CN111063435B - Diabetes typing diagnosis system - Google Patents

Abstract

The invention provides a diabetes typing diagnosis system, which collects the glucose measurement value of a participant and carries out sectional processing to obtain a trend fluctuation function, and completes the diagnosis of diabetes according to the trend fluctuation function. The system can overcome the current situation that the measured instantaneous blood sugar fluctuation data information is redundant and is caused by various interference factors, the fluctuation rule of the dynamic blood sugar change process of different types of diabetes mellitus is revealed by using a data analysis technology, and a new index, namely a trend-removing fluctuation function F, based on instantaneous blood sugar monitoring data and used for evaluating the internal insulin generation of a patient and assisting the diabetes mellitus classification is preliminarily established _d (l) To aid in the diagnosis of diabetes and to reveal the nature of diabetes.

Description

A Diabetes Type Diagnosis System

technical field

The invention relates to a diabetes classification and diagnosis system, which belongs to the technical field of medical signal processing.

Background technique

Diabetes is a metabolic disease characterized by hyperglycemia. Chronic hyperglycemia will lead to microvascular complications such as diabetic retinopathy, diabetic nephropathy, and diabetic neuropathy. It is collectively referred to as the four major chronic diseases along with respiratory diseases, cardiovascular diseases and tumors. non-communicable diseases.

Human blood glucose homeostasis is mainly regulated by insulin and glucagon. Among them, glucagon is secreted by α cells in the pancreas, which can increase the blood glucose concentration, and insulin is secreted by the β cells in the pancreatic islet cells, which can reduce the blood glucose concentration. At present, the diagnosis of diabetes is mainly based on islet β-cell function. Diabetes is divided into type 1 diabetes and type 2 diabetes. The etiology of type 1 diabetes focuses on the destruction of islet β cells by the autoimmune system, resulting in an absolute lack of insulin secretion. Type 2 diabetes focuses more on insulin resistance. On the basis of relatively insufficient insulin secretion.

The accurate classification of diabetes is of great significance for individualized treatment and is the direct basis for guiding the formulation of treatment plans. The current diabetes classification method is limited to clinical manifestations and observation of the dynamic changes of pancreatic β-cell function with the course of the disease. It relies on the individualized experience of clinicians based on practice and lacks a widely recognized differential diagnosis process. Type homogenization. Typically, β-cell function is achieved by serum C-peptide measurement, which is produced in the same amount as insulin and is considered the best representation of endogenous insulin secretion. Numerous studies have reported the diagnostic performance of C-peptide for type 1 and type 2 diabetes. Literature [1] (Torn C L-OM, Schersten B. Predictability of C-peptide for autoimmune diabetes in young adult diabetic patients. Pract Diabetes Int. 2001; 18:83-88.), Literature [2] (Thunander M, Torn C, Petersson C, Ossiansson B, Fornander J, Landin-Olsson M. Levels of C-peptide, BMI, and age, and their utility for classification of diabetes in relation to autoimmunity, in adults with newly diagnosed diabetes in Kronoberg, Sweden. Eur J Endocrinol. 2012; 166: 1021-1029.), literature [3] (Ludvigsson J, Carlsson A, Forsander G, Ivarsson S, Kockum I, Lernmark A, etal. C-peptide in the classification of diabetes in children and adolescents. Pediatr Diabetes. 2012;13:45-50.). However, due to the lack of standardized detection methods for serum C-peptide levels, there are considerable problems in their application. At the same time, serum C-peptide levels correlate with blood glucose levels, which makes quantitative evaluation difficult to achieve. Literature [4] (Little RR, Wielgosz RI, Josephs R, Kinumi T, Takatsu A, LiH, Stein D, Burns C. Implementing a Reference Measurement System for C-Peptide: Successes and Lessons Learned. Clin Chem. 2017 Sep; 63 (9):1447-1456.).

