CN114145722B - Pulse pathological feature mining method for pancreatitis patients - Google Patents

Abstract

The invention provides a method for mining pulse pathological characteristics of pancreatitis patients, which comprises the following steps: obtaining pulse signal curve samples marked with labels, wherein the labels are used for distinguishing the pulse signal curve of the sample as the pulse signal curve of a normal person or a pancreatitis patient, then carrying out periodic division on the signal curve, extracting basic time domain characteristics, stability indexes and double peak index characteristics to form input characteristic vectors, inputting the input characteristic vectors into a classification model, supervising the input characteristic vectors by the labels corresponding to all the samples, carrying out training, obtaining a trained classification model, and then diagnosing that the person to be diagnosed is ill. The extracted dual-peak index represents two peaks and an obvious concave arc structure in a single-period curve, and the stability index is used for representing the change rate of the latter half of the single-period curve, so that pancreatitis patients can be well pointed, case feature vectors can be provided for subsequent training classification models, and the obtained trained classification models have positive significance on classification patients.

Description

A pulse pathological feature mining method for patients with pancreatitis

technical field

The invention belongs to the technical field of traditional Chinese medicine pulse collection system, and in particular relates to a pulse pathological feature mining method for patients with pancreatitis.

Background technique

Traditional Chinese medicine occupies a very important position in the world medical system. Chinese pulse diagnosis has been summed up in the medical field for more than two thousand years, and has accumulated rich experience. One of the representative diagnostic methods. At present, TCM diagnosis mainly relies on TCM physicians to touch the patient's veins with their hands. However, the difference between subject and object in this diagnosis method has a great influence on the diagnosis results. There is no quantitative data in the diagnosis of pulse condition, which lacks objectivity. With the development of sensors and artificial intelligence technology, computer-aided medicine has been widely used.

Accurately analyzing and diagnosing pulse conditions through more human body information, and collecting multi-modal information is of great significance for pulse condition diagnosis in traditional Chinese medicine.

The blood flows through the whole body of the human body, and the physiological changes of the human body can affect the change of the blood vessel pressure in the arm arteries. In the theory of traditional Chinese medicine, the pulse contains various information of the human body. Pulse is a quasi-periodic signal. After the pulse signal is preprocessed and divided into cycles, its single-cycle pulse can reflect a variety of human physiological and pathological information, such as age, gender, physical condition, and health status. However, at present, doctors of traditional Chinese medicine use manual measurement and only rely on experience and pulse diagnosis when making a diagnosis, which causes the uncertainty of the diagnosis result.

Contents of the invention

The technical problem to be solved by the present invention is to provide a pulse pathological feature mining algorithm for patients with pancreatitis in view of the deficiencies in the prior art above. The characteristic design formula, including stability index and double-peak index, compared with non-pancreatitis patients, the pulse condition of pancreatitis patients has two peaks and obvious concave arc structure, the above two characteristics can well describe the single pulse For the stability of the periodic signal, the above two pathological features are fused with the basic time-domain features to train the classification model to achieve the preