WO2018107915A1 - 一种基于时序分析的通用无创血糖预测方法 - Google Patents

一种基于时序分析的通用无创血糖预测方法 Download PDF

Info

Publication number
WO2018107915A1
WO2018107915A1 PCT/CN2017/108525 CN2017108525W WO2018107915A1 WO 2018107915 A1 WO2018107915 A1 WO 2018107915A1 CN 2017108525 W CN2017108525 W CN 2017108525W WO 2018107915 A1 WO2018107915 A1 WO 2018107915A1
Authority
WO
WIPO (PCT)
Prior art keywords
blood glucose
sequence
model
eigenvalue
invasive
Prior art date
Application number
PCT/CN2017/108525
Other languages
English (en)
French (fr)
Inventor
唐飞
耿占潇
王晓浩
丁亚东
范志伟
Original Assignee
博邦芳舟医疗科技（北京）有限公司
清华大学
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 博邦芳舟医疗科技（北京）有限公司, 清华大学 filed Critical 博邦芳舟医疗科技（北京）有限公司
Priority to US15/744,438 priority Critical patent/US10825569B2/en
Priority to EP17823009.0A priority patent/EP3358485A4/en
Publication of WO2018107915A1 publication Critical patent/WO2018107915A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/50—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for simulation or modelling of medical disorders
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H10/00—ICT specially adapted for the handling or processing of patient-related medical or healthcare data
    • G16H10/40—ICT specially adapted for the handling or processing of patient-related medical or healthcare data for data related to laboratory analysis, e.g. patient specimen analysis
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/70—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00—Details relating to CAD techniques
    • G06F2111/10—Numerical modelling
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
    • Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation

Definitions

  • the invention relates to non-invasive blood glucose detection for human body, and belongs to the field of non-invasive blood glucose test research, in particular to a general non-invasive blood glucose prediction method based on time series analysis.
  • Diabetes is a group of metabolic diseases characterized by hyperglycemia, and there is currently no cure for diabetes.
  • the treatment of diabetes is mainly to monitor and control blood sugar levels.
  • the traditional method of blood glucose measurement has obvious defects, which brings trauma and pain to the patient during the measurement process, and it is not convenient to achieve continuity detection.
  • Non-invasive blood glucose detection technology overcomes the shortcomings of traditional detection methods and can effectively meet the needs of diabetics to monitor blood glucose concentration in real time and frequently. It is the development direction of blood glucose detection technology. However, the accuracy of the current non-invasive blood glucose test is not sufficient.
  • the currently used non-invasive blood glucose prediction method uses the non-invasive measurement parameters at this moment to predict the blood glucose at that time.
  • the impedance spectrum characteristic of the test time is used to predict blood glucose.
  • Chinese patent CN105662434A uses the mid-infrared light characteristic at the test time to predict blood glucose
  • China Patent CN104490403A uses the spectral information of the test time to predict blood glucose.
  • a method for predicting blood glucose using potential time-series dynamics of blood glucose is disclosed in the Chinese patent document (CN103310113A), which utilizes a subcutaneous blood glucose level that has been tested by an invasive method for a certain period of time, and predicts blood glucose for a period of time thereafter. This is the use of the dynamic changes in the body's own blood sugar itself, through the invasive method to collect blood glucose values for a period of time to predict the blood sugar for a period of time.
  • the blood glucose is tested by testing the physiological parameters related to blood sugar levels, and there is a certain delay between the blood sugar changes and the physiological parameters of the human body.
  • the delay of different physiological parameters may be different, and the physiological at the current time cannot be simply used.
  • the parameter value is used to predict the blood glucose at the current time; while the historical blood glucose information is used to predict the blood glucose.
  • the time series of blood glucose is used, the historical blood sugar needs to be obtained through an invasive method, and the risk of trauma and infection caused by the invasive method cannot be overcome.
  • the object of the present invention is to overcome the deficiencies of the existing non-invasive blood glucose prediction methods, and to establish a blood glucose model by using a time series analysis method to overcome the delay between human physiological parameters and blood sugar level changes.
  • a general non-invasive blood glucose prediction method based on time series analysis characterized in that the method comprises the following steps:
  • t is the sampling point number
  • T is the number of eigenvalues calculated by the collected physiological parameters
  • Z is the sequence length; and the invasive method is used to obtain the measured blood glucose level sequence Glu(t), and the eigenvalue sequence and the measured blood glucose value sequence are obtained.
  • the time series analysis model is used to express the relationship between the sequence of relevant eigenvalues and the sequence of measured blood glucose values, and the single eigenvalue model and the single eigenvalue model blood glucose sequence are obtained.
  • Multi-feature value fusion each single eigenvalue model blood glucose sequence is fused using a weighted average model to obtain a multi-feature value model
  • a universal non-invasive blood glucose prediction method based on time series analysis is characterized in that: step 2) when eigenvalue screening, using a cross-correlation function to obtain the similarity between the eigenvalue sequence and the measured blood glucose value sequence, for the eigenvalue sequence x i (t), its correlation function with the measured blood glucose value sequence is:
  • N is the calculated cross-correlation sequence length
  • R( ⁇ ) is the cross-correlation function value
  • ⁇ is the independent variable of the cross-correlation function
  • a universal non-invasive blood glucose prediction method based on time series analysis is characterized in that: step 3) based on the single eigenvalue model of time series analysis, using the moving average model in the time series analysis method to express the relevant feature value sequence and the measured blood glucose
  • step 3 based on the single eigenvalue model of time series analysis, using the moving average model in the time series analysis method to express the relevant feature value sequence and the measured blood glucose
  • m is the model order, 0 ⁇ n ⁇ m, b jn is the model coefficient, and ⁇ j (t) is the residual;
  • the model coefficient b jn is obtained by the least squares method, and the intermediate variable g j (t) is obtained:
  • the delay time T j between the two is obtained by g j (t) and Glu(t), and the final single eigenvalue model blood glucose sequence G j (t) is obtained:
  • a universal non-invasive blood glucose prediction method based on time series analysis is characterized in that: step 4) when multi-feature value fusion is performed, the weighted average model is used for fusion, and the multi-eigenvalue model parameters K j , K j are obtained as G j (t) Corresponding weights, obtained by linear regression model as follows:
  • step 5 performs non-invasive blood glucose prediction, and the specific steps are as follows:
  • step 2) using the relevant feature value information obtained in step 2) of the modeling to extract the feature values, and obtain a subset of the feature values, the number of related feature values in the feature value subset is M, and the correlation feature value number is j;
  • the final blood glucose prediction sequence is obtained.
  • physiological parameters of non-invasive acquisition include infrared spectral characteristics, impedance characteristics, temperature, humidity, blood flow rate, blood oxygen saturation, pulse, sound speed, Sound impedance, photoacoustic characteristics.
  • a general non-invasive blood glucose prediction method based on time series analysis is characterized in that: the eigenvalue sequence and the measured blood glucose value sequence are normalized in the preprocessing process and then filtered by wavelet filtering.
  • the invention Compared with the prior art, the invention has the following advantages and outstanding effects: 1.
  • the general non-invasive blood glucose prediction method based on time series analysis proposed by the invention is simple and easy to use, and for each diabetic patient, only one test is required for about 3 hours.
  • the model can be established; 2 overcome the delay of changes in human physiological parameters and changes in blood glucose concentration, so that the blood glucose value measured by the non-invasive method is more accurate; 3 the method proposed by the invention is applicable to a variety of different non-invasive blood glucose testing methods, and has versatility .
  • Figure 1 is a block flow diagram of a general non-invasive blood glucose prediction method based on time series analysis.
  • Fig. 2 is an example of a non-invasive eigenvalue sequence and a measured blood glucose value sequence.
  • Figure 3 is an example of wavelet filtering results.
  • Figure 4 is an example of a single eigenvalue model result.
  • Figure 5 is a comparison of a single eigenvalue model and multiple eigenvalue fusion results.
  • Figure 6 is a schematic diagram of the data collection process.
  • Fig. 7 is an example of blood glucose prediction results.
  • Figure 8 is a multi-sensor non-invasive blood glucose detecting device probe based on an impedance spectrum-optical method.
  • 1-low frequency electrode 2-temperature humidity sensor; 3-high frequency electrode; 4-light-emitting diode array; 5-photosensor; 6-contact plate; 7-shield electrode; L is the matching inductance of high-frequency electrode.
  • a multi-sensor non-invasive blood glucose detecting device based on an impedance spectrum-optical method includes a temperature and humidity sensor 2, an LED array 4, a photosensor 5, a low frequency electrode 1 and a high frequency electrode 3.
  • the high-frequency electrode adopts a parallel electrode, and the matched inductor is directly soldered on the positive electrode or the negative electrode of the electrode, and the shield electrode 7 is provided.
  • the low-frequency electrode has a distance of 15cm, which can stably test the low-frequency impedance of the tissue.
  • the multi-sensor non-invasive blood glucose detecting device based on the impedance spectrum-optical method is used for data acquisition.
  • the data acquisition process is as shown in Fig. 6.
  • the specific steps are as follows:
  • the tester starts the test in the fasting state, and can not perform strenuous exercise 30 minutes before the test. It is best to sit in a comfortable position in a comfortable environment.
  • the tester wears a non-invasive test device and begins to continuously collect relevant physiological parameter information.
  • the non-invasive test equipment here can be different equipment based on different principles.
  • Non-invasive blood glucose testing equipment based on impedance spectroscopy and optical methods requires collection of low-frequency impedance, high-frequency impedance, temperature, humidity, and light transmittance of the tissue over time. Each parameter is calculated every 1 minute and stored in a file.
  • T, T are the number of eigenvalues
  • the non-invasive blood glucose test method T based on the impedance spectrum and the optical method takes 10
  • x i (t) is the time series of the i-th eigenvalue.
  • the input uses the conventional method to obtain the measured blood glucose level sequence Glu(t).
  • the eigenvalues and the measured blood glucose values are normalized.
  • the sequence of eigenvalues and the sequence of measured blood glucose values are interpolated to obtain a time-matched sequence of eigenvalues and a sequence of blood glucose values.
  • Figure 2 An example of normalization of each feature value sequence and measured blood glucose value sequence is shown in Figure 2, which shows the normalized results of the tissue resonance frequency, light transmission, temperature, humidity parameter sequence, and measured blood glucose value sequence.
  • Wavelet filtering In order to eliminate high frequency noise, the original time series is filtered by wavelet filtering. Wavelet filtering first selects the wavelet base.
  • the db8 wavelet can be used to perform six-layer decomposition on the original signal, and the first layer and the second layer are reconstructed to eliminate high-frequency noise.
  • the effect of wavelet filtering is shown in Figure 3.
  • N is the calculated cross-correlation sequence length
  • R( ⁇ ) is the cross-correlation function value
  • ⁇ is the independent variable of the cross-correlation function.
  • m is the model order, 0 ⁇ n ⁇ m, b jn is the model coefficient, and ⁇ j (t) is the residual;
  • the model coefficient b jn is obtained by the least squares method, and the intermediate variable g j (t) is obtained:
  • the delay time T j between the two is obtained by g j (t) and Glu(t), and the final single eigenvalue model blood glucose sequence G j (t) is obtained:
  • the cross-correlation function can be used to obtain the delay T j of g j (t) and Glu(t), and the cross-correlation function is as shown in equation (5):
  • N is the calculated cross-correlation sequence length
  • R( ⁇ ) is the cross-correlation function value
  • ⁇ is the independent variable of the cross-correlation function.
  • the value of ⁇ corresponding to the maximum value R max of the cross-correlation function is T j .
  • the model order can be set to 10, and the single eigenvalue blood glucose model results are shown in Fig. 4.
  • the relationship between the measured blood glucose level, the original eigenvalue and the single eigenvalue model blood glucose sequence is given.
  • the blood glucose is tested by testing the physiological parameters related to blood sugar levels, and there is a certain delay between the blood sugar changes and the physiological parameters of the human body.
  • the delay of different physiological parameters may be different, and the physiological at the current time cannot be simply used.
  • the parameter value is used to predict the blood glucose at the current time.
  • the method of time series analysis is used to establish a model to overcome the delay between physiological parameters and blood glucose levels.
  • Multi-feature value fusion Each single eigenvalue model blood glucose sequence is fused using a weighted average model to obtain multi-eigenvalue model parameters, and the weight K j corresponding to G j (t) is obtained by a linear regression model:
  • the multi-eigenvalue blood glucose model sequence is obtained by weighted averaging:
  • G(t) is a multi-eigenvalue blood glucose model sequence
  • the multi-eigenvalue blood glucose model sequence is better than the single eigenvalue model blood glucose sequence, and the information of each eigenvalue is integrated, and the prediction result is more stable, as shown in Fig. 5.
  • step 2) using the relevant feature value information obtained in step 2) of the modeling to extract the feature values, and obtain a subset of the feature values, the number of related feature values in the feature value subset is M, and the correlation feature value number is j;
  • the final blood glucose prediction sequence is obtained.
  • Fig. 7 The results of blood glucose prediction using the above method are shown in Fig. 7. Three tests were performed on one user, and one of the experiments was used to model and the other two experiments were predicted. The picture on the gray background represents the modeling result, and the white background is the prediction result. The dotted line in the figure is the model prediction result, which is realized as the measured blood glucose result. It can be seen that the results of predicting blood glucose using this method are more accurate.
  • the model can be established after one test, and the whole process is about three hours. If different test methods are replaced, only the obtained feature values are different, the modeling method does not need to be changed, and it is versatile.
  • Optical methods can be used to non-invasively acquire spectral features of the tissue, such as the mid-infrared band, the near-infrared band, and the visible band. The characteristic spectra associated with changes in blood glucose are extracted from the spectrum to obtain eigenvalues. Ultrasonic methods can be used to non-invasively collect the acoustic properties of the tissue, and the characteristics such as sound velocity and acoustic impedance can be extracted as characteristic values.

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Public Health (AREA)
  • Medical Informatics (AREA)
  • Primary Health Care (AREA)
  • General Health & Medical Sciences (AREA)
  • Epidemiology (AREA)
  • Data Mining & Analysis (AREA)
  • Biomedical Technology (AREA)
  • Databases & Information Systems (AREA)
  • Pathology (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Geometry (AREA)
  • Evolutionary Computation (AREA)
  • Computer Hardware Design (AREA)
  • Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)

Abstract

一种基于时序分析的通用无创血糖预测方法。方法包括数据输入和预处理、特征值筛选、基于时序分析的单特征值模型、多特征值融合、无创血糖预测5个步骤。无创血糖预测时输入新的无创测试数据，利用建模得到的相关特征值信息、单特征值模型和多特征值模型计算出血糖预测序列。该预测方法易于实施，可以克服人体生理参数变化和血糖变化之间的延迟，得到更准确的无创血糖测试结果。该预测方法适用于不同的无创血糖测试方法，具有通用性。

Description

一种基于时序分析的通用无创血糖预测方法 技术领域
本发明涉及对人体进行无创血糖检测，属于无创血糖测试研究领域，具体为一种基于时序分析的通用无创血糖预测方法。
背景技术
糖尿病是一组以高血糖为特征的代谢性疾病，目前还没有根治糖尿病的方法。糖尿病的治疗以频繁地监测、控制血糖水平为主。传统的有创取血测量血糖的方法存在明显缺陷，在测量过程中给患者带来创伤和痛觉，不便于实现连续性的检测。无创血糖检测技术克服了传统检测方法的缺点，能有效地满足糖尿病人实时、频繁监测血糖浓度的需求，是血糖检测技术发展的方向。但目前无创血糖测试的准确度还不满足需求。
目前使用的无创血糖预测方法采用该时刻的无创测量参数预测该时刻的血糖，如美国专利US20120101351A1中利用测试时刻的阻抗谱特性预测血糖，中国专利CN105662434A利用测试时刻的中红外光特性预测血糖，中国专利CN104490403A中利用测试时刻的光谱信息预测血糖。
中国专利文献(CN103310113A)中公开了利用血糖的潜在时序动态特性进行血糖预测的方法，其利用的是之前一段时间通过有创方法测试的皮下血糖值，预测之后一段时间的血糖。这属于利用人体血糖本身的动态变化特性，通过有创方法采集一段时间的血糖值来预测之后一段时间的血糖。
无创血糖测试时通过测试人体与血糖水平相关的生理参数来预测血糖，人体的血糖变化和生理参数变化之间有一定的延迟，不同的生理参数的延迟可能不同，不能简单的用当前时刻的生理参数值去预测当前时刻的血糖；而利用历史血糖信息去预测血糖，虽然使用了血糖的时间序列，但历史血糖需要通过有创的方法得到，无法克服有创方法带来的创伤和感染风险。
发明内容
本发明的目的是克服现有无创血糖预测方法的不足，利用时序分析的方法建立血糖模型，克服人体生理参数和血糖水平变化之间的延迟。
本发明的技术方案如下：
一种基于时序分析的通用无创血糖预测方法，其特征在于该方法包括如下步骤：
1)数据输入和预处理：通过无创方法连续采集人体的相关生理参数得到特征值序列xi(t)，i＝1，…，T，t＝1，…，Z，i为特征值编号，t为采样点编号，T为通过采集的生理参数计算得到的特征值个数，Z为序列长度；同时利用有创方法得到实测血糖值序列Glu(t)，将特征值序列和实测血糖值序列进行归一化处理；
2)特征值筛选：根据特征值序列和实测血糖值序列的相似性进行特征值筛选，筛选出与血糖变化相关度较大的特征值子集，记录筛选出的相关特征值信息；
3)基于时序分析的单特征值模型：采用时序分析模型表达相关特征值序列和实测血糖值序列之间的关系，得到单特征值模型和单特征值模型血糖序列；
4)多特征值融合：将各个单特征值模型血糖序列使用加权平均模型进行融合，得到多特征值模型；
5)利用建立的相关特征值信息、单特征值模型和多特征值模型进行无创血糖预测。
上述方案中的一种基于时序分析的通用无创血糖预测方法，其特征在于：步骤2)特征值筛选时，使用互相关函数得到特征值序列和实测血糖值序列的相似性，对于特征值序列xi(t)，其与实测血糖值序列的相关函数为：
Figure PCTCN2017108525-appb-000001
其中，N为设定计算的互相关序列长度，R(τ)为互相关函数值，τ是互相关函数的自变量；进行特征值筛选时，当R(τ)的最大值Rmax超过设定阈值，认为两个序列相似，把这个特征值作为相关特征值，加入到特征值子集，特征值子集中相关特征值个数为M，相关特征值编号为j。
上述方案中的一种基于时序分析的通用无创血糖预测方法，其特征在于：步骤3)基于时序分析的单特征值模型时，采用时序分析方法中的滑动平均模型表达相关特征值序列和实测血糖值序列之间的关系，如下式：
Figure PCTCN2017108525-appb-000002
其中：m为模型阶数，0≤n＜m，bjn为模型系数，εj(t)为残差；
利用最小二乘法得到模型系数bjn，从而得到中间变量gj(t)：
Figure PCTCN2017108525-appb-000003
由gj(t)和Glu(t)得到两者之间的延时Tj，得到最终的单特征值模型血糖序列Gj(t)：
Gj(t)＝gj(t-Tj)
上述方案中的一种基于时序分析的通用无创血糖预测方法，其特征在于：步骤4)多特征值融合时，使用加权平均模型进行融合，得到多特征值模型参数Kj，Kj为Gj(t)对应的权重，通过线性回归模型得到如下式：
Figure PCTCN2017108525-appb-000004
其中ε(t)为拟合残差。
上述方案中的一种基于时序分析的通用无创血糖预测方法，其特征在于步骤5)进行无创血糖预测时，具体步骤为：
1)通过无创方法连续采集人体的相关生理参数重新得到特征值序列xi(t)，对特征值序列进行预处理；
2)利用建模中步骤2)得到的相关特征值信息进行特征值提取，得到特征值子集，特征值子集中相关特征值个数为M，相关特征值编号为j；
3)根据建模中步骤3)得到的单特征模型模型参数m、bjn和Tj，进行单特征值预测，得到gj(t)和Gj(t)，其中
Figure PCTCN2017108525-appb-000005
Gj(t)＝gj(t-Tj)；
4)根据建模中步骤4)得到的多特征值模型参数Kj，得到最终血糖预测序列
Figure PCTCN2017108525-appb-000006
上述方案中的一种基于时序分析的通用无创血糖预测方法，其特征在于：无创采集的生理参数包括红外光谱特性、阻抗特性、温度、湿度、血流速、血氧饱和度、脉搏、声速、声阻抗、光声谱特性。
上述方案中的一种基于时序分析的通用无创血糖预测方法，其特征在于：特征值序列和实测血糖值序列在预处理过程中归一化后采用小波滤波进行滤波。