Elevated blood sugar level is the most intuitive manifestation of diabetes. Studying the fluctuation law of ambulatory blood sugar changes in different types of diabetes is helpful to reveal the essence of diabetes. Thanks to the development of blood glucose monitoring devices, the use of continuous glucose monitoring (CGM) and instantaneous glucose monitoring (FGM) has grown rapidly in the past few years. With the application of machine learning, data-driven blood glucose pattern classification and anomaly detection have also been attempted in type 1 diabetes and insulin pump closed-loop development. Reference [5] (Hall H, Perelman D, Breschi A, Limcaoco P, Kellogg R, McLaughlin T, Snyder M. Glucotypes reveal new patterns of glucose dysregulation. PLoS Biol. 2018 Jul 24;16(7):e2005143.).

FGM data show that the dynamics of blood glucose contain a lot of implicit information. It is critical to establish a link between endogenous etiology and information on glycemic fluctuations. To date, no study has used glycemic dynamic curve-based indicators to classify diabetes.

In the present invention, we use the detrended fluctuation function to describe the characteristics of blood glucose fluctuations, and establish the intrinsic relationship between the calculated index and islet function while further guiding the classification of diabetes. This type of function is derived from Detrended Volatility Analysis (DFA), a method commonly used to analyze long-range correlations in time series. Literature [6] (Peng C K, Mietus J, Hausdorff J M, et al. Long-range anticorrelations and non-Gaussian behavior of the heart beat. Phys Rev Lett, 1993, 70(9): 1343-1346.), Literature [7 ] (Peng C K, Buldyrev S V, Havlin S, et al. Mosaic organization of DNA nucleotides. Phys Rev E, 1994, 49(2): 1685-1689.), literature [8] (Peng CK, Havlin S, Stanley HE , Goldberger AL. Quantification of scaling exponents and crossover phenomena in nonstationaryheartbeat time series. Chaos 1995;5(1):82–87.).

Here, we established new metrics for predicting beta cell function and diabetes classification based on monitoring instantaneous glucose data, analyzed using a detrended fluctuation function. This could potentially pave the way for making further use of the vast amounts of data provided by blood glucose monitoring devices and studying digital precision drug delivery in diabetes research.

SUMMARY OF THE INVENTION

In view of this, the present invention provides a type of diabetes diagnosis system, which can overcome the current situation of complex information of measured instantaneous blood glucose fluctuation data caused by various interference factors, and use data analysis technology to reveal the dynamic blood glucose change process of different types of diabetes. Based on the fluctuation law, a new index based on instantaneous blood glucose monitoring data to evaluate the intrinsic insulin production of patients and assist in the classification of diabetes, namely the detrended fluctuation function F _d (l), is initially established to assist in the diagnosis of diabetes and reveal the essence of diabetes.

The present invention is achieved through the following technical solutions:

A diabetes classification and diagnosis system, wherein the diagnosis system collects the measured values of the participant's glucose and performs segmentation processing to obtain a trend fluctuation function, and completes the type diagnosis of diabetes according to the trend fluctuation function.

Further, the diagnosis system of the present invention includes a data acquisition and preprocessing module, a full data processing module, a segmented data processing module, a trend fluctuation function calculation module and a diabetes diagnosis module;

The data acquisition and preprocessing module collects the glucose measurements of the participants to obtain _a glucose time series G(k) of length N1, and divides the entire sequence into n segments of length l

再计算去除均值后的序列

然后计算求和序列Y(k)为：Full data processing module, calculate the mean of the sequence G(k)

Recalculate the series after removing the mean

Then calculate the summation sequence Y(k) as:

Segment data processing module, calculate the data value of each segment

Among them, i=1,2,...,n, k=(i-1)l+1,(i-1)l+2,...,il;

ω _i =[ω _0,i ,ω _1,i ] ^T

ω _i =(X _i ^T X _i ) ^-1 X _i ^T Y _i

where the coefficients k _j,i =(i-1)l+j, j=1,2...l, Yi is a column vector containing the data point information in the _ith segment of Y(k) ;

整合为Y_t(k)，计算去趋势序列Y_l(k)；Trend fluctuation function calculation module, which calculates the data value of each segment

Integrate as Y _t (k), calculate the detrended series Y _l (k);

Y _l (k)=Y(k)-Y _t (k)

The detrended fluctuation function F _d (l) is defined as:

The diabetes diagnosis module compares F _d (1) with the set threshold, and when it is less than or equal to the set threshold, it is determined as type 2 diabetes, and when it is greater than the set threshold, it is determined as type 1 diabetes.

