification prediction data of a pulse wave blood pressure measuring device.
- blood pressure is one of the most important physiological parameters of the human body and is of great significance for preventing diseases such as hypertension, stroke, myocardial infarction or heart failure.
- the general idea of the algorithm is to collect the human PPG (photoplethy smography) signal, and then the data is preprocessed to extract the time domain or frequency domain features, using the correlation model (linear, SVM (support vector machine), ANN (artificial nerve) Network)) Conduct training and regression prediction.
- the actual measured data has a certain randomness, including the randomness of the collected data and the randomness established by the regression prediction model. This randomness will adversely affect the practicality of the method.
- the object of the present invention is to solve the problem that the randomness in the prior art adversely affects the applicability of the method, and proposes a method for processing classification and prediction data of a pulse wave blood pressure measuring device.
- the present invention adopts the following technical solutions:
- the method for processing classification and prediction data of a pulse wave blood pressure measuring device comprises the following steps:
- S500 Calling the optimized classification training model, predicting the blood pressure value interval category of the test object to obtain a blood pressure value interval category, thereby predicting the blood pressure value.
- step S100 includes:
- step S300 includes:
- the collected pulse wave characteristic parameters and the corresponding blood pressure value interval categories are used as sample data, and are divided into sample data with significant beat waves and sample data with insignificant beat signals;
- steps S100 through S400 are repeated with the addition of data samples.
- step S500 includes:
- S520 Calling the optimized classification training model according to the calculated characteristic parameter, and predicting a blood pressure value interval category of the test object;
- the blood pressure value is predicted according to the obtained blood pressure value interval type.
- the manner of predicting the blood pressure value in step S530 includes: a regression analysis method and a median blood pressure interval range corresponding to the predicted category as the final predicted blood pressure value.
- step S130 includes:
- the classification algorithm in step S300 includes: binary logic classification, support vector machine classification, artificial neural network classification, decision tree, and random forest.
- the pre-processing of step S110 includes removing baseline drift, filtering to remove power frequency interference, and myoelectric interference.
- the present invention also provides a computer readable storage medium:
- a computer readable storage medium storing a computer program for use with a computing device, the computer program being executed by a processor to implement the method described above.
- the specific blood pressure value regression prediction is transformed into a classification decision within a certain blood pressure range.
- the classification algorithm is used to construct the classification training model, which can reduce the prediction difficulty and reduce the actual measurement while maintaining the prediction accuracy.
- the randomness of the data affects the measurement and improves the usability.
- Figure 1 is a flow chart of the present invention
- FIG. 2 is a flow chart of step S100
- FIG. 3 is a flow chart of step S130;
- FIG. 4 is a waveform showing a significant beat wave of a periodic pulse wave according to the present invention.
- FIG. 5 is a waveform of an inconspicuous beat wave of a periodic pulse wave according to the present invention.
- Figure 6 is a flow chart of step S300
- FIG. 7 is a flowchart of step S500.
- a classification prediction data processing method of a pulse wave blood pressure measuring device includes the following steps:
- step S100 Extract features from the collected pulse wave signals and record corresponding blood pressure values.
- step S100 specifically includes sub-steps S110, S120, and S130:
- the pulse wave data (PPG signal) is separately collected for different people (total number: n), and the acquisition of the pulse wave data is mainly measured by a pulse wave collecting device, and the time length of data collected by each object is t seconds, t seconds
- the data corresponds to the effective full pulse wave waveform number d, and the feature matrix dimension is n*d.
- the blood pressure value corresponding to the pulse wave data is a blood pressure value that can be ensured by the blood pressure measuring device during the time period during which the pulse wave is acquired, including diastolic blood pressure and systolic blood pressure.
- n sets of pulse wave data and their corresponding blood pressure values Preferably, t>20s, n>>100.
- S120 Perform preprocessing on the data. Specifically: preprocessing each piece of data (t seconds) under each acquisition object, mainly designing a bandpass filter to remove baseline drift, filtering to remove power frequency interference and myoelectric interference, and adopt FIR bandpass filtering.
- the passband frequency is 1-5Hz.
- the characteristic points to be detected are the aortic valve opening point A, the systolic peak, and the dicrotic notch.
- the highest pressure point D (dicrotic peak).
- the feature points to be detected mainly include a trough point of the aortic valve and a systolic peak of the systolic period.
- the maximum value points B and D in the data can be detected using the findpeaks function.
- the minimum value points A and C in the data can be detected by inverting the data and then using the findpeaks function.
- the findpeaks is to use the difference method to realize the extreme point detection of the data, that is, the pulse wave data d 1 , d 2 , d are provided. 3 , . . . , d i , . . . , if d i >d i-1 and d i >d i+1 , it is determined that d i is a maximum point.
- the extracted characteristic parameters are as follows:
- ⁇ T time delay between the systolic and diastolic peaks
- T1-T4 time domain characteristics related to blood pressure systolic and diastolic
- T complete waveform period
- Enhanced Index is a measure of the contribution of wave reflection to systolic arterial pressure:
- Inflection point area ratio (IPA): A1 and A2 are the area of the area under the entire PPG wave separated at the inflection point:
- W1, W2 pulse width
- H/T pulse height to cycle ratio
- SI Large arteriosclerosis index
- R_slope H/T1, the slope of the rising edge of the waveform
- F_slope H / (T2 + T3 + T4), the waveform falling slope
- H, H1, H2, H3 the relative height of the pulse wave normalized
- P m Mean arterial pressure
- P s Systolic blood pressure
- P d Diastolic blood pressure
- the extracted characteristic parameters are as follows:
- Pulse cycle T systolic rise time SUT, diastolic time DT, IPA (knee area ratio, A1/A2), R_slope (rise slope: H/SUT), F_slope (fall slope: H/DT) and K (pulse wave) Waveform eigenvalue), pulse height percentage (10%, 25%, 33%, 50%, 66%, 75%) corresponds to the time width as shown in Figure 5 and Table 1 below:
- P m Mean arterial pressure
- P s Systolic blood pressure
- P d Diastolic blood pressure
- Classification label setting For the two types of blood pressure values (systolic pressure, diastolic blood pressure) corresponding to the n*d group pulse wave characteristics, a corresponding label is produced.
- the common blood pressure range of systolic blood pressure is 90-140mmHg
- the collected pulse wave characteristic parameters and the corresponding blood pressure value interval categories are used as sample data, and are divided into sample data with distinct beat waves and sample data with insignificant beat signals. Specifically: the n*d group pulse wave characteristic parameters and the corresponding blood pressure value interval categories are taken as the 1 ⁇ n*d group of sample data, wherein the number of samples with significant beat waves is n1*d, and the number of inconspicuous samples n2 *d.
- a classification algorithm selecting a classification algorithm to construct a training model for the classification training data. Aiming at the characteristic parameters of the above pulse wave data and its blood pressure classification label, a suitable classification algorithm is used to construct a classification training model for the systolic and diastolic pressures of two typical pulse wave types.
- Common classification algorithms include binary logic classification, support vector machine classification, artificial neural network classification, decision tree, random forest, etc., preferably using random forest.
- Support vector machine classification Find the best classification hyperplane classification prediction sample category, and search for the best one of all possible linear classifiers according to the distribution of training samples.
- the sample that determines the classification hyperplane is not all training data, but two of the two different categories of data points with the smallest interval. This type of data point that can be used to really help determine the optimal linear classification model is called a "support vector.”
- Integrated model classification comprehensively consider the prediction results of multiple classifiers, making decisions, divided into two types:
- the typical model is a random forest classifier, which is to build multiple decision trees simultaneously on the same training data.
- a standard decision tree will sort the influence of each feature on the prediction results, thus determining the different features to build the split from top to bottom.
- the order of the nodes in this way, the decision trees in all random forests will be constructed consistently by this strategy, thus losing diversity. Therefore, in the construction process of the random forest classifier, each decision tree will abandon this fixed sorting algorithm and randomly select features.
- the other is to build multiple classification models in a certain order. There is a dependency relationship between these models. Generally speaking, the addition of each subsequent model needs to contribute to the comprehensive performance of the existing integrated model, and then continuously improve after the update. The performance of the integrated model, and ultimately hope to build a model with stronger classification capabilities by integrating multiple classifiers with weaker classification capabilities. The more representative one is the gradient elevation decision tree. Unlike the random forest classifier model, each decision tree here will reduce the fitting error of the overall integration model on the training set as much as possible.
- Steps S100 to S400 are repeated with the addition of the data samples.
- S500 Calling the optimized classification training model, predicting the blood pressure value interval category of the test object to obtain a blood pressure value interval category, thereby predicting the blood pressure value.
- S510, S520, and S530 specifically including sub-steps S510, S520, and S530:
- S510 Collect PPG data of the test object sample, process the data, and perform feature parameter calculation. Specifically, the PPG data of the test object sample is collected for t seconds, the data is preprocessed, the waveform type is identified, and then the feature parameters are calculated.
- S520 Calling the optimized classification training model according to the calculated characteristic parameter, and predicting a blood pressure value interval category of the test object.
- the blood pressure value is predicted according to the obtained blood pressure value interval type.
- the exact blood pressure value can be predicted by combining common regression analysis methods (such as linear regression, SVR, etc.).
- the invention also provides a computer readable storage medium storing a computer program for use with a computing device, the computer program being executed by a processor to implement the method described above.
- the present invention converts the specific blood pressure value regression prediction into a classification decision within a certain blood pressure range, and sets a corresponding blood pressure interval and classifies it, and combines the selection of the effective features of the PPG signal to test a plurality of classification algorithms to obtain the most suitable one.
- the high-accuracy classification algorithm which constructs the classification training model, can reduce the prediction difficulty while maintaining the prediction accuracy, avoiding the problem that the regression model requires purely the requirements of the model feature relationship, and reduces the randomness of the actual measured data. The impact on measurement improves practicality.
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Cardiology (AREA)
- Engineering & Computer Science (AREA)
- Heart & Thoracic Surgery (AREA)
- Physiology (AREA)
- Biophysics (AREA)
- Pathology (AREA)
- Vascular Medicine (AREA)
- Biomedical Technology (AREA)
- Physics & Mathematics (AREA)
- Medical Informatics (AREA)
- Molecular Biology (AREA)
- Surgery (AREA)
- Animal Behavior & Ethology (AREA)
- General Health & Medical Sciences (AREA)
- Public Health (AREA)
- Veterinary Medicine (AREA)
- Measuring Pulse, Heart Rate, Blood Pressure Or Blood Flow (AREA)
Abstract
一种脉搏波血压测量装置的分类预测数据处理方法,包括如下步骤:从采集到的脉搏波信号中提取特征并记录相应的血压值(S100);对血压值按常见血压范围区间分类并设置分类标签(S200);将分类后的血压值数据分成训练数据和测试数据,选用分类算法构建分类训练模型(S300);对创建成功的分类训练模型,利用测试数据进行分类预测,统计分类预测准确率,根据准确率调整优化分类训练模型(S400);调用优化后的分类训练模型,对测试对象的血压值区间类别进行预测以得到血压值区间类别,从而对血压值进行预测(S500)。该方法可以在保持预测精度的前提下降低预测难度,减少实际测量的数据的随机性对测量的影响。
Description
本发明涉及数据处理,特别涉及一种脉搏波血压测量装置的分类预测数据处理方法。
本项研究工作得到了中国国家自然科学基金项目(NO.61571268)、广东省科技厅重大科技专项项目-基于智能***护仪的远程人体生理多参数实时监测与分析物联网平台与示范工程、以及深圳市***重大科技项目-基于智能手机的远程人体生理多参数实时监测与分析网络平台产业化的资助。
近年来,随着可穿戴产品越发流行,尤其是针对医疗健康领域。其中,血压作为人体最重要的生理参数之一,对预防高血压,中风,心肌梗塞或心力衰竭等疾病具有重要意义。
就传统测量血压方法来讲,医生更倾向于选择柯氏听音法或示波法。虽然该种方法测量精度高,但是柯氏听音法的准确性往往因人而异,受到临床经验的影响,示波法测量血压需要佩戴袖带,往往携带不方便。因此,无创高精度可穿戴血压测量设备受到很多人的期待。
其中,针对无创血压测量算法,相关学者做了大量的研究。算法的普遍思路是通过采集人体PPG(photoplethy smography,脉搏波)信号,然后数据进过预处理后提取时域或频域特征,利用相关模型(线性、SVM(支持向量机)、ANN(人工神经网络))进行训练与回归预测。实际测量的数据具有一定的随机性,包括采集数据的随机性及回归预测模型建立的随机性,这种随机性会对方法的实用性产生不利的影响。
发明内容
本发明的目的是为了解决现有技术中随机性会对方法的实用性产生不利影响的问题,提出一种脉搏波血压测量装置的分类预测数据处理方法。
为解决上述技术问题,本发明采用以下技术方案:
脉搏波血压测量装置的分类预测数据处理方法,包括如下步骤:
S100、从采集到的脉搏波信号中提取特征并记录相应的血压值;
S200、对血压值按常见血压范围区间分类并设置分类标签;
S300、将分类后的血压值数据分成训练数据和测试数据,选用分类算法构建分类训练模型;
S400、对创建成功的分类训练模型,利用测试数据进行分类预测,统计分类预测准确率,根据准确率调整优化分类训练模型;
S500、调用优化后的分类训练模型,对测试对象的血压值区间类别进行预测以得到血压值区间类别,从而对血压值进行预测。
在一些优选的实施方式中,步骤S100包括:
S110、采集脉搏波数据样本;
S120、对数据进行预处理;
S130、提取脉搏波时域参数。
在一些优选的实施方式中,步骤S300包括:
S310、对采集的脉搏波特征参数和对应的血压值区间类别作为样本数据,分为重搏波明显的样本数据和重搏波不明显样本数据;
S320、针对重搏波明显的样本数据,随机选择一部分数据用作重搏明显分类训练数据,其余数据用作重搏明显分类测试数据;针对重搏波不明显样本数据,随机选择一部分数据用作重搏不明显分类训练数据,其余数据用作重搏不明显分类测试数据;
S330、选用分类算法对分类训练数据构建训练模型。
在一些优选的实施方式中,在增加数据样本的情况下重复步骤S100至S400。
在一些优选的实施方式中,步骤S500包括:
S510、采集测试对象样本PPG数据,对数据进行处理以及进行特征参数计算;
S520、根据计算所得的特征参数调用优化后的分类训练模型,对测试对象的血压值区间类别进行预测;
S530、根据得到的血压值区间类别,对血压值进行预测。
在进一步优选的实施方式中,步骤S530中对血压值进行预测的方式包括:回归分析方法和通过求取预测类别对应血压区间范围中值作为最终预测血压值。
在进一步优选的实施方式中,步骤S130包括:
S131、识别统计重搏波明显与不明显的两种典型脉搏波;
S132、检测特征点;
S133、计算特征参数。
在一些优选的实施方式中,步骤S300中的分类算法包括:二元逻辑分类、支持向量机分类、人工神经网络分类、决策树和随机森林。
在进一步优选的实施方式中,步骤S110的预处理包括去除基线漂移、滤波去除工频干扰和肌电干扰。
在另一方面,本发明还提供一种计算机可读存储介质:
一种计算机可读存储介质,其存储有与计算设备结合使用的计算机程序,所述计算机程序被处理器执行以实现上述方法。
与现有技术相比,本发明的有益效果有:
将具体血压值回归预测转化为一定血压范围内的分类决策,通过设定相应血压区间并进行分类,选用分类算法构建分类训练模型,可以在保持预测精度的前提下降低预测难度,减少了实际测量的数据的随机性对测量的影响,提高实用性。
图1是本发明的流程图;
图2是步骤S100的流程图;
图3是步骤S130的流程图;
图4是本发明涉及的一个周期脉搏波的重搏波明显的波形;
图5是本发明涉及的一个周期脉搏波的重搏波不明显的波形;
图6是步骤S300的流程图;
图7是步骤S500的流程图。
以下对本发明的实施方式作详细说明。应该强调的是,下述说明仅仅是示例性的,而不是为了限制本发明的范围及其应用。
参考图1,脉搏波血压测量装置的分类预测数据处理方法包括如下步骤:
S100、从采集到的脉搏波信号中提取特征并记录相应的血压值。参考图2,步骤S100具体包括子步骤S110、S120和S130:
S110、采集脉搏波数据样本。具体为:针对不同人(总人数:n)分 别采集脉搏波数据(PPG信号),脉搏波数据的获取主要是通过脉搏波采集设备测量得到,每个对象采集数据的时长为t秒,t秒数据对应有效完整脉搏波波形数d,则特征矩阵维数为n*d。脉搏波数据对应的血压值为采集脉搏波的时间段内通过血压测量设备测量得到的可保证准确性的血压值,包括舒张压和收缩压。这里,需要测量n组脉搏波数据及其对应的血压值。优选地,t>20s,n>>100。
S120、对数据进行预处理。具体为:对每个采集对象下的每段数据(t秒)进行预处理,主要设计一个带通滤波器,实现去除基线漂移,滤波去除工频干扰和肌电干扰,可采用FIR带通滤波器,通带频率1-5Hz。
S130、提取脉搏波时域参数。针对滤波平滑后的数据,进行脉搏波时域参数的提取,参考图3,具体包括子步骤S131、S132和S133:
S131、识别统计重搏波明显与不明显的两种典型脉搏波。具体为:识别统计两种典型脉搏波,分别为重搏波明显与不明显的脉搏波,对应人数n1、n2,则n=n1+n2。
S132、检测特征点。针对两种典型脉搏波波形提取对应波形的特征点:
参考图4,针对重搏波明显的波形,需要检测的特征点有主动脉瓣开放点A(trough point),收缩期最高压力点B(systolic peak),重搏波起点C(dicrotic notch),重搏波最高压力点D(dicrotic peak)。
参考图5,针对重搏波不明显的波形,需要检测的特征点主要有主动脉瓣开放点A(trough point),收缩期最高压力点B(systolic peak)。
利用findpeaks函数可以检测出数据中的极大值点B和D。通过对数据取反再利用findpeaks函数可以检测出数据中的极小值点A和C,findpeaks是利用差分法实现对数据的极值点检测,即设有脉搏波数据d
1,d
2,d
3,……,d
i,……,如果有d
i>d
i-1且d
i>d
i+1则判定d
i为极大值点。
S133、计算特征参数。根据特征点计算对应波形类型下的特征参数:
参考图4,针对重搏波明显波形(dicrotic peak obvious),提取的特征参数如下:
△T:收缩期和舒张期峰值之间的时间延迟;
T1-T4:血压收缩期和舒张期有关时域特征;
T:完整的波形周期;
增强指数(AI):增强压力(AG)是波反射对收缩期动脉压力贡献的度量:
拐点面积比(IPA):A1和A2是在拐点处分离的整个PPG波下的区域面积:
W1、W2:脉冲宽度;
H/T:脉搏高度与周期比;
大动脉硬化指数(SI):与动脉僵硬度相关
R_slope:H/T1,波形上升沿斜率;
F_slope:H/(T2+T3+T4),波形下降斜率;
H、H1、H2、H3:脉搏波归一化下相对高度;
K:脉搏波波形特征值(Pulse wave waveform eigenvalue),由如下公式计算:
其中,P
m(Mean arterial pressure)为平均动脉压,P
s(Systolic bloodpressure)为收缩压,P
d(Diastolic blood pressure)为舒张压。
参考图5,针对重搏波不明显波形(dicrotic peak not obvious),提取的特征参数如下:
脉搏周期T,收缩期上升时间SUT,舒张期时间DT,IPA(拐点面积比,A1/A2),R_slope(上升斜率:H/SUT),F_slope(下降斜率:H/DT)和K(脉搏波波形特征值),脉搏高度百分比(10%,25%,33%,50%,66%,75%)对应时间宽度如图5和下表一:
表一 脉搏高度百分比对应的时间宽度
脉搏波波形特征值K由如下公式计算:
其中,P
m(Mean arterial pressure)为平均动脉压,P
s(Systolic blood pressure)为收缩压,P
d(Diastolic blood pressure)为舒张压。
S200、对血压值按常见血压范围区间分类并设置分类标签。对采集脉搏波信号对应的血压值按常见血压范围区间分类,设舒张压血压值区间种类数为c1,收缩压血压值区间种类数为c2。
分类标签的设置:对于n*d组脉搏波特征对应的两类血压值(收缩压,舒张压)分类制作对应标签。收缩压常见血压范围为90-140mmHg,舒张压常见血压范围为60-90mmHg。若划分血压区间间隔选为dis=10mmHg,那么对于收缩压分类标签及对应血压范围为:1(<=90mmHg),2(90~100mmHg),3(100~110mmHg),4(110~120mmHg),5(120~130mmHg),6(130~140mmHg),7(>140mmHg)。对于舒张压分类标签及对应血压范围为:1(<=60mmHg),2(60~70mmHg),3(70~80mmHg),4(80~90mmHg),5(>90mmHg)。
S300、将分类后的血压值数据分成训练数据和测试数据,选用分类算法构建分类训练模型。参考图6,具体包括子步骤S310、S320和S330:
S310、对采集的脉搏波特征参数和对应的血压值区间类别作为样本数据,分为重搏波明显的样本数据和重搏波不明显样本数据。具体的:对采集的n*d组脉搏波特征参数和对应的血压值区间类别作为样本数据的1~n*d组,其中重搏波明显的样本数为n1*d,不明显样本数n2*d。
S320、针对重搏波明显的样本数据,随机选择一部分数据用作重搏明显分类训练数据,其余数据用作重搏明显分类测试数据;针对重搏波不明显样本数据,随机选择一部分数据用作重搏不明显分类训练数据,其余数据用作重搏不明显分类测试数据。具体的:针对两类数据,随机选择k1*d组数据用作重搏明显分类训练数据,其余(n1-k1)*d数据用作重搏明显分类测试数据。同理重搏不明显分类训练数据组数为k2*d,测试数据组数为(n2-k2)*d,优选地,k1/n1=75%=k2/n2。
S330、选用分类算法对分类训练数据构建训练模型。针对上述脉搏波数据的特征参数及其血压分类标签,选用合适的分类算法对两种典型脉搏波类型的收缩压和舒张压分别构建分类训练模型。常见分类算法有二元逻辑分类、支持向量机分类、人工神经网络分类、决策树、随机森林等,优选用随机森林。
支持向量机分类:寻找最佳分类超平面分类预测样本类别,根据训练样本的分布,搜索所有可能的线性分类器中最佳的那个。决定分类超平面的样本并不是所有训练数据,而是其中的两个间隔最小的两个不同类别的数据点。这种可以用来真正帮助决策最优线性分类模型的数据点叫做“支持向量”。
集成模型分类:综合考量多个分类器的预测结果,作出决策,分为两种:
一种是利用相同的分类训练数据同时搭建多个独立的分类模型,然后通过投票的方式,以少数服从多数做出最终的分类决策。典型模型为随机森林分类器,即在相同训练数据上同时搭建多棵决策树,一株标准的决策树会根据每位特征对预测结果的影响进行排序,从而决定不同特征从上至下构建***节点的顺序,如此一来,所有随机森林中的决策树都会受这一策略影响而构建得一致,从而丧失多样性。因此随机森林分类器在构建过程中,每一个决策树都会放弃这一固定的排序算法,转而随机选取特征。
另一种是按照一定次序搭建多个分类模型,这些模型之间存在依赖关系,一般而言,每一个后续模型的加入都需要对现有集成模型的综合性能 有所贡献,进而不断提升更新过后的集成模型的性能,并最终期望借助整合多个分类能力较弱的分类器,搭建出具有更强分类能力的模型。比较有代表性的当属梯度提升决策树,与随机森林分类器模型不同的是这里每一棵决策树在生成过程中都会尽可能降低整体集成模型在训练集上的拟合误差。
S400、对创建成功的分类训练模型,利用测试数据进行分类预测,统计分类预测准确率,根据准确率调整优化分类训练模型。
在增加数据样本的情况下重复步骤S100至S400。
S500、调用优化后的分类训练模型,对测试对象的血压值区间类别进行预测以得到血压值区间类别,从而对血压值进行预测。参考图7,具体包括子步骤S510、S520和S530:
S510、采集测试对象样本PPG数据,对数据进行处理以及进行特征参数计算。具体为:采集测试对象样本PPG数据t秒,对数据进行预处理,识别波形种类,然后进行特征参数计算。
S520、根据计算所得的特征参数调用优化后的分类训练模型,对测试对象的血压值区间类别进行预测。
S530、根据得到的血压值区间类别,对血压值进行预测。
在得到分类区间类别情况下,可结合常见回归分析方法(如线性回归,SVR等)对确切血压值进行预测。也可通过求取预测类别对应血压区间范围中值作为最终预测血压值,例如在dis=10mmHg的前提下,如对收缩压所在区间预测标签为3,即预测血压所在区间范围为100~110mmHg,则预测收缩压值为(100+110)/2=105mmHg。
在另一方面,本发明还提供一种计算机可读存储介质,其存储有与计算设备结合使用的计算机程序,所述计算机程序被处理器执行以实现上述方法。
根据上述可知,本发明将具体血压值回归预测转化为一定血压范围内的分类决策,通过设定相应血压区间并进行分类,结合PPG信号有效特征的选取,试验多种分类算法,获得最适合的高精确性分类算法,从而构建分类训练模型,可以在保持预测精度的前提下降低预测难度,避免了单纯 利用回归模型要求对模型特征关系刻画要求高的问题,减少了实际测量的数据的随机性对测量的影响,提高实用性。
以上内容是结合具体/优选的实施方式对本发明所作的进一步详细说明,不能认定本发明的具体实施只局限于这些说明。对于本发明所属技术领域的普通技术人员来说,在不脱离本发明构思的前提下,其还可以对这些已描述的实施方式做出若干替代或变型,而这些替代或变型方式都应当视为属于本发明的保护范围。
Claims (10)
- 一种脉搏波血压测量装置的分类预测数据处理方法,其特征在于包括如下步骤:S100、从采集到的脉搏波信号中提取特征并记录相应的血压值;S200、对血压值按常见血压范围区间分类并设置分类标签;S300、将分类后的血压值数据分成训练数据和测试数据,选用分类算法构建分类训练模型;S400、对创建成功的分类训练模型,利用测试数据进行分类预测,统计分类预测准确率,根据准确率调整优化分类训练模型;S500、调用优化后的分类训练模型,对测试对象的血压值区间类别进行预测以得到血压值区间类别,从而对血压值进行预测。
- 根据权利要求1所述的方法,其特征在于步骤S100包括:S110、采集脉搏波数据样本;S120、对数据进行预处理;S130、提取脉搏波时域参数。
- 根据权利要求1所述的方法,其特征在于步骤S300包括:S310、对采集的脉搏波特征参数和对应的血压值区间类别作为样本数据,分为重搏波明显的样本数据和重搏波不明显样本数据;S320、针对重搏波明显的样本数据,随机选择一部分数据用作重搏明显分类训练数据,其余数据用作重搏明显分类测试数据;针对重搏波不明显样本数据,随机选择一部分数据用作重搏不明显分类训练数据,其余数据用作重搏不明显分类测试数据;S330、选用分类算法对分类训练数据构建训练模型。
- 根据权利要求1所述的方法,其特征在于:在增加数据样本的情况下重复步骤S100至S400。
- 根据权利要求1所述的方法,其特征在于步骤S500包括:S510、采集测试对象样本PPG数据,对数据进行处理以及进行特征参数计算;S520、根据计算所得的特征参数调用优化后的分类训练模型,对测试对象的血压值区间类别进行预测;S530、根据得到的血压值区间类别,对血压值进行预测。
- 根据权利要求5所述的方法,其特征在于步骤S530中对血压值进行预测的方式包括:回归分析方法和通过求取预测类别对应血压区间范围中值作为最终预测血压值。
- 根据权利要求2所述的方法,其特征在于步骤S130包括:S131、识别统计重搏波明显与不明显的两种典型脉搏波;S132、检测特征点;S133、计算特征参数。
- 根据权利要求1所述的方法,其特征在于步骤S300中的分类算法包括:二元逻辑分类、支持向量机分类、人工神经网络分类、决策树和随机森林。
- 根据权利要求2所述的方法,其特征在于步骤S110的预处理包括去除基线漂移、滤波去除工频干扰和肌电干扰。
- 一种计算机可读存储介质,其存储有与计算设备结合使用的计算机程序,所述计算机程序被处理器执行以实现权利要求1-9任一项所述方法。
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711214783.3 | 2017-11-28 | ||
CN201711214783.3A CN109833035B (zh) | 2017-11-28 | 2017-11-28 | 脉搏波血压测量装置的分类预测数据处理方法 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2019105105A1 true WO2019105105A1 (zh) | 2019-06-06 |
Family
ID=66665376
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/CN2018/106183 WO2019105105A1 (zh) | 2017-11-28 | 2018-09-18 | 脉搏波血压测量装置的分类预测数据处理方法 |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN109833035B (zh) |
WO (1) | WO2019105105A1 (zh) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111358453A (zh) * | 2020-03-17 | 2020-07-03 | 乐普(北京)医疗器械股份有限公司 | 一种血压分类预测方法和装置 |
CN111685748A (zh) * | 2020-06-15 | 2020-09-22 | 广州视源电子科技股份有限公司 | 一种基于分位数的血压预警方法、装置、设备及存储介质 |
WO2021038237A1 (en) * | 2019-08-30 | 2021-03-04 | Coventry University | Blood pressure measurement device |
CN113180623A (zh) * | 2021-06-01 | 2021-07-30 | 山东大学 | 无袖式血压测量方法、***、设备及存储介质 |
CN113288091A (zh) * | 2021-05-06 | 2021-08-24 | 广东工业大学 | 用于血压分类的模型训练方法、装置及可穿戴设备 |
CN114340483A (zh) * | 2019-09-25 | 2022-04-12 | 长桑医疗(海南)有限公司 | 一种血压标定选择方法及其建模方法 |
CN115399909A (zh) * | 2022-08-31 | 2022-11-29 | 宋人杰 | 一种为超声透皮给药机建立超声参数模型的方法 |
CN115770028A (zh) * | 2022-12-16 | 2023-03-10 | 深圳市爱都科技有限公司 | 血压检测方法、***、装置及存储介质 |
CN116383617A (zh) * | 2023-04-21 | 2023-07-04 | 复旦大学 | 一种基于脉搏波波形特征的智能血压检测方法及*** |
Families Citing this family (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20200383579A1 (en) * | 2019-06-10 | 2020-12-10 | Apple Inc. | Projecting Blood Pressure Measurements With Limited Pressurization |
CN111027629B (zh) * | 2019-12-13 | 2024-02-27 | 国网山东省电力公司莱芜供电公司 | 基于改进随机森林的配电网故障停电率预测方法及*** |
CN113243902B (zh) * | 2021-05-31 | 2021-12-07 | 之江实验室 | 一种基于光电容积脉搏波的特征提取方法 |
CN116919363A (zh) * | 2022-04-12 | 2023-10-24 | 乐普(北京)医疗器械股份有限公司 | 一种基于大数据特征的个性化血压预测方法和装置 |
CN117137456A (zh) * | 2023-10-13 | 2023-12-01 | 电子科技大学 | 一种基于可见光视觉的非接触式血压测量方法 |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20130296723A1 (en) * | 2012-05-03 | 2013-11-07 | Samsung Electronics Co., Ltd. | Portable blood pressure measuring apparatus and blood pressure measuring method in portable terminal |
CN104699931A (zh) * | 2013-12-09 | 2015-06-10 | 广州华久信息科技有限公司 | 一种基于人脸的神经网络血压预测方法及手机 |
CN105286815A (zh) * | 2015-11-02 | 2016-02-03 | 重庆大学 | 一种基于波形时域特征的脉搏波信号特征点检测方法 |
CN106037694A (zh) * | 2016-05-13 | 2016-10-26 | 吉林大学 | 一种基于脉搏波的连续血压测量装置 |
CN106413534A (zh) * | 2015-08-08 | 2017-02-15 | 深圳先进技术研究院 | 连续血压测量装置、测量模型建立方法和*** |
CN106821356A (zh) * | 2017-02-23 | 2017-06-13 | 吉林大学 | 基于Elman神经网络的云端连续血压测量方法及*** |
CN107184194A (zh) * | 2017-07-06 | 2017-09-22 | 中国科学院合肥物质科学研究院 | 基于数字控制的血压自动测量装置及方法 |
Family Cites Families (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US4917098A (en) * | 1989-02-23 | 1990-04-17 | Colin Electronics Co., Ltd. | Method and apparatus for measuring blood pressure |
DE69832782T2 (de) * | 1997-11-20 | 2006-09-21 | Seiko Epson Corp. | Vorrichtung zur pulswellenuntersuchung |
JP5151690B2 (ja) * | 2008-05-27 | 2013-02-27 | オムロンヘルスケア株式会社 | 血圧情報測定装置および指標取得方法 |
CN101732033A (zh) * | 2008-11-07 | 2010-06-16 | 中国科学院计算技术研究所 | 一种人体波形中的特征参数的提取方法和装置 |
KR101298838B1 (ko) * | 2010-11-29 | 2013-08-23 | (주)더힘스 | 혈관경화도 진단을 위한 정보 제공 방법 |
CN102894964B (zh) * | 2011-07-26 | 2014-08-20 | 深圳大学 | 一种无创血压测量方法和装置 |
CN103027667B (zh) * | 2011-09-30 | 2017-01-18 | Ge医疗***环球技术有限公司 | 脉搏波的特征参数提取 |
CN102499669B (zh) * | 2011-10-26 | 2014-12-24 | 中国科学院深圳先进技术研究院 | 心脏参数的测量装置 |
CN102397064B (zh) * | 2011-12-14 | 2014-02-19 | 中国航天员科研训练中心 | 连续血压测量装置 |
CN106073742A (zh) * | 2013-05-13 | 2016-11-09 | 天津点康科技有限公司 | 一种血压测量***及方法 |
KR101560521B1 (ko) * | 2014-06-05 | 2015-10-14 | 길영준 | 혈압을 실시간으로 모니터링하기 위한 방법, 시스템 및 비일시성의 컴퓨터 판독 가능한 기록 매체 |
CN105595979A (zh) * | 2016-01-21 | 2016-05-25 | 中山大学 | 一种基于脉搏波传播时间的无创连续血压监测方法及装置 |
GB2551201A (en) * | 2016-06-10 | 2017-12-13 | Polar Electro Oy | Multi-sensor system for estimating blood pulse wave characteristics |
CN206499449U (zh) * | 2016-09-13 | 2017-09-19 | 深圳市岩尚科技有限公司 | 一种脉搏波采集手环 |
CN106691406A (zh) * | 2017-01-05 | 2017-05-24 | 大连理工大学 | 一种基于单探头光电容积脉搏波的血管弹性和血压的检测方法 |
CN112006669B (zh) * | 2020-08-19 | 2021-06-04 | 北京雪扬科技有限公司 | 一种基于血压测量模型的双通道计算方法的血压仪 |
-
2017
- 2017-11-28 CN CN201711214783.3A patent/CN109833035B/zh active Active
-
2018
- 2018-09-18 WO PCT/CN2018/106183 patent/WO2019105105A1/zh active Application Filing
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20130296723A1 (en) * | 2012-05-03 | 2013-11-07 | Samsung Electronics Co., Ltd. | Portable blood pressure measuring apparatus and blood pressure measuring method in portable terminal |
CN104699931A (zh) * | 2013-12-09 | 2015-06-10 | 广州华久信息科技有限公司 | 一种基于人脸的神经网络血压预测方法及手机 |
CN106413534A (zh) * | 2015-08-08 | 2017-02-15 | 深圳先进技术研究院 | 连续血压测量装置、测量模型建立方法和*** |
CN105286815A (zh) * | 2015-11-02 | 2016-02-03 | 重庆大学 | 一种基于波形时域特征的脉搏波信号特征点检测方法 |
CN106037694A (zh) * | 2016-05-13 | 2016-10-26 | 吉林大学 | 一种基于脉搏波的连续血压测量装置 |
CN106821356A (zh) * | 2017-02-23 | 2017-06-13 | 吉林大学 | 基于Elman神经网络的云端连续血压测量方法及*** |
CN107184194A (zh) * | 2017-07-06 | 2017-09-22 | 中国科学院合肥物质科学研究院 | 基于数字控制的血压自动测量装置及方法 |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2021038237A1 (en) * | 2019-08-30 | 2021-03-04 | Coventry University | Blood pressure measurement device |
CN114340483A (zh) * | 2019-09-25 | 2022-04-12 | 长桑医疗(海南)有限公司 | 一种血压标定选择方法及其建模方法 |
CN111358453A (zh) * | 2020-03-17 | 2020-07-03 | 乐普(北京)医疗器械股份有限公司 | 一种血压分类预测方法和装置 |
CN111358453B (zh) * | 2020-03-17 | 2022-07-29 | 乐普(北京)医疗器械股份有限公司 | 一种血压分类预测方法和装置 |
CN111685748A (zh) * | 2020-06-15 | 2020-09-22 | 广州视源电子科技股份有限公司 | 一种基于分位数的血压预警方法、装置、设备及存储介质 |
CN113288091A (zh) * | 2021-05-06 | 2021-08-24 | 广东工业大学 | 用于血压分类的模型训练方法、装置及可穿戴设备 |
CN113288091B (zh) * | 2021-05-06 | 2023-10-03 | 广东工业大学 | 用于血压分类的模型训练方法、装置及可穿戴设备 |
CN113180623A (zh) * | 2021-06-01 | 2021-07-30 | 山东大学 | 无袖式血压测量方法、***、设备及存储介质 |
CN115399909A (zh) * | 2022-08-31 | 2022-11-29 | 宋人杰 | 一种为超声透皮给药机建立超声参数模型的方法 |
CN115770028A (zh) * | 2022-12-16 | 2023-03-10 | 深圳市爱都科技有限公司 | 血压检测方法、***、装置及存储介质 |
CN116383617A (zh) * | 2023-04-21 | 2023-07-04 | 复旦大学 | 一种基于脉搏波波形特征的智能血压检测方法及*** |
CN116383617B (zh) * | 2023-04-21 | 2023-09-22 | 复旦大学 | 一种基于脉搏波波形特征的智能血压检测方法及*** |
Also Published As
Publication number | Publication date |
---|---|
CN109833035A (zh) | 2019-06-04 |
CN109833035B (zh) | 2021-12-07 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2019105105A1 (zh) | 脉搏波血压测量装置的分类预测数据处理方法 | |
Sk et al. | Coronary Heart Disease Prediction and Classification using Hybrid Machine Learning Algorithms | |
TWI596600B (zh) | 識別生理聲音的方法以及系統 | |
WO2019100560A1 (zh) | 基于人工智能自学习的心电图自动分析方法和装置 | |
Pathinarupothi et al. | Instantaneous heart rate as a robust feature for sleep apnea severity detection using deep learning | |
CN102488503B (zh) | 连续血压测量装置 | |
WO2019161610A1 (zh) | 心电信息处理方法和心电工作站*** | |
WO2017024457A1 (zh) | 连续血压测量装置、测量模型建立方法和*** | |
CN107028603B (zh) | 使用脉搏触诊信号来检测人体内的糖尿病的装置和方法 | |
WO2019161611A1 (zh) | 心电信息处理方法和心电工作站 | |
Tang et al. | Classification of normal/abnormal heart sound recordings based on multi-domain features and back propagation neural network | |
Fang et al. | Dual-channel neural network for atrial fibrillation detection from a single lead ECG wave | |
CN107358556A (zh) | 基于物联网的健康监测及评价平台 | |
WO2021164347A1 (zh) | 一种对血压进行预测的方法和装置 | |
CN111557658B (zh) | Ppg实时心率信号质量评估方法及装置、存储介质 | |
CN116451110A (zh) | 基于信号能量特征和脉搏周期的血糖预测模型构建方法 | |
Yang et al. | Automated intra-patient and inter-patient coronary artery disease and congestive heart failure detection using EFAP-Net | |
Banerjee et al. | Non-invasive detection of coronary artery disease based on clinical information and cardiovascular signals: A two-stage classification approach | |
CN111329467A (zh) | 一种基于人工智能的心脏疾病辅助检测方法 | |
Chou et al. | Modelling arterial blood pressure waveforms for extreme bradycardia and tachycardia by curve fitting with Gaussian functions | |
US20180153415A1 (en) | Aortic stenosis classification | |
Horoba et al. | Recognition of atrial fibrilation episodes in heart rate variability signals using a machine learning approach | |
Celler et al. | Blood pressure estimation using time domain features of auscultatory waveforms and GMM-HMM classification approach | |
Al Fahoum et al. | PPG signal-based classification of blood pressure stages using wavelet transformation and pre-trained deep learning models | |
CN113679369A (zh) | 一种心率变异性的评估方法、智能穿戴设备及存储介质 |
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: 18883741 Country of ref document: EP Kind code of ref document: A1 |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 18883741 Country of ref document: EP Kind code of ref document: A1 |