本发明与现有技术相比，具有以下优点及突出性效果：①本发明提出的基于时序分析的通用无创血糖预测方法简单易用，对于每位糖尿病患者，只需要通过一次约3小时的测试即可建立模型；②克服了人体生理参数变化和血糖浓度变化的延迟，使无创方法测得的血糖值更准确；③本发明提出的方法适用于多种不同的无创血糖测试方法，具有通用性。
附图说明：
图1是基于时序分析的通用无创血糖预测方法的流程框图。
图2是无创得到的特征值序列和实测血糖值序列示例。
图3是小波滤波结果示例。
图4是单特征值模型结果示例。
图5是单特征值模型和多特征值融合结果对比。
图6是数据采集流程示意图。
图7是血糖预测结果示例。
图8是基于阻抗谱-光学方法的多传感器无创血糖检测设备探头。
图中：1-低频电极；2-温湿度传感器；3-高频电极；4-发光二极管阵列；5-光电传感器；6-接触板；7-屏蔽电极；L为高频电极的匹配电感。
具体实施方式
下面结合附图对本发明提出的一种基于时序分析的通用无创血糖预测方法具体过程做进一步的说明。
1、第1实施例：
下面结合基于阻抗谱-光学方法的多传感器无创血糖检测设备对该种一种基于时序分析的通用无创血糖预测方法具体过程做进一步的说明。
基于阻抗谱-光学方法的多传感器无创血糖检测设备，检测探头如附图8，包括温湿度传感器2、发光二极管阵列4、光电传感器5、低频电极1和高频电极3。高频电极采用平行电极，电极的正极或负极上直接焊接匹配的电感，并设有屏蔽电极7。低频电极的距离位15cm，能稳定测试组织低频阻抗。
首先要使用基于阻抗谱-光学方法的多传感器无创血糖检测设备进行数据采集，数据采集的过程如附图6，具体步骤为：
1)测试者空腹状态开始测试，测试前30分钟不能进行剧烈运动，最好是在舒适的环境中以舒适的姿势静坐。
2)测试者佩戴无创测试设备，开始连续采集相关生理参数信息。这里的无创测试设备可以是基于不同原理的不同设备。
3)测试者佩戴无创设备20分钟后开始进餐，进餐在15分钟内完成；进餐的量最好能够控制。
4)餐后至少连续采集数据140分钟。最好选择采集餐后3个小时的数据，这样可以得到餐后血糖升高到降低的整个过程。每次实验采集的数据长度不需要完全一致。
在采用无创设备进行数据采集的同时，利用有创方法得到标准血糖值，可以每隔30 分钟采集一次指尖血数据，得到实测血糖值序列Glu(t)。
基于阻抗谱和光学方法的无创血糖测试设备，需要采集组织的低频阻抗、高频阻抗、温度、湿度、透光性等特征值随时间的变化。各个参数每隔1分钟计算一次，并存储在文件中。
采集数据之后，根据采集的数据进行建模，具体步骤如下：
1)数据输入和预处理。通过无创方法连续采集人体的相关生理参数，包括低频阻抗、高频阻抗、谐振频率、温度、湿度、组织透光性，从生理参数中计算得到特征值序列xi(t)，i＝1，…，T，T为特征值个数，针对基于阻抗谱和光学方法的无创血糖测试方法T取10，xi(t)为第i个特征值的时间序列。输入利用传统方法得到实测血糖值序列Glu(t)。将特征值和实测血糖值进行归一化处理。并对特征值序列和实测血糖值序列进行插值，得到时间匹配的特征值序列与血糖值序列。
各特征值序列和实测血糖值序列归一化的示例如附图2，图中列出了组织谐振频率、透光性、温度、湿度参数序列和实测血糖值序列的归一化结果。
为了消除高频噪声用小波滤波对原始时间序列进行滤波。小波滤波首先选定小波基，这里可以采用db8小波，对原始信号做六层分解，取第一层和第二层信号进行重构，消除高频噪声。小波滤波的效果如附图3。
2)特征值筛选：根据特征值序列和实测血糖值序列的相似性进行特征值筛选，筛选出与血糖变化相关度较大的特征值子集，记录筛选出的相关特征值信息；使用互相关函数得到特征值序列和实测血糖值序列的相似性，对于特征值序列xi(t)，其与实测血糖值序列的相关函数为：
Figure PCTCN2017108525-appb-000007
其中N为设定计算的互相关序列长度，R(τ)为互相关函数值，τ是互相关函数的自变量。进行特征值筛选时，当特征值与实测血糖值序列的互相关函数的最大值Rmax超过设定阈值，认为两个信号相似，把这个特征值作为相关特征值，加入到特征值子集，特征值子集中相关特征值个数为M，相关特征值编号为j。
3)基于时序分析的单特征值模型：采用时序分析方法中的滑动平均模型表达相关特征值序列和实测血糖值序列之间的关系，如下式：
Figure PCTCN2017108525-appb-000008
其中：m为模型阶数，0≤n＜m，bjn为模型系数，εj(t)为残差；
利用最小二乘法得到模型系数bjn，从而得到中间变量gj(t)：
Figure PCTCN2017108525-appb-000009
由gj(t)和Glu(t)得到两者之间的延时Tj，得到最终的单特征值模型血糖序列Gj(t)：
Gj(t)＝gj(t-Tj)     (4)
可以使用互相关函数得到gj(t)和Glu(t)的延时Tj，互相关函数如式(5)：
Figure PCTCN2017108525-appb-000010
其中N为设定计算的互相关序列长度，R(τ)为互相关函数值，τ是互相关函数的自变量。互相关函数的最大值Rmax对应的τ值就是Tj。
这里可以设置模型阶数为10，单特征值血糖模型结果如图4，其中给出了实测血糖值、原始特征值和单特征值模型血糖序列的关系。
无创血糖测试时通过测试人体与血糖水平相关的生理参数来预测血糖，人体的血糖变化和生理参数变化之间有一定的延迟，不同的生理参数的延迟可能不同，不能简单的用当前时刻的生理参数值去预测当前时刻的血糖。这里采用时序分析的方法建立模型可以克服生理参数和血糖值之间的延迟。
4)多特征值融合：将各个单特征值模型血糖序列使用加权平均模型进行融合，得到多特征值模型参数，Gj(t)对应的权重Kj通过线性回归模型得到：
Figure PCTCN2017108525-appb-000011
其中ε(t)为拟合残差；
得到Kj后，通过加权平均得到多特征值血糖模型序列：
Figure PCTCN2017108525-appb-000012
其中，G(t)是多特征值血糖模型序列；
多特征值血糖模型序列的会比单特征值模型血糖序列要好，综合了各个特征值的信息，预测结果也更稳定，如图5。
5)利用建立的模型进行无创血糖预测：
1)通过无创方法连续采集人体的相关生理参数得到新的特征值序列xi(t)，对特征值序列进行预处理；
2)利用建模中步骤2)得到的相关特征值信息进行特征值提取，得到特征值子集，特征值子集中相关特征值个数为M，相关特征值编号为j；
3)根据建模中步骤3)得到的单特征模型模型参数m、bjn和Tj，进行单特征值预测，得到gj(t)和Gj(t)，其中
Figure PCTCN2017108525-appb-000013
Gj(t)＝gj(t-Tj)；
4)根据建模中步骤4)得到的多特征值模型参数Kj，得到最终血糖预测序列
Figure PCTCN2017108525-appb-000014
利用上述方法进行血糖预测的结果见图7。对1个使用者进行3次测试，利用其中一次实验进行建模，对其余两次实验进行预测。其中灰色背景的图片表示建模结果，白色背景是预测结果。图中虚线为模型预测结果，实现为实测血糖结果。可以看出使用该方法预测血糖的结果较准确。
从上述过程可以发现，经过一次测试就可以建立模型，整个过程大约3个小时。如果更换不同的测试方法，只是得到的特征值不同，建模方法不需要改变，具有通用性。可以采用光学方法无创采集组织的光谱特征，如中红外波段、近红外波段、可见光波段等。从光谱提取与血糖变化相关的特征光谱，得到特征值。可以采用超声波方法无创采集组织的声学特性，从中提取出声速、声阻抗等特性作为特征值。

Claims (7)

  1. 一种基于时序分析的通用无创血糖预测方法，其特征在于该方法包括如下步骤：
    1)数据输入和预处理：通过无创方法连续采集人体的相关生理参数得到特征值序列xi(t)，i＝1，…，T，t＝1，…，Z，i为特征值编号，t为采样点编号，T为通过采集的生理参数计算得到的特征值个数，Z为序列长度；同时利用有创方法得到实测血糖值序列Glu(t)，将特征值序列和实测血糖值序列进行归一化处理；
    2)特征值筛选：根据特征值序列和实测血糖值序列的相似性进行特征值筛选，筛选出与血糖变化相关度较大的特征值子集，记录筛选出的相关特征值信息；
    3)基于时序分析的单特征值模型：采用时序分析模型表达相关特征值序列和实测血糖值序列之间的关系，得到单特征值模型和单特征值模型血糖序列；
    4)多特征值融合：将各个单特征值模型血糖序列使用加权平均模型进行融合，得到多特征值模型；
    5)利用建立的相关特征值信息、单特征值模型和多特征值模型进行无创血糖预测。
  2. 如权利要求1所述的一种基于时序分析的通用无创血糖预测方法，其特征在于：步骤2)特征值筛选时，使用互相关函数得到特征值序列和实测血糖值序列的相似性，对于特征值序列xi(t)，其与实测血糖值序列的相关函数为：
    Figure PCTCN2017108525-appb-100001
    其中，N为设定计算的互相关序列长度，R(τ)为互相关函数值，τ是互相关函数的自变量；进行特征值筛选时，当R(τ)的最大值Rmax超过设定阈值，认为两个序列相似，把这个特征值作为相关特征值，加入到特征值子集，特征值子集中相关特征值个数为M，相关特征值编号为j。
  3. 如权利要求1所述的一种基于时序分析的通用无创血糖预测方法，其特征在于：步骤3)基于时序分析的单特征值模型时，采用时序分析方法中的滑动平均模型表达相关特征值序列和实测血糖值序列之间的关系，如下式：
    Figure PCTCN2017108525-appb-100002
    其中：m为模型阶数，0≤n＜m，bjn为模型系数，εj(t)为残差；
    利用最小二乘法得到模型系数bjn，从而得到中间变量gj(t)：
    Figure PCTCN2017108525-appb-100003
    由gj(t)和Glu(t)得到两者之间的延时Tj，得到最终的单特征值模型血糖序列Gj(t)：
    Gj(t)＝gj(t-Tj)
  4. 如权利要求1所述的一种基于时序分析的通用无创血糖预测方法，其特征在于：步骤4)多特征值融合时，使用加权平均模型进行融合，得到多特征值模型参数Kj，Kj为Gj(t)对应的权重，通过线性回归模型得到如下式：
    Figure PCTCN2017108525-appb-100004
    其中ε(t)为拟合残差。
  5. 如权利要求1所述的一种基于时序分析的通用无创血糖预测方法，其特征在于步骤5)进行无创血糖预测时，具体步骤为：
    1)通过无创方法连续采集人体的相关生理参数重新得到特征值序列xi(t)，对特征值序列进行预处理；
    2)利用建模中步骤2)得到的相关特征值信息进行特征值提取，得到特征值子集，特征值子集中相关特征值个数为M，相关特征值编号为j；
    3)根据建模中步骤3)得到的单特征模型模型参数m、bjn和Tj，进行单特征值预测，得到gj(t)和Gj(t)，其中
    Figure PCTCN2017108525-appb-100005
    4)根据建模中步骤4)得到的多特征值模型参数Kj，得到最终血糖预测序列
    Figure PCTCN2017108525-appb-100006
  6. 如权利要求1所述的一种基于时序分析的通用无创血糖预测方法，其特征在于：无创采集的生理参数包括红外光谱特性、阻抗特性、温度、湿度、血流速、血氧饱和度、脉搏、声速、声阻抗、光声谱特性。
  7. 如权利要求1所述的一种基于时序分析的通用无创血糖预测方法，其特征在于：特征值序列和实测血糖值序列在预处理过程中归一化后采用小波滤波进行滤波。
PCT/CN2017/108525 2016-12-16 2017-10-31 一种基于时序分析的通用无创血糖预测方法 WO2018107915A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
US15/744,438 US10825569B2 (en) 2016-12-16 2017-10-31 Universal non-invasive blood glucose estimation method based on time series analysis
EP17823009.0A EP3358485A4 (en) 2016-12-16 2017-10-31 GENERAL METHOD FOR PREDICTING NON-EFFRACTIVE GLYCEMIA BASED ON TIME ANALYSIS

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201611170162.5A CN106980746B (zh) 2016-12-16 2016-12-16 一种基于时序分析的通用无创血糖预测方法
CN201611170162.5 2016-12-16

Publications (1)

Publication Number Publication Date
WO2018107915A1 true WO2018107915A1 (zh) 2018-06-21

Family

ID=59340484

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/108525 WO2018107915A1 (zh) 2016-12-16 2017-10-31 一种基于时序分析的通用无创血糖预测方法

Country Status (4)

Country Link
US (1) US10825569B2 (zh)
EP (1) EP3358485A4 (zh)
CN (1) CN106980746B (zh)
WO (1) WO2018107915A1 (zh)

Families Citing this family (25)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106980746B (zh) * 2016-12-16 2021-01-26 清华大学 一种基于时序分析的通用无创血糖预测方法
CN107563426B (zh) * 2017-08-25 2020-05-22 清华大学 一种机车运行时序特征的学习方法
CN109712715A (zh) * 2017-10-25 2019-05-03 普天信息技术有限公司 一种生理数据的评分处理方法及装置
CN109934370B (zh) * 2017-12-15 2024-03-05 中国农业大学 一种鱼类存活率预测方法及装置
CN109276258B (zh) * 2018-08-10 2021-08-03 北京大学深圳研究生院 基于dtw的血糖趋势预测方法、系统及医疗设备
CN109065176B (zh) * 2018-08-20 2020-11-10 清华-伯克利深圳学院筹备办公室 一种血糖预测方法、装置、终端和存储介质
CN109448855A (zh) * 2018-09-17 2019-03-08 大连大学 一种基于cnn和模型融合的糖尿病血糖预测方法
CN111110949B (zh) * 2018-11-01 2021-07-09 上海市第六人民医院 胰岛素注射量确定方法及装置、计算机存储介质及设备
CN109872821A (zh) * 2019-04-24 2019-06-11 北京信息科技大学 一种构建血糖预测模型的方法及装置
CN110164553B (zh) * 2019-05-31 2023-03-03 东北大学 一种基于vmd-lssvm模型的在线动态血糖预测方法
BR102020003529A2 (pt) * 2020-02-20 2021-08-31 Nilton Braz Giraldelli Sistema para interpretação da medição de bioanalitos
CN110942831B (zh) * 2019-11-26 2022-10-11 北京航空航天大学 基于自注意力机制的孕期数据建模方法
CN111150395A (zh) * 2020-01-15 2020-05-15 桂林电子科技大学 一种基于生物电阻抗的自适应血糖测量方法
CN111599470B (zh) * 2020-04-23 2022-07-29 中国科学院上海技术物理研究所 一种用于提高近红外无创血糖检测精度的方法
CN112635054A (zh) * 2020-11-30 2021-04-09 新绎健康科技有限公司 一种基于脉图参数对目标血糖值进行预测的系统及方法
WO2022261917A1 (zh) * 2021-06-17 2022-12-22 广州视源电子科技股份有限公司 可实现无创血糖检测的装置及设备
CN113729698B (zh) * 2021-08-26 2023-07-28 朱斌 一种无创血糖检测方法及系统
CN114239658B (zh) * 2021-12-20 2024-05-10 桂林电子科技大学 一种基于小波分解与gru神经网络的血糖预测方法
CN114093521B (zh) * 2022-01-20 2022-04-12 广东工业大学 基于随机森林的均匀化样本重构血糖估计方法及系统
CN114548158B (zh) * 2022-01-28 2022-11-11 广东工业大学 一种用于血糖预测的数据处理方法
CN115969366A (zh) * 2023-03-05 2023-04-18 北京大学第三医院（北京大学第三临床医学院） 基于近红外吸收光谱-阻抗谱分析联用的血糖测量方法
CN117770806A (zh) * 2023-12-13 2024-03-29 苏州泽声微电子有限公司 血糖监测方法
CN117530684B (zh) * 2024-01-09 2024-04-16 深圳市双佳医疗科技有限公司 一种基于健康大数据的血糖异常检测与预警系统及方法
CN117582222B (zh) * 2024-01-18 2024-03-29 吉林大学 信息化血糖监测系统及方法
CN117612737B (zh) * 2024-01-24 2024-05-03 胜利油田中心医院 一种糖尿病护理数据智能优化方法