Further, in the present invention, l≥10, preferably l=34.

Further, the threshold value of the present invention is 0.7.

beneficial effect

The present invention utilizes data analysis technology and combines the idea of de-trending fluctuation analysis to propose a diagnosis system for the condition of diabetic patients, and the specific effects are as follows:

(1) The most prominent advantage of this system is that it can reduce the influence of various external factors on the data, which is convenient for revealing the inherent fluctuation characteristics of the data, mining effective information on blood sugar fluctuations, and reflecting the endogenous cause of diabetes.

(2) The detrended fluctuation function of this system is a quantitative index obtained by data analysis method, which provides ideas for in-depth use of a large amount of data provided by blood glucose monitoring equipment and research on digital accurate drug delivery in diabetes research. To a certain extent, it can help Aid in the diagnosis of diabetes.

Description of drawings

Fig. 1 is the sociodemographic and clinical characteristics of the participants in the examples of the present invention;

Fig. 2 is the Spearman correlation coefficient variation diagram with fasting C-peptide when the fluctuation function takes different 1 values in the example of the present invention;

Fig. 3 is the wave function distribution histogram calculated according to the FGM information of all participants in the example of the present invention, wherein the red curve is obtained by fitting a bimodal Gaussian mixture model;

Fig. 4 is the Spearman correlation coefficient comparison table of detrended fluctuation function, MAGE, SD, Mean BG, TIR and fasting C-peptide in the example of the present invention;

Fig. 5 is the fluctuation function distribution diagram of the type 1 diabetes group and the type 2 diabetes group in the example of the present invention;

FIG. 6 is a receiver operating characteristic curve diagram for evaluating the effect of wave function typing in an example of the present invention.

Detailed ways

To make the purposes, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention.

The diagnosis system of the embodiment of the present invention needs to perform diagnosis based on the glucose measurement values of the participants. Therefore, the measurement and collection of data in conjunction with the diagnosis system is realized by the FGM equipment. The FGM equipment measures the blood sugar of each participating patient. The authors provide transient monitoring equipment including a glucose oxidase-based electrochemical sensor and a receiver. The sensor was placed under the skin, and participants were required to wear it daily, with the device relaying glucose measurements at that moment to the receiver every 15 minutes, and changing it every 14 days.

According to an embodiment of the present invention, the diabetes classification and diagnosis system, wherein the diagnosis system collects the measured value of the participant's glucose and performs segmentation processing to obtain a trend fluctuation function, and completes the diagnosis of diabetes according to the trend fluctuation function.

The diagnosis system of the present invention includes a data acquisition and preprocessing module, a full data processing module, a segmented data processing module, a trend fluctuation function calculation module and a diabetes diagnosis module;

The data acquisition and preprocessing module collects the glucose measurement values of the participants and performs preprocessing to obtain _a glucose time series G(k) of length N1, and divides the entire sequence into n segments of length l,

为向下取整符。Specifically: after the data collection performed by this module, _a glucose time series G(k) of length N1 is obtained, and the sampling period of the sequence is T:=15 [min]; then preprocessing is performed to divide the entire sequence into equal parts. Multiple fragments of length l, that is, each fragment contains l data points, the actual sequence length used in the analysis is

is the round-down character.

再计算去除均值后的序列，然后计算求和序列Y(k)为：Full data processing module, calculate the mean of the sequence G(k)

Then calculate the sequence after removing the mean, and then calculate the summation sequence Y(k) as:

The mean-removed sequence satisfies

The segmented data processing module divides Y(k) into n=N/l segments, and each segment contains 1 glucose data points.

Calculate trend data values for each segment

Among them, i=1,2,...,n, k=(i-1)l+1,(i-1)l+2,...,il;

Define ω _i =[ω _0,i ,ω _1,i ] ^T , the value of ω _i is calculated by least squares, so that ω _i =(X _i ^T X _i ) ^-1 X _i ^T Y _i

Among them, X _i is an l×2 matrix, the coefficients k _j,i =(i-1)l+j, j=1,2...l, Y _i is the i-th matrix including Y(k) A column vector of data point information in each slice.

整合为Y_t(k)，计算去趋势序列Y_l(k)，去趋势序列Y_l(k),k＝1,2,...,N被定义为Y(k)和Y_t(k)的差值：Trend fluctuation function calculation module, which calculates the data value of each segment

Integrate as Y _t (k), calculate the detrended series Y _l (k), the detrended series Y _l (k), k=1,2,...,N are defined as Y(k) and Y _t (k ) difference:

Y _l (k)=Y(k)-Y _t (k)

Among them, the values of Y _l (k) and Y _t (k) depend on the value of l;

The detrended fluctuation function F _d (l) is defined as:

The diabetes diagnosis module compares F _d (1) with the set threshold, and when it is less than or equal to the set threshold, it is determined as type 2 diabetes, and when it is greater than the set threshold, it is determined as type 1 diabetes.

From a systems biology perspective, diabetic glycemic fluctuations are the net result of the activity of a complex metabolic system, both interfered with by behaviors, such as physical activity and food intake, and in order to reduce postprandial glycemic fluctuation levels and reduce the risk of hypoglycemia and work done to ensure stable fasting blood glucose levels; in turn regulated by intrinsic hormone networks, including those of the pancreas, liver, gut, adipose tissue, kidneys, and brain. Under normal physiological conditions, behavioral influences can also cause blood sugar fluctuations, but they can be controlled within a small range. In the diabetic state, behavioral-influenced blood sugar fluctuations are magnified due to a dysregulated hormonal network. The blood sugar fluctuation under the influence of behavior is regarded as the trend fluctuation, and the detrended fluctuation function index makes it possible to start from the internal regulation ability of the hormone system, so that the hormone system can converge on its own under the condition of excluding external influences. The most prominent advantage of this system is that by filtering the trend components in the glucose time series, the influence of external factors on blood sugar fluctuations is reduced, and it is convenient to reveal the inherent fluctuation characteristics of the data. Glucose time series G(k) firstly obtains Y(k) by removing the mean value and accumulating and summing, thus eliminating the overall bias of the time series. Next, the summation sequence Y(k) is divided into equal length segments, and the trend component _Yt (k) is obtained by linear regression method according to the data point information contained in each segment. Finally, the variance of the detrended fluctuation component Y _l (k), that is, F _d (l), is calculated to reflect the systematic fluctuation after removing the trend term, thereby revealing the endogenous etiology of diabetes.

The detrended fluctuation function obtained by this system is a quantitative index obtained by data analysis method, which provides ideas for in-depth use of a large amount of data provided by blood glucose monitoring equipment and research on digital precise drug delivery in diabetes research. Currently, the classification of diabetes still relies on clinical judgment, lacks a widely accepted differential diagnosis process, and is difficult to homogenize. Although serum C-peptide is usually used to evaluate the intrinsic secretion of insulin and guide the classification of diabetes, there are also external interference factors such as non-standard measurement methods. At the same time, in medicine, Beta cell function is the primary consideration in determining the grading diagnosis of diabetes, which is usually reflected by serum C-peptide levels. Quantitative assessment is difficult because C-peptide levels interfere with blood glucose levels, renal function, and individual insulin resistance. The use of detrended fluctuation functions can be quantitatively used, which can assist in the diagnosis of diabetes to a certain extent. After obtaining the basic data of the patient's blood sugar fluctuation through the FGM device in this embodiment of the present invention, the diagnosis system can be implemented through MATLAB 2016bfor MAC programming, and the diagnosis system cleans, processes and subsequently calculates the data obtained by the blood sugar testing device to obtain the detrended fluctuation function F. The value of _d (l), the type of diabetes can be diagnosed by the trend fluctuation function F _d (l).

At the same time, the value of the detrended fluctuation function F _d (l) can also be combined with the mean blood glucose fluctuation amplitude (MAGE), blood glucose standard deviation (SD), blood glucose mean value (Mean BG) and 70-180 mg/dL time range percentage (TIR). ) indicators, and in the follow-up experiments, the optimization selection, comparative evaluation and effectiveness evaluation of indicators were carried out through SPSS software version 16.0 (SPSS Inc., Chicago, IL, US) for Windows.

The following is a verification of diabetes diagnosis using the detrended fluctuation function F _d (l):

In order to study the present invention, from January 2018 to June 2019, the Department of Endocrinology and Metabolism, Peking University People's Hospital recruited 113 inpatients with diabetes, the age range of these patients was 18-75 years old, and according to the 1999 World Health Organization (World Health Organization (WHO)) Diagnosis of diabetes mellitus according to WHO) criteria. Diabetes was diagnosed and classified clearly by an endocrinologist and independently confirmed by another specialist. Exclusion criteria for these patients included those who were currently using another CGM system, did not receive this new method of blood glucose monitoring, had a substantially altered medical condition, were unable to use the FGM system, and were known to be allergic to medical adhesives and pregnant. The study was approved by the Institutional Review Board of Peking University People's Hospital, and written informed consent was obtained from all participants.

This research invention includes the following steps:

In the first step, shown in Figure 1, all participants underwent a physical examination including height, weight, and blood pressure measurements before proceeding to the study. BMI is calculated by dividing weight in kilograms by height in square meters. Blood pressure was measured three times using a standard mercury sphygmomanometer and the measurements were averaged. Before the study, triglycerides (TG), total cholesterol (TC), high density lipoprotein cholesterol (HDL-C), low density lipoprotein cholesterol (LDL-C), fasting C-peptide, fasting insulin and fasting blood glucose (FPG) ) was determined by an enzyme immunoassay using a biochemical analyzer (7600-120; Hitachi, Tokyo, Japan). Hemoglobin A1c (HbA1c) levels in whole blood were measured using automated high performance liquid chromatography (PrimusUltra 2, Trinity Biotechnology, Bray, Cowicklow, Ireland) and standard procedures. In addition, participants were required to wear an instantaneous blood glucose monitoring (Freestyle Libre H, Abbott, USA) device daily to continuously monitor subcutaneous interstitial glucose. The device includes a glucose oxidase-based electrochemical sensor and a receiver. The sensor is placed subcutaneously and replaced every 14 days, and the receiver wirelessly transmits and stores interstitial glucose measurements every 15 minutes.

The second step is to carry out data processing and analysis according to the blood glucose monitoring data, and use this system to obtain the fluctuation function of all participants. Figure 2 shows the correlation analysis results of dividing the blood glucose data into different segments l. It can be seen that when l= At 34, the correlation coefficient is the highest, and this detrended fluctuation function is represented by F _d .

The third step is to calculate the indicators commonly used to reflect blood sugar fluctuations, including: mean blood sugar fluctuation range (MAGE), blood sugar standard deviation (SD), blood sugar average (Mean BG) and 70-180mg/dL time range percentage (TIR), For comparison with the resulting wave function performance. Figure 3 shows the Spearman correlation analysis results of the above five indicators, showing that there is a significant correlation between the fluctuation function and fasting C-peptide (r=-0.751; p<0.01), and the higher the F _d , the higher the fasting C-peptide. Low. Compared with the other four performance indicators, the proposed detrended fluctuation function F _d (r=-0.751p<0.01) had the strongest correlation with fasting C-peptide, among which MAGE had the lowest correlation coefficient (r=-0.334; p <0.01), even though it also represents the degree of blood glucose variability. The definition of TIR determines that it can reflect certain blood glucose fluctuation information, and its relationship with fasting C-peptide is higher than the average blood glucose (r=0.620, P<0.01). SD describes the degree of dispersion between a group of individuals (r=-0.678; p<0.01), outperforming Mean BG and TIR, but lower than the wave function _Fd .

The fourth step is to evaluate the classification effect of the proposed indicators. Figure 4 is the histogram of the fluctuation function of all participants, and the value of the fluctuation function F _d of different subjects is between 3.380 and 15.361. The mean was 8.188, the interquartile range (IQR) was 4.400, and the histograms of the F _d distributions for all participants showed a bimodal distribution. In addition, the performance of the detrended fluctuation function F _d was evaluated separately for patients with type 1 diabetes and type 2 diabetes. Among them, the median and IQR of the type 1 diabetes group were 10.007 and 2.578, respectively; the median of the type 2 diabetes group was 5.844, and the IQR was 1.883, as shown in Figure 5. Figure 6 is a graph of receiver operating characteristics for evaluating the effect of wave function typing. The area under the curve (AUC) reached 0.866, p=0.000 for significance. The Youden Index (Yi) was 0.693. These statistics show that F _d can be used to guide the classification of diabetes, and the cut-off point is preferably 0.7 (sensitivity 88.5%, specificity 81.8%).

The embodiments of the present invention are described in detail above in conjunction with the accompanying drawings. Based on the instantaneous monitoring of blood glucose data, we use the detrended fluctuation function for analysis to establish new indicators for predicting beta cell function and diabetes classification, and use data analysis and evaluation methods to evaluate the indicators. effectiveness was evaluated. This will shed light on leveraging the vast amount of data provided by blood glucose monitoring devices and studying digital precision drug delivery in diabetes research.

To sum up, the above are only preferred embodiments of the present invention, and are not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims (4)

1. A diabetes typing diagnosis system is characterized in that the diagnosis system collects the glucose measurement values of participants and carries out sectional processing to obtain a trend fluctuation function, and the diabetes is diagnosed according to the trend fluctuation function;

the diagnosis system comprises a data acquisition and preprocessing module, a full data processing module, a segmented data processing module, a trend fluctuation function calculation module and a diabetes diagnosis module;

a data acquisition and preprocessing module for acquiring glucose measurement values of participants to obtain a length N ₁ The glucose time sequence G (k) of (1), the entire sequence being divided into n segments of length l

A full data processing module for calculating the mean value of the sequence G (k)

Recalculate the mean-removed sequence

The summation sequence y (k) is then calculated as:

a segment data processing module for calculating data value of each segment

Wherein i ═ 1,2, …, n, k ═ i-1) l +1, (i-1) l +2, …, il;

definition of ω _i ＝[ω _0,i ,ω _1,i ] ^T ，ω _i Is obtained by least squares calculation, thereby omega _i ＝(X _i ^T X _i ) ^-1 X _i ^T Y _i ；

Wherein the coefficient k _j,i ＝(i-1)l+j，j＝1,2……l，Y _i Is a column vector containing the information of data points in the ith segment of Y (k);

a trend fluctuation function calculation module for calculating the data value Y of each segment _t ⁱ (k) Integration into Y _t (k) Calculating a detrending sequence Y _l (k)；

Y _l (k)＝Y(k)-Y _t (k)

Defining a detrending fluctuation function F _d (l) Comprises the following steps:

diabetes diagnostic Module, will F _d (l) And comparing with a set threshold, judging as type 2 diabetes when the value is less than or equal to the set threshold, and judging as type 1 diabetes when the value is greater than the set threshold.

2. The diabetes typing diagnosis system according to claim 1, wherein l.gtoreq.10.

3. The diabetes typing diagnostic system according to claim 1, wherein l-34.

4. The diabetes typing diagnostic system according to claim 1, wherein the threshold value is 0.7.

Images