SYSTEM AND METHOD FOR THE SIMULTANEOUS, NON-INVASIVE ESTIMATION OF BLOOD GLUCOSE, GLUCOCORTICOID LEVEL AND BLOOD PRESSURE

• US 20130267796A1
• Filed: 11/30/2011
• Published: 10/10/2013
• Est. Priority Date: 12/01/2010
• Status: Abandoned Application

• Current Assignee: Universitat Politecnica De Catalunya
• Original Assignee: Universitat Politecnica De Catalunya
• Inventors: Enric Monte Moreno, Enric
• Application Number:
  US13/991,034
• Publication Number:
  US 20130267796A1
• Time in Patent Office: Days
• US Class Current: 600/301
• CPC Class Codes: A61B 5/021 Measuring pressure in heart or blood vessels A61B5/0205 takes precedence
• CPC Class Codes: A61B 5/14532 for measuring glucose, e.g. by tissue impedance measurement
• CPC Class Codes: A61B 5/14551 for measuring blood gases
• CPC Class Codes: A61B 5/7235 Details of waveform analysis A61B5/349 takes precedence
• CPC Class Codes: A61B 5/7257 using Fourier transforms
• CPC Class Codes: A61B 5/7278 Artificial waveform generation or derivation, e.g. synthesising signals from measured signals A61B5/327 takes precedence

Abstract

System and method for the simultaneous and non-invasive estimation of blood glucose, glucocorticoid and pressure levels. It comprises an activity module detector for a digitalized signal acquired by a sensor, representing the distal heartbeat of a person, selecting a segment of consecutive signal samples of fixed duration S_window(t) and using the same to generate sub-windows S_frame(t, n) of shorter duration; a signal processing module which receives the two signals S_window(t) and S_frame(t, n) and which delivers a vector X_F at its output with the parameters of a physiological model and module based on automatic learning, which receives the X_F vector and information on a person'"'"'s characteristics and provides an estimation of blood glucose levels (BGL), systolic pressure levels (SPL), diastolic pressure levels (DPL) and glucocorticoid levels (GCL) at its output.

Claims

1. System for the simultaneous and non-invasive estimation of glucose in blood, glucocorticoid levels and blood pressure, based on a person'"'"'s distal heartbeat waveform, acquired from a sensor, which provides a digitalized signal, comprising:

an activity detection module for said digitalized signal, which selects a segment of consecutive samples from said digitalized signal, with a fixed duration and which are referred to as S_window(t) and, based on the same, generates a sequence of sub-windows or frames referred to as S_frame(t, n) which are shorter in duration than those of S_window(t), wherein the index t indicates the sample number within a frame and n represents the frame number;

a signal processing module which receives the two signals S_window(t) and S_frame(t, n) containing the waveform of the distal heartbeat in a fixed duration module, and which provides an X_F vector as output, which in turn contains the parameters of a physiological model; and

an automatic learning based module to which said X_F vector is fed alongside information on the characteristics of the person and which provides an estimation of blood glucose levels (BGL), systolic pressure levels (SPL), diastolic pressure levels (DPL) and glucocorticoid levels (GCL) as output.

2. System, according to claim 1, wherein said sensor is of the plethysmograph type selected from optical, acoustic or mechanical, including a pulse oximeter system (SpO2) and in that said activity detection module comprises:

a frame segmentation sub-module which provides said S_frame(t, n) sequence;

a sub-module to calculate the energy of each frame;

a sub-module to calculate the spectral entropy of each frame; and

a sub-module to detect zero crosses,

using said sub-modules to generate an X_AD(n) vector, which is fed to a classifier sub-module, which implements a decision function and whose index for each frame indicates whether the frame belongs to a “

useful signal”

category or corresponds to an “

absence of signal”

.

3. System according to claim 2, wherein said classifier sub-module is associated with a database labeled with the category to which each frame belongs for its training.

4. System, according to claim 3, wherein said classifier is selected from the group consisting of:

k-nearest neighbor, linear discriminants, including the Fisher discriminant, quadratic discriminants, decision trees and “

support vector machines”

.

5. System, according to claim 2, wherein said activity detection module also includes a sub-module in its output for determining the useful segment the sequence of categories, to which each frame belongs, receives as input and said sub-module integrates a finite-state automaton used to determine segments of consecutive frames of useful signal, comprising the following states:

s1=“

absence of useful signal”

s2=“

useful signal”

s3=“

ambiguous area”

and each state of the automaton have counters defined as follows;

C_Sⁱ;

number of consecutive frames in the i^th state classified as “

useful signal”

; and

C_ASⁱ;

number of consecutive frames in the i^th state classified as “

absence of signal”

.

6. System, according to claim 1, wherein said automatic learning module uses a committee of predictors trained by means of the “

AdaBoost”

algorithm, using a “

radial basic function”

type neuronal network as a basic predictor of the committee of predictors, which takes advantage of the interactions between the pressure values, glucocorticoid level and glucose level in order to improve estimations.

7. Method for the simultaneous and non-invasive estimation of glucose in blood, glucocorticoid levels and blood pressure, based on a person'"'"'s distal heartbeat waveform, acquired from a sensor, which provides a digitalized signal, comprising:

detecting the activity of said digitalized signal by selecting a segment of consecutive samples of said digitalized signal, of fixed duration and referred to as S_window(t) and using said segment to generate a sequence of sub-windows or frames referred to as S_frame(t, n) and of lesser duration than those of S_window(t), wherein the index t indicates the sample number within a frame and n represents the frame number making it possible to eliminate those signal segments which are not useful, comprising the initial transition, false clicks, signal losses, noise and saturation;

processing the two signals S_window(t) and S_frame(t, n), which contain the distal heartbeat waveform in a fixed duration segment and based on the same, generating an X_F vector, which contains the parameters of a physiological model; and

feeding said X_F vector and information on a person'"'"'s characteristics into a module based on automatic learning and which provides an estimate of blood glucose levels (BGL), systolic pressure levels (SPL), diastolic pressure levels (DPL) and glucocorticoid levels (GCL) as output.

8. Method, according to claim 8, characterized in that claim 7, wherein said activity detection stage comprises:

segmenting the digitalized signal into frames S_window(t) providing a sequence of S_frame(t, n);

calculating the energy of each frame;

calculating the Fast Fourier Transform (FFT) of the frame for the spectral entropy of each frame before normalizing the energy spectrum of the frame and;

detecting zero crosses,

using these energy and spectral energy values to generate an X_AD(n) vector per frame, which is fed into a classifier sub-module, which in turn implements a decision function and whose index for each frame indicates whether the frame belongs to a “

useful signal”

category or corresponds to an “

absence of signal”

.

9. Method according to claim 8, characterized in that wherein said segmentation of S_window(t) is carried out in 5 second frames with a 50% digitalized signal overlap.

10. Method according to claim 8, wherein said zero cross detection is carried out by using the average value of S_frame(t, n) to eliminate the continuous component before counting the number of times the signal crosses the zero threshold.

11. Method according to claim 8, wherein for each frame, the parameters of energy, spectral entropy and zero crosses are grouped together in a vector, which will be the input of a classifier, which will decide whether or not the n^th frame corresponds to a useful frame:

X_AD(n)=[E_frame(n),E_frame^σ

(n),E_frame^skew(n),H_frame^s(n),Z_frame^c(n)]^T

12. Method according to claim 11, wherein said classifier is selected from the group consisting of:

k nearest neighbor, linear discriminants, including the Fisher discriminant, quadratic discriminants, decision trees and support vector machines and in that the classifier is trained with a database previously labeled with the category to which each frame belongs.

13. Method, according to claim 9, wherein in order to determine a useful segment, the sequence of categories to which each frame obtained at the output of said classifier sub-module belongs, is fed into a sub-module, which contains a finite-state automaton applied to certain segments of consecutive frames of useful signal, comprising the following states:

s1=“

absence of useful signal”

s2=“

useful signal”

s3=“

ambiguous area”

and each state of the machine will contain counters, defined as follows;

C_Sⁱ;

number of consecutive frames in the i^th state, classified as “

useful signal” and

;

C_ASⁱ;

number of consecutive frames in the i^th state classified as “

absence of signal”

.

14. Method, according to claim 8, wherein the signal processing stage entails generating an X_F vector containing the parameters, which characterize a person'"'"'s physiological state, using overall parameters calculated based on S_window(t) and aggregated parameters of the characteristics calculated in the sequence S_frame(t, 1), S_frame(t, 2), . . . , S_frame(t, N_frame), which are useful for modeling the short-term evolution of various physiological parameters and respiratory rate, using the cepstral analysis to obtain information on the spectral content.

15. Method according to claim 14, wherein the average value of each sequence of cepstral parameters is subtracted to offset the effect of the specific sensor, calculating an average value of the cepstrums, each time a sensor is changed, for each group of parameters and storing said average value to carry out the subtraction during the signal processing stage.

16. Method, according to claim 7, wherein said module based on automatic learning receives the X_F vector at its input, along with the person'"'"'s physical characteristics, including at least gender, age and body mass index and enters the three variables of interest at its output:

blood glucose levels (BGL), glucocorticoid levels (GCL) and blood pressure levels (SPL and DPL) applying an algorithm, which implements a non-linear regression between said X_F input and the three variables cited, selecting a regression method, which models the interaction between the three variables.

17. Method, according to claim 16, wherein said automatic learning uses a committee of predictors trained by means of the “

AdaBoost”

algorithm, using a “

radial basis function”

type neuronal network as a basic predictor of the committee of predictors, which takes advantage of the interactions between pressure values, glucocorticoid levels and glucose level in order to improve estimations.