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120101351A1 (en) 2009-04-17 2012-04-26 Andreas Caduff Wide band field response measurement for glucose determination
CN103310113A (zh) 2013-06-24 2013-09-18 浙江大学 一种基于频带分离和数据建模的通用血糖预测方法
CN104490403A (zh) 2014-12-06 2015-04-08 深圳市前海安测信息技术有限公司 基于光谱技术的无创血糖测量系统及其测量方法
CN105662434A (zh) 2016-04-12 2016-06-15 北京科宇佳科技有限公司 一种中红外无创血糖检测设备
CN105962949A (zh) * 2016-06-14 2016-09-28 上海理工大学 一种基于近红外光能量守恒法的无创血糖计算方法以及信号采集装置
CN106980746A (zh) * 2016-12-16 2017-07-25 清华大学 一种基于时序分析的通用无创血糖预测方法

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
SE0400456D0 (sv) * 2004-02-26 2004-02-26 Lars Liljeryd Multiparameter metabolic monitoring, a method and device for the improvement of management and control in borderline or manifest type 2-diabetes
EP2152895A2 (en) * 2007-05-11 2010-02-17 Sigmed, Inc. Non-invasive characterization of a physiological parameter
US20110160555A1 (en) * 2008-07-31 2011-06-30 Jacques Reifman Universal Models for Predicting Glucose Concentration in Humans
EP2544124A1 (en) * 2011-07-04 2013-01-09 Sabirmedical, S.L. Methods and systems for non-invasive measurement of glucose levels
WO2016054079A1 (en) * 2014-09-29 2016-04-07 Zyomed Corp. Systems and methods for blood glucose and other analyte detection and measurement using collision computing

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120101351A1 (en) 2009-04-17 2012-04-26 Andreas Caduff Wide band field response measurement for glucose determination
CN103310113A (zh) 2013-06-24 2013-09-18 浙江大学 一种基于频带分离和数据建模的通用血糖预测方法
CN104490403A (zh) 2014-12-06 2015-04-08 深圳市前海安测信息技术有限公司 基于光谱技术的无创血糖测量系统及其测量方法
CN105662434A (zh) 2016-04-12 2016-06-15 北京科宇佳科技有限公司 一种中红外无创血糖检测设备
CN105962949A (zh) * 2016-06-14 2016-09-28 上海理工大学 一种基于近红外光能量守恒法的无创血糖计算方法以及信号采集装置
CN106980746A (zh) * 2016-12-16 2017-07-25 清华大学 一种基于时序分析的通用无创血糖预测方法

Also Published As

Publication number Publication date
US20190295729A1 (en) 2019-09-26
EP3358485A4 (en) 2019-03-06
CN106980746B (zh) 2021-01-26
CN106980746A (zh) 2017-07-25
US10825569B2 (en) 2020-11-03
EP3358485A1 (en) 2018-08-08

Similar Documents

Publication Publication Date Title
WO2018107915A1 (zh) 一种基于时序分析的通用无创血糖预测方法
CN109758160B (zh) 一种基于lstm-rnn模型的无创血糖预测方法
CN101357062B (zh) 基于容积脉搏波检测的生命体征参数测评装置
US20100022895A1 (en) Diagnosis system of deficient and forceful pulse
CN110575181A (zh) 近红外光谱无创血糖检测网络模型训练方法
CN110420019A (zh) 一种心冲击图信号的深度回归心率估计方法
RU2010146145A (ru) Неинвазивный способ и устройство для определения стадий поверхностного и глубокого сна
WO2006010310A1 (fr) Appareil et technique de surveillance non invasive de composants sanguins via des spectres d'impedance de forme d'impulsion
CN103263271A (zh) 非接触式自动血氧饱和度测量系统及测量方法
CN108542402B (zh) 基于自组织竞争神经网络模型和红外光谱的血糖检测方法
CN106264499A (zh) 一种量化心肺系统交互作用的分析方法
CN111481190A (zh) 基于双路ppg信号的稳健心率测量方法和装置
CN113662538A (zh) 一种基于时频域综合分析的无创血糖检测方法
CN110313900A (zh) 一种基于深度学习对膝关节摆动信号的分类方法
CN112294282A (zh) 基于rppg的情绪检测装置的自标定方法
CN109523545A (zh) 一种非接触式心率检测方法及系统
CN113576475B (zh) 一种基于深度学习的无接触血糖测量方法
WO2022063047A1 (zh) 一种基于光电容积脉搏波的无创糖尿病预测系统及方法
Lu et al. Pulse feature extraction based on improved gaussian model
CN113069091A (zh) 一种ppg信号的脉象分类装置及方法
CN105167759A (zh) 一种基于智能手机的人体脉搏波速度测量方法及其系统
CN106175698B (zh) 睡眠状态分析中睡眠周期检测装置
CN109036552A (zh) 中医诊断终端及其存储介质
CN114145725B (zh) 一种基于无创连续血压测量的ppg采样率估算方法
Lu et al. Pulse waveform analysis for pregnancy diagnosis based on machine learning

Legal Events

Date Code Title Description
WWE Wipo information: entry into national phase

Ref document number: 2017823009

Country of ref document: EP

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

Ref document number: 17823009

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE