CN112992337B - Lung function assessment algorithm, device, medium and equipment for cervical and spinal cord injury patient - Google Patents

Lung function assessment algorithm, device, medium and equipment for cervical and spinal cord injury patient Download PDF

Info

Publication number
CN112992337B
CN112992337B CN202110168260.XA CN202110168260A CN112992337B CN 112992337 B CN112992337 B CN 112992337B CN 202110168260 A CN202110168260 A CN 202110168260A CN 112992337 B CN112992337 B CN 112992337B
Authority
CN
China
Prior art keywords
lung function
function
cough
sample
patient
Prior art date
Legal status (The legal status 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 status listed.)
Active
Application number
CN202110168260.XA
Other languages
Chinese (zh)
Other versions
CN112992337A (en
Inventor
莫鸿强
周樊
曾键沣
章臻
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
South China University of Technology SCUT
Original Assignee
South China University of Technology SCUT
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 South China University of Technology SCUT filed Critical South China University of Technology SCUT
Priority to CN202110168260.XA priority Critical patent/CN112992337B/en
Publication of CN112992337A publication Critical patent/CN112992337A/en
Application granted granted Critical
Publication of CN112992337B publication Critical patent/CN112992337B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • G16H50/20—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
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/08—Detecting, measuring or recording devices for evaluating the respiratory organs
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching

Landscapes

  • Health & Medical Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biomedical Technology (AREA)
  • Public Health (AREA)
  • Medical Informatics (AREA)
  • Data Mining & Analysis (AREA)
  • Pathology (AREA)
  • General Health & Medical Sciences (AREA)
  • Physics & Mathematics (AREA)
  • Animal Behavior & Ethology (AREA)
  • Surgery (AREA)
  • Theoretical Computer Science (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Physiology (AREA)
  • Molecular Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Biophysics (AREA)
  • Veterinary Medicine (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Signal Processing (AREA)
  • Psychiatry (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Pulmonology (AREA)
  • Databases & Information Systems (AREA)
  • Epidemiology (AREA)
  • Primary Health Care (AREA)
  • Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)

Abstract

The invention provides an algorithm, a device, a medium and equipment for evaluating lung function of a cervical and spinal cord injury patient; the algorithm comprises the following steps: acquiring a cough tone signal to be decided; deconvoluting the cough tone signal to reflect an excitation component of the pulmonary exhaled air flow; calculating the short-time energy maximum value of the excitation source component and the descending time of the root-mean-square curve to form a feature vector of the sample to be decided, and inputting the feature vector into a classification decision function to obtain a decision value of the sample to be decided; the classification decision function is obtained by training the initial classification decision function; and evaluating the lung function according to the decision value. The method solves the problem that the method is easily influenced by the resonance peak of the sound channel sound transmission transfer function when the intensity of the lung function is evaluated by directly using the time domain peak intensity of the cough sound; meanwhile, the algorithm model is simple and small in calculation amount, and is convenient to realize in wearable equipment.

Description

Lung function assessment algorithm, device, medium and equipment for cervical and spinal cord injury patient
Technical Field
The invention relates to the technical field of medical instruments and medical signal processing, in particular to an algorithm, a device, a medium and equipment for evaluating lung functions of a patient with cervical and spinal cord injury.
Background
With the rapid development of industrialization and the increase of traffic flow, the incidence of cervical spinal cord injury is remarkably increased. Cervical spinal cord injury is a destructive nerve injury that can lead to paralysis of the patient's respiratory muscles. While paralysis of the respiratory muscles results in a decrease in the function of the patient's inspiratory and expiratory muscles, which in turn leads to a failure of the patient's lung function. Respiratory complications resulting from lung failure are the leading cause of early death in patients with cervical spinal cord injury. Therefore, lung function from cervical spinal cord injury is a very important indicator during a patient's stay. The strength of lung function determines the severity and survival probability of the patient's condition and also determines whether the patient needs additional adjuvant therapy.
Detection of pulmonary function in patients with cervical spinal cord injury is a difficult problem. The lung function can be accurately evaluated by using a lung function instrument, but the measurement mode is complex to operate and high in cost, and daily measurement is difficult to carry out in clinical practice.
One possible alternative at present is to assess the respiratory capacity of a patient based on the peak of the sound intensity of his cough sound, the principle of which is: when a patient coughs forcefully, the intensity of the expiratory airflow determines the sound intensity of the cough sound, so that the peak sound intensity of the cough sound can reflect the flow rate of the expiratory airflow in the lung to a certain extent, and the lung function can be indirectly evaluated. However, if the cough sound is regarded as the modulation of the flow sounds exhaled by the lungs by the vocal tract, the cough sound intensity is dependent on the vocal tract sound transfer function in addition to the flow rate of the flow exhaled by the lungs; even with a low flow rate of the pulmonary expiratory airflow, the cough sounds may produce a high sound intensity peak due to the vocal tract modulation, giving a false lung function assessment result.
How to more accurately evaluate lung function of a cervical-spinal cord injury patient based on cough sound is not solved at present.
Disclosure of Invention
To overcome the defects and shortcomings of the prior art, the invention aims to provide an algorithm, a device, a medium and equipment for evaluating the lung function of a cervical spine marrow injury patient
In order to achieve the purpose, the invention is realized by the following technical scheme: an algorithm for evaluating lung function of a patient with cervical spinal cord injury is characterized in that: the method comprises the following steps:
s1, acquiring a cough tone signal to be decided;
s2, deconvoluting the cough tone signal to reflect the exciting source component y of the lung expired air flowi(k) (ii) a Calculating the excitation source component yi(k) Short time energy maximum of
Figure BDA0002938037790000021
And the fall time of the root mean square curve
Figure BDA0002938037790000022
S3 maximum energy in short time
Figure BDA0002938037790000023
And the fall time of the root mean square curve
Figure BDA0002938037790000024
Feature vectors forming a sample to be decided
Figure BDA0002938037790000025
And inputting the decision value g (phi) into a classification decision function to obtain a decision value g (phi) of a sample to be decidedu) (ii) a The classification decision function is obtained by training the initial classification decision function;
s4, according to the decision value g (phi)u) Assessment of lung function: if g (phi)u) If the lung function is more than 0, the sample is evaluated to be strong, otherwise, the lung function is weak.
Preferably, the step S2 includes the following sub-steps:
s21, framing and windowing the cough sound signal; wherein, the frame length is set to wlen, the frame shift (i.e. the displacement of the next frame to the previous frame) is set to inc, the window function length win is set to be the same as the frame length, the ith frame signal is marked as xi(k) K is 1,2, …, wlen, i is 1,2, …, l, l is the total number of frames after framing, and wlen, inc, i, k, l are positive integers;
s22, for the ith frame signal xi(k) Deconvoluting an excitation sequence y reflecting the pulmonary expiratory airflow by linear predictive codingi(k)；
S23, calculating the excitation source frame sequence yi(k) Short-time energy en ofiAnd searching for a short-term energy maximum en in all excitation source framesmax；
S24, calculating the excitation source frame sequence yi(k) Root mean square value of
Figure BDA0002938037790000031
To pair
Figure BDA0002938037790000032
Plotting a root mean square curve of the cough sound excitation source signal and calculating the falling time t of the root mean square curved。
Preferably, in S23, the source frame sequence y is excitedi(k) Short-time energy en ofiThe calculation formula is as follows:
Figure BDA0002938037790000033
preferably, in S24, the source frame sequence y is excitedi(k) Root mean square value of (Yrms)iThe calculation formula is as follows:
Figure BDA0002938037790000034
time of fall t of root mean square curvedThe time required for the rms curve amplitude to first drop to 0.25 times the maximum amplitude from the time point of the maximum amplitude of the rms curve to that time point is taken.
Preferably, in S21, the cough tone signal is subjected to end point detection, and then is framed and windowed.
Preferably, in S21, the frame length wlen is set to a value according to a frame duration of 20 to 30ms, the frame shift inc is set to a value of 0.25 to 0.75 times the frame length, and the window function is a hanning window;
in the step S22, the solution of the equation set in the linear predictive coding is implemented by using a Levinson-Durbin algorithm.
Preferably, the training of the initial classification decision function refers to:
establishing a linear classifier with a decision function of g (phi) ═ wTφ+w0(ii) a Wherein g (φ) is a decision value of sample φ, [ en ═max，td]TIs the feature vector of the sample, w ═ w1，w2]TAnd w0Is the undetermined parameter of the linear classifier;
collecting cough tone signals of patients with strong lung function and weak lung function with neck and spinal cord injuries as training samples to form a sample set, wherein the number of the cough tone signal samples of the patients with strong lung function is m, and the labels of the samples are set to be 1; the number of cough tone signal samples of the patient with weak lung function is n, and the label of each sample is set to be-1;
calculating the short-time energy maximum value en of the cough tone signals in the sample set according to the step S2maxAnd the fall time t of the root mean square curvedThe eigenvector phi forming the corresponding sample is ═ enmax，td]T(ii) a The feature vector set corresponding to the cough sound signal sample of the patient with strong memory lung function is
Figure BDA0002938037790000041
Figure BDA0002938037790000042
The feature vector set corresponding to the cough sound signal sample of the patient with weak lung function is
Figure BDA0002938037790000043
Wherein
Figure BDA0002938037790000044
Is the characteristic vector of the ith patient with strong lung function,
Figure BDA0002938037790000045
a feature vector for the ith patient with weak lung function;
training a linear classifier to obtain a classification decision function by taking the feature vectors in the feature vector set as input and the sample label values corresponding to the feature vectors as expected output
Figure BDA0002938037790000046
Wherein, w*And
Figure BDA0002938037790000047
respectively the best projection direction and the best threshold value obtained by training.
Preferably, the linear classifier is a Fisher linear classifier.
The principle of the evaluation method of the invention is as follows:
the cough tone signal may be viewed as a modulation of the flow sounds exhaled by the lungs by the vocal tract. The flow rate of the exhaled air flow, particularly the peak flow rate, can reflect the respiratory muscle force and further can reflect the lung function; therefore, the peak sound intensity of the cough sound may reflect the lung function to some extent. The peak sound intensity of the cough sound, however, is dependent on the flow rate of the expired air stream from the lungs as well as on the resonance peak of the vocal tract transfer function. Therefore, the excitation source component of the lung exhalation airflow is deconvolved from the cough tone signal, and the influence of the vocal tract can be eliminated, thereby more accurately reflecting the lung function. The linear predictive coding can estimate the vocal tract model more accurately and has simple calculation. Therefore, the present invention S22 uses linear predictive coding to obtain the excitation source component of the cough sound signal. Since the cough signal has a short-term stationary characteristic, the present invention S21 requires framing and windowing of the signal to maintain its "quasi-stationary" characteristic. And an overlapping portion needs to be provided between two frames in order to ensure that the characteristics of the two adjacent frames change smoothly.
The short-time energy maximum of the cough tone signal driver component and the fall time of the root mean square curve can be used as features for effectively evaluating lung function: the short-time energy maximum value can directly reflect the respiratory muscle strength of the lung airflow during exhalation, the short-time energy maximum value of a patient with weak lung function is smaller, and the short-time energy maximum value of a patient with strong lung function is larger; patients with weak lung function have a slower rate of root mean square drop of the stimulus component than patients with strong lung function because of insufficient respiratory muscle strength and longer duration of exhalation of the pulmonary airflow than patients with strong lung function. Therefore, the invention S2 calculates the short-time energy maximum of the cough tone signal excitation source component and the fall time of the root-mean-square curve as the feature vector of the linear classifier respectively.
The calculation algorithm of the invention only uses a small amount of characteristics to evaluate the strength of the lung function according to the characteristics of the cough sound excitation source components of the patient with strong lung function and the patient with weak lung function, and simultaneously greatly reduces the calculation amount, thereby reducing the requirements for hardware realization.
A device for evaluating lung function of a patient with cervical spinal cord injury is characterized in that: the method comprises the following steps:
the signal acquisition module is used for acquiring a cough sound signal to be decided;
a preprocessing module for deconvolving the cough tone signal to an excitation component y reflecting the pulmonary exhaled air flowi(k) (ii) a Calculating the excitation source component yi(k) Short time energy maximum of
Figure BDA0002938037790000051
And the fall time of the root mean square curve
Figure BDA0002938037790000052
A classification decision module for maximizing short-time energy
Figure BDA0002938037790000053
And the fall time of the root mean square curve
Figure BDA0002938037790000054
Feature vectors forming a sample to be decided
Figure BDA0002938037790000055
And inputting the decision value g (phi) into a classification decision function to obtain a decision value g (phi) of a sample to be decidedu) (ii) a The classification decision function is obtained by training the initial classification decision function;
an evaluation module for evaluating the decision value g (phi)u) Assessment of lung function: if g (phi)u) If the lung function is more than 0, the sample is evaluated to be strong, otherwise, the lung function is weak.
A storage medium, wherein the storage medium stores a computer program which, when executed by a processor, causes the processor to execute the above-described cervical and spinal cord injury patient lung function assessment algorithm.
A computing device comprising a processor and a memory for storing a program executable by the processor, wherein the processor implements the above algorithm for evaluating lung function in a patient with cervical spinal cord injury when executing the program stored in the memory.
Compared with the prior art, the invention has the following advantages and beneficial effects:
1. the method overcomes the defects of the conventional lung function evaluation algorithm, and eliminates the influence of a resonance peak value of a sound channel sound transmission transfer function by deconvolving an excitation source component of the airflow exhaled from the lung from the cough sound by using linear predictive coding;
2. the method only utilizes the short-time energy of the cough sound signal excitation source and the descending time of the root-mean-square curve as characteristics to evaluate the lung function, has simple model, small calculated amount and low requirement on hardware realization, and is convenient to realize in wearable equipment.
Drawings
FIG. 1 is a flow chart of an embodiment of an algorithm for evaluating lung function in a patient with cervical spinal cord injury;
FIGS. 2(a) -2 (d) are time-domain graphs of the cough tone signal and its RMS curve of patients with strong lung function and patients with weak lung function;
3(a) -3 (d) are time domain plots of the short-time energy of the coughing tone driver component and the rms curve of the driver component for patients with strong and weak lung function;
FIG. 4 is a flowchart of an algorithm for evaluating lung function of a patient with cervical and spinal cord injury according to an embodiment.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings and specific embodiments.
Example one
The lung function assessment algorithm for the cervical and spinal cord injury patient is characterized by collecting cough tone signals of forceful coughs of patients with strong lung functions and weak lung functions; eliminating the influence of sound channels by utilizing linear predictive coding, and deconvoluting an excitation source component of the pulmonary exhaled air flow from the cough sound; calculating the short-time energy and the root mean square value of the excitation source component and drawing a root mean square curve of the short-time energy and the root mean square value; establishing a linear classifier by taking the short-time energy maximum value of the excitation source component and the falling time of the root-mean-square curve as characteristics, and training the linear classifier by taking patient cough sound samples with strong lung function and weak lung function as a sample set; and evaluating the lung function strength of the cervical spinal cord injury patient by a trained linear classifier.
The work flow is shown in fig. 1, and comprises the following steps:
and S1, acquiring the cough tone signal to be decided.
S2, deconvoluting the cough tone signal to reflect the exciting source component y of the lung expired air flowi(k) (ii) a Calculating the excitation source component yi(k) Short time energy maximum of
Figure BDA0002938037790000071
And the fall time of the root mean square curve
Figure BDA0002938037790000072
The method comprises the following steps:
s21, framing and windowing the cough sound signal; wherein, the frame length is set to wlen, the frame shift (i.e. the displacement of the next frame to the previous frame) is set to inc, the window function length win is set to be the same as the frame length, the ith frame signal is marked as xi(k) K is 1,2, …, wlen, i is 1,2, …, l, l is the total number of frames after framing, and wlen, inc, i, k, l are positive integers; preferably, the method comprises the following steps: the frame length wlen is taken as a value according to the frame duration of 20-30 ms, the frame shift inc is taken as 0.25-0.75 time of the frame length, and the window function is a Hanning window;
s22, for the ith frame signal xi(k) Deconvoluting an excitation sequence y reflecting the pulmonary expiratory airflow by linear predictive codingi(k) (ii) a The solution of the equation set in the linear predictive coding is preferably realized by adopting the existing algorithm Levinson-Durbin algorithm;
s23, calculating the excitation source frame sequence yi(k) Short-time energy en ofiAnd searching for a short-term energy maximum en in all excitation source framesmax；
Preferably a sequence of excitation source frames yi(k) Short-time energy en ofiThe calculation formula is as follows:
Figure BDA0002938037790000073
short-term energy maximum enmaxThe calculation formula of (2) is as follows:
enmax＝max{en1，…，eni，…，enl}
in practical application, other formulas can be adopted for calculating the short-time energy;
s24, calculating the excitation source frame sequence yi(k) Root mean square value of
Figure BDA0002938037790000081
To pair
Figure BDA0002938037790000082
Plotting a root mean square curve of the cough sound excitation source signal and calculating the falling time t of the root mean square curved；
Preferably a sequence of excitation source frames yi(k) Root mean square value of (Yrms)iThe calculation formula is as follows:
Figure BDA0002938037790000083
time of fall t of root mean square curvedTaking the time required for the amplitude of the root-mean-square curve to firstly drop to 0.25 times of the maximum amplitude after the time point of the maximum amplitude of the root-mean-square curve reaches the time point;
other methods may be used to calculate the time of descent of the rms and rms curves in practical applications.
S3 maximum energy in short time
Figure BDA0002938037790000084
And the fall time of the root mean square curve
Figure BDA0002938037790000085
Feature vectors forming a sample to be decided
Figure BDA0002938037790000086
And input into a classification decision function to obtain the information to be classifiedDecision value g (phi) of decision sampleu) (ii) a The classification decision function is obtained by training the initial classification decision function;
s4, according to the decision value g (phi)u) Assessment of lung function: if g (phi)u) If the lung function is more than 0, the sample is evaluated to be strong, otherwise, the lung function is weak.
The training process of the initial classification decision function refers to:
establishing a linear classifier, wherein the Fisher linear classifier is preferably adopted by the linear classifier, and other forms can be adopted by the linear classifier; the decision function is g (phi) ═ wTφ+w0(ii) a Wherein g (φ) is a decision value of sample φ, [ en ═max，td]TIs the feature vector of the sample, w ═ w1，w2]TAnd w0Is the undetermined parameter of the linear classifier;
collecting cough tone signals of patients with strong lung function and weak lung function with neck and spinal cord injuries as training samples to form a sample set, wherein the number of the cough tone signal samples of the patients with strong lung function is m, and the labels of the samples are set to be 1; the number of cough tone signal samples of the patient with weak lung function is n, and the label of each sample is set to be-1;
calculating the short-time energy maximum value en of the cough tone signals in the sample set according to the step S2maxAnd the fall time t of the root mean square curvedThe eigenvector phi forming the corresponding sample is equal to [ en ]max，td]T(ii) a The feature vector set corresponding to the cough sound signal sample of the patient with strong memory lung function is
Figure BDA0002938037790000091
Figure BDA0002938037790000092
The feature vector set corresponding to the cough sound signal sample of the patient with weak lung function is
Figure BDA0002938037790000093
Wherein
Figure BDA0002938037790000094
Is the characteristic vector of the ith patient with strong lung function,
Figure BDA0002938037790000095
a feature vector for the ith patient with weak lung function;
training a linear classifier to obtain a classification decision function by taking the feature vectors in the feature vector set as input and the sample label values corresponding to the feature vectors as expected output
Figure BDA0002938037790000096
Wherein w*And
Figure BDA0002938037790000097
respectively the best projection direction and the best threshold value obtained by training.
Fig. 2(a) is a time-domain diagram of short-time energy of a cough tone signal of a patient with strong lung function, fig. 2(b) is a time-domain diagram of short-time energy of a cough tone signal of a patient with weak lung function, fig. 2(c) is a time-domain diagram of a root-mean-square curve of a cough tone signal of a patient with strong lung function, and fig. 2(d) is a time-domain diagram of a root-mean-square curve of a cough tone signal of a patient with weak lung function.
Fig. 3(a) is a time domain diagram of the short-term energy of the cough sound stimulus component of the patient with strong lung function, fig. 3(b) is a time domain diagram of the short-term energy of the cough sound stimulus component of the patient with weak lung function, fig. 3(c) is a time domain diagram of the root mean square curve of the cough sound stimulus component of the patient with strong lung function, and fig. 3(d) is a time domain diagram of the root mean square curve of the cough sound stimulus component of the patient with weak lung function.
The principle of the evaluation method of the invention is as follows:
the cough sound signal can be regarded as the modulation of the sound channel on the flow sound of the exhaled air of the lung, and the flow speed of the exhaled air flow, particularly the peak flow speed, can reflect the respiratory muscle force and further can reflect the lung function; therefore, the peak sound intensity of the cough sound may reflect the lung function to some extent. However, the peak intensity of the cough sound depends on the flow rate of the exhaled airflow in the lung and is also related to the resonance peak of the vocal tract transfer function, so that the signal short-time energy time domain diagram and the root mean square curve of the cough sound of the patient with strong lung function and the patient with weak lung function are not greatly different when the influence of the resonance peak of the vocal tract transfer function is not eliminated (as shown in fig. 2(a) to 2 (d)). After the vocal tract influence is removed by deconvolving the cough sound signal, the short-time energy maximum of the cough sound excitation source component of the patient with strong lung function is significantly greater than that of the patient with weak lung function (as shown in fig. 3(a) and 3 (b)), and the root mean square value of the cough sound excitation source component of the patient with strong lung function is significantly faster than that of the patient with weak lung function (as shown in fig. 3(c) and 3 (d)), so the short-time energy maximum and the root mean square curve of the excitation source component calculated in step S2 of the present invention can be used as an effective feature for distinguishing the two. The calculation algorithm of the invention only uses a small amount of characteristics to evaluate the strength of the lung function according to the characteristics of the cough sound excitation source components of the patient with strong lung function and the patient with weak lung function, and simultaneously greatly reduces the calculation amount, thereby reducing the requirements for hardware realization.
The evaluation algorithm of the present invention is described below with specific examples:
an algorithm for evaluating lung function of a cervical spinal cord injury patient comprises the following steps:
s1, acquiring a cough tone signal to be decided;
s2, framing and windowing all the cough sound signals, where the frame length is set to 20ms, i.e., wlen is 0.02 fs, the frame shift (i.e., the displacement of the next frame to the previous frame) is set to 0.25 times the frame length, i.e., inc is 0.25 wlen, the window function length win is set to be the same as the frame length, and the ith frame signal is marked as xi(k) K is 1,2, …, wlen, i is 1,2, …, l, l is the total number of frames after framing, and wlen, inc, i, k, l are positive integers;
for the ith frame signal xi(k) Excitation source frame sequence y for decoding and uncoiling lung exhalation airflow by 15-order linear predictioni(k) The predictor adopts an all-pole model, and the formula is as follows:
Figure BDA0002938037790000101
Figure BDA0002938037790000111
wherein the coefficient anAnd the gain factor G is a parameter of the model, and p is a model order. The prediction error gu (k) is the excitation source component of the cough tone signal that is unwound in the present invention.
It should be noted that the vocal tract model is not completely an all-pole model when coughing, but the all-pole model is the easiest to calculate, and the zero can be approximated by a pole
Figure BDA0002938037790000112
This approximation is completely reasonable as long as the order p of the model is high enough, and if the rate of convergence of the denominator polynomial is fast enough, the pole on the right of the equation can replace the zero on the left of the equation well with only a few terms;
computing an excitation source frame sequence yi(k) Short-time energy en ofiAnd searching for the maximum value en of the short-term energy in all excitation source framesmax. Excitation source frame sequence yi(k) The short-time energy calculation formula is as follows:
Figure BDA0002938037790000113
wherein eniA short-time energy value of an ith excitation source frame signal; other formulas may be used for the calculation of the short-time energy.
Computing an excitation source frame sequence yi(k) Root mean square value of (Yrms)iThe calculation formula is as follows:
Figure BDA0002938037790000114
wherein, YrmsiFor the ith excitation source frame yi(k) The root mean square value of; for YrmsiPlotting the cough sound excitation source signalAnd calculating the falling time t of the root mean square curvedThe time for the decrease is the time required for the curve amplitude to decrease to 0.25 times of the maximum amplitude for the first time from the time point of the maximum amplitude of the curve to the time point, and the time for the decrease of the root mean square curve is the time required for the curve amplitude to decrease to 0.25 times of the maximum amplitude for the first time from the time point of the maximum amplitude to the time point and the time required for the sum of the minimum amplitude of the curve because the amplitude of the root mean square curve does not necessarily decrease to less than 0.25 times of the maximum amplitude; it should be noted that, the calculation method of the root mean square value and the value of the root mean square curve descending time are not unique, and other calculation methods can be adopted;
s3 maximum energy in short time
Figure BDA0002938037790000121
And the fall time of the root mean square curve
Figure BDA0002938037790000122
Feature vectors forming a sample to be decided
Figure BDA0002938037790000123
And inputting the decision value g (phi) into a classification decision function to obtain a decision value g (phi) of a sample to be decidedu) (ii) a The classification decision function is obtained by training the initial classification decision function;
s4, according to the decision value g (phi)u) Assessment of lung function: if g (phi)u) If the lung function is more than 0, the sample is evaluated to be strong, otherwise, the lung function is weak.
The training processing is carried out on the initial classification decision function, and the method comprises the following steps:
establishing a linear classifier with a decision function of g (phi) wTφ+w0(ii) a Wherein g (φ) is a decision value of sample φ, [ en ═max，td]TIs the feature vector of the sample, w ═ w1，w2]TAnd w0Is the undetermined parameter of the classifier;
the method comprises the steps of taking cough tone signals which are acquired by a hospital and have strong lung functions and weak lung functions and are used for patients to cough forcibly as training samples to form a sample set, wherein 25 cough tone signal samples of the patients with strong lung functions are obtained, and the labels of the samples are set to be 1; 25 cough tone signal samples of patients with weak lung function, and the labels of the samples are set to be-1;
for all samples in the sample set, the respective short-time energy maximum en is obtained according to step S2maxAnd the root mean square curve fall time tdThe eigenvector phi forming the corresponding sample is ═ enmax，td]T(ii) a The feature vector set corresponding to the cough sound signal sample of the patient with strong memory lung function is
Figure BDA0002938037790000124
Figure BDA0002938037790000125
The feature vector set corresponding to the cough sound signal sample of the patient with weak lung function is
Figure BDA0002938037790000126
Wherein
Figure BDA0002938037790000131
Is the characteristic vector of the ith patient with strong lung function,
Figure BDA0002938037790000132
a feature vector for the ith patient with weak lung function;
taking the feature vector in the feature vector set as input, taking a sample label value corresponding to the feature vector as expected output, and training a linear classifier to obtain a classification decision function
Figure BDA0002938037790000133
Wherein, w*And
Figure BDA0002938037790000134
respectively obtaining the optimal projection direction and the optimal threshold value obtained by training; the method comprises the following steps:
1) calculating the mean-class vector score of two classes of samplesBiede
Figure BDA0002938037790000135
2) Calculating the in-class dispersion matrix of the two types of samples respectively to obtain
Figure BDA0002938037790000136
3) Calculating the dispersion matrix in the general class
Figure BDA0002938037790000137
4) Calculating the optimal projection direction
Figure BDA0002938037790000138
Wherein,
Figure BDA0002938037790000139
5) calculating the center of the two samples after projection
Figure BDA00029380377900001310
6) Selecting a classification threshold of
Figure BDA00029380377900001311
7) The final Fisher classification decision function is:
Figure BDA00029380377900001312
it should be noted that the selection of the classification threshold is not limited to the above calculation formula, and may be selected manually or by other calculation methods according to the trained result.
In order to realize the above-mentioned lung function assessment algorithm for patients with cervical and spinal cord injuries, the present embodiment provides a lung function assessment device for patients with cervical and spinal cord injuries, including:
the signal acquisition module is used for acquiring a cough sound signal to be decided;
a preprocessing module for deconvoluting the cough tone signal to reflect the lung exhalationExcitation source component y of the air flowi(k) (ii) a Calculating the excitation source component yi(k) Short time energy maximum of
Figure BDA00029380377900001313
And the fall time of the root mean square curve
Figure BDA00029380377900001314
A classification decision module for maximizing short-time energy
Figure BDA0002938037790000141
And the fall time of the root mean square curve
Figure BDA0002938037790000142
Feature vectors forming a sample to be decided
Figure BDA0002938037790000143
And inputting the decision value g (phi) into a classification decision function to obtain a decision value g (phi) of a sample to be decidedu) (ii) a The classification decision function is obtained by training the initial classification decision function;
an evaluation module for evaluating the decision value g (phi)u) Assessment of lung function: if g (phi)u) If the lung function is more than 0, the sample is evaluated to be strong, otherwise, the lung function is weak.
Example two
The present embodiment is an algorithm for evaluating lung function strength of a cervical spinal cord injury patient, and the workflow thereof is shown in fig. 4, and the difference from the first embodiment is that: in this embodiment, in step S1, before framing and windowing the cough sound signal, the end point detection is performed on the cough sound signal, and then framing and windowing are performed.
The remaining steps of this embodiment are the same as those of the first embodiment.
EXAMPLE III
The present embodiment is a storage medium, wherein the storage medium stores a computer program, and the computer program, when executed by a processor, causes the processor to execute the lung function assessment algorithm for a cervical and spinal cord injury patient according to the first embodiment or the second embodiment.
Example four
The present embodiment is a computing device, which includes a processor and a memory for storing a program executable by the processor, and when the processor executes the program stored in the memory, the processor implements the lung function assessment algorithm for a cervical and spinal cord injury patient according to the first embodiment or the second embodiment.
The above embodiments are preferred embodiments of the present invention, but the present invention is not limited to the above embodiments, and any other changes, modifications, substitutions, combinations, and simplifications which do not depart from the spirit and principle of the present invention should be construed as equivalents thereof, and all such changes, modifications, substitutions, combinations, and simplifications are intended to be included in the scope of the present invention.

Claims (10)

1. An algorithm for evaluating lung function of a patient with cervical spinal cord injury is characterized in that: the method comprises the following steps:
s1, acquiring a cough tone signal to be decided;
s2, deconvoluting the cough tone signal to reflect the exciting source component y of the lung expired air flowi(k) (ii) a Calculating the excitation source component yi(k) Short time energy maximum of
Figure FDA0002938037780000011
And the fall time of the root mean square curve
Figure FDA0002938037780000012
S3 maximum energy in short time
Figure FDA0002938037780000013
And the fall time of the root mean square curve
Figure FDA0002938037780000014
Feature vectors forming a sample to be decided
Figure FDA0002938037780000015
And inputting the decision value g (phi) into a classification decision function to obtain a decision value g (phi) of a sample to be decidedu) (ii) a The classification decision function is obtained by training the initial classification decision function;
s4, according to the decision value g (phi)u) Assessment of lung function: if g (phi)u)>0, the sample is evaluated as strong in lung function, otherwise, the lung function is weak.
2. The cervical and spinal cord injury patient lung function assessment algorithm of claim 1, wherein: the step S2 includes the following sub-steps:
s21, framing and windowing the cough sound signal; wherein, the frame length is set to wlen, the frame shift is set to inc, the window function length win is set to be the same as the frame length, the ith frame signal is marked as xi(k) K is 1,2, wlen, i is 1,2, l, l is total frame number after framing, and wlen, inc, i, k, l are positive integers;
s22, for the ith frame signal xi(k) Deconvoluting an excitation sequence y reflecting the pulmonary expiratory airflow by linear predictive codingi(k)；
S23, calculating the excitation source frame sequence yi(k) Short-time energy en ofiAnd searching for a short-term energy maximum en in all excitation source framesmax；
S24, calculating the excitation source frame sequence yi(k) Root mean square value of
Figure FDA0002938037780000016
To pair
Figure FDA0002938037780000017
Plotting a root mean square curve of the cough sound excitation source signal and calculating the falling time t of the root mean square curved。
3. The cervical and spinal cord injury patient lung function assessment algorithm of claim 2, wherein: in the step S23, the source frame sequence y is excitedi(k) Short-time energy en ofiThe calculation formula is as follows:
Figure FDA0002938037780000021
4. the cervical and spinal cord injury patient lung function assessment algorithm of claim 2, wherein: in the step S24, the source frame sequence y is excitedi(k) Root mean square value of (Yrms)iThe calculation formula is as follows:
Figure FDA0002938037780000022
time of fall t of root mean square curvedThe time required for the rms curve amplitude to first drop to 0.25 times the maximum amplitude from the time point of the maximum amplitude of the rms curve to that time point is taken.
5. The cervical and spinal cord injury patient lung function assessment algorithm of claim 2, wherein: in S21, the cough sound signal is subjected to end point detection, and then is framed and windowed.
6. The cervical and spinal cord injury patient lung function assessment algorithm of claim 2, wherein: in the step S21, the frame length wlen is set to a value according to a frame duration of 20-30 ms, the frame shift inc is set to a value of 0.25-0.75 times the frame length, and the window function is a Hanning window;
in the step S22, the solution of the equation set in the linear predictive coding is implemented by using a Levinson-Durbin algorithm.
7. The cervical and spinal cord injury patient lung function assessment algorithm of claim 1, wherein: the training process of the initial classification decision function refers to:
establishing a linear classifier with a decision function of g (phi) ═ wTφ+w0(ii) a Wherein g (φ) is a decision value of sample φ, φ ═ enmax,td]TIs the feature vector of the sample, w ═ w1,w2]TAnd w0Is the undetermined parameter of the linear classifier;
collecting cough tone signals of patients with strong lung function and weak lung function with neck and spinal cord injuries as training samples to form a sample set, wherein the number of the cough tone signal samples of the patients with strong lung function is m, and the labels of the samples are set to be 1; the number of cough tone signal samples of the patient with weak lung function is n, and the label of each sample is set to be-1;
calculating the short-time energy maximum value en of the cough tone signals in the sample set according to the step S2maxAnd the fall time t of the root mean square curvedThe eigenvector phi forming the corresponding sample is ═ enmax,td]T(ii) a The feature vector set corresponding to the cough sound signal sample of the patient with strong memory lung function is
Figure FDA0002938037780000031
Figure FDA0002938037780000032
The feature vector set corresponding to the cough sound signal sample of the patient with weak lung energy is
Figure FDA0002938037780000033
Wherein
Figure FDA0002938037780000034
Is the characteristic vector of the ith patient with strong lung function,
Figure FDA0002938037780000035
a feature vector for the ith patient with weak lung function;
training a linear classifier to obtain a classification decision function by taking the feature vectors in the feature vector set as input and the sample label values corresponding to the feature vectors as expected output
Figure FDA0002938037780000036
Wherein, w*And
Figure FDA0002938037780000037
respectively the best projection direction and the best threshold value obtained by training.
8. A device for evaluating lung function of a patient with cervical spinal cord injury is characterized in that: the method comprises the following steps:
the signal acquisition module is used for acquiring a cough sound signal to be decided;
a preprocessing module for deconvolving the cough tone signal to an excitation component y reflecting the pulmonary exhaled air flowi(k) (ii) a Calculating the excitation source component yi(k) Short time energy maximum of
Figure FDA0002938037780000038
And the fall time of the root mean square curve
Figure FDA0002938037780000039
A classification decision module for maximizing short-time energy
Figure FDA00029380377800000310
And the fall time of the root mean square curve
Figure FDA00029380377800000311
Feature vectors forming a sample to be decided
Figure FDA00029380377800000312
And inputting the decision value g (phi) into a classification decision function to obtain a decision value g (phi) of a sample to be decidedu) (ii) a The classification decision function is obtained by training the initial classification decision function;
an evaluation module for evaluating the decision value g (phi)u) Assessment of lung function: if g (phi)u)>0, the sample is evaluated as strong in lung function, otherwise, the lung function is weak.
9. A storage medium, wherein the storage medium stores a computer program which, when executed by a processor, causes the processor to execute the cervical and spinal cord injury patient lung function assessment algorithm of any one of claims 1 to 7.
10. A computing device comprising a processor and a memory for storing processor-executable programs, wherein the processor, when executing the programs stored in the memory, implements the cervical and spinal cord injury patient lung function assessment algorithm of any one of claims 1 to 7.
CN202110168260.XA 2021-02-07 2021-02-07 Lung function assessment algorithm, device, medium and equipment for cervical and spinal cord injury patient Active CN112992337B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110168260.XA CN112992337B (en) 2021-02-07 2021-02-07 Lung function assessment algorithm, device, medium and equipment for cervical and spinal cord injury patient

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110168260.XA CN112992337B (en) 2021-02-07 2021-02-07 Lung function assessment algorithm, device, medium and equipment for cervical and spinal cord injury patient

Publications (2)

Publication Number Publication Date
CN112992337A CN112992337A (en) 2021-06-18
CN112992337B true CN112992337B (en) 2022-05-24

Family

ID=76348961

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110168260.XA Active CN112992337B (en) 2021-02-07 2021-02-07 Lung function assessment algorithm, device, medium and equipment for cervical and spinal cord injury patient

Country Status (1)

Country Link
CN (1) CN112992337B (en)

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7539533B2 (en) * 2006-05-16 2009-05-26 Bao Tran Mesh network monitoring appliance
US11097122B2 (en) * 2015-11-04 2021-08-24 The Regents Of The University Of California Magnetic stimulation of the spinal cord to restore control of bladder and/or bowel
US11864880B2 (en) * 2017-12-21 2024-01-09 The University Of Queensland Method for analysis of cough sounds using disease signatures to diagnose respiratory diseases
US20210219925A1 (en) * 2018-06-14 2021-07-22 Srados Labs, Inc. Apparatus and method for detection of physiological events
US20200138337A1 (en) * 2018-11-06 2020-05-07 The Regents Of The University Of Colorado, A Body Corporate Non-Contact Breathing Activity Monitoring And Analyzing System Through Thermal On Projection Medium Imaging
US11948690B2 (en) * 2019-07-23 2024-04-02 Samsung Electronics Co., Ltd. Pulmonary function estimation
CN111179967B (en) * 2019-12-17 2022-05-24 华南理工大学 Algorithm, medium and equipment for linear classification of true and false cough sounds of patients with cervical and spinal cord injuries
CN111261300A (en) * 2020-01-16 2020-06-09 复旦大学附属中山医院 Method for acquiring normal predicted value of lung function parameter

Also Published As

Publication number Publication date
CN112992337A (en) 2021-06-18

Similar Documents

Publication Publication Date Title
Sankur et al. Comparison of AR-based algorithms for respiratory sounds classification
Jin et al. Adventitious sounds identification and extraction using temporal–spectral dominance-based features
CN102469954B (en) Method and apparatus of determining exhaled nitric oxide
Melcon et al. Age and laryngeal airway resistance during vowel production
Ternström et al. The voice range profile: Its function, applications, pitfalls and potential
WO2011010384A1 (en) Sleep apnea syndrome examination device and program
Dhupati et al. A novel drowsiness detection scheme based on speech analysis with validation using simultaneous EEG recordings
CN112822976B (en) Estimating lung volume by speech analysis
CN105448291A (en) Parkinsonism detection method and detection system based on voice
US20140066724A1 (en) System and Methods for Evaluating Vocal Function Using an Impedance-Based Inverse Filtering of Neck Surface Acceleration
McKenna et al. Magnitude of neck-surface vibration as an estimate of subglottal pressure during modulations of vocal effort and intensity in healthy speakers
JP7291319B2 (en) Evaluation method and apparatus for difficult airway based on speech technique by machine learning
Yiu et al. Reliability and applicability of aerodynamic measures in dysphonia assessment
CN112992337B (en) Lung function assessment algorithm, device, medium and equipment for cervical and spinal cord injury patient
Dar et al. Lung anomaly detection from respiratory sound database (sound signals)
CN118133209A (en) Data processing method based on breath trainer data analysis
Kriboy et al. A novel method for obstructive sleep apnea severity estimation using speech signals
Himeshima et al. Detection of abnormal lung sounds taking into account duration distribution for adventitious sounds
CN117198340B (en) Dysarthria correction effect analysis method based on optimized acoustic parameters
Stemple Voice research: so what? A clearer view of voice production, 25 years of progress; the speaking voice
CN116530943B (en) Anesthesia depth detection device based on blood gas data
CN111179967B (en) Algorithm, medium and equipment for linear classification of true and false cough sounds of patients with cervical and spinal cord injuries
CN116469416A (en) Breathing sound signal identification method and system
JPH0998964A (en) Coughing measurement apparatus
CN111228751B (en) Respiratory quantitative training method and device

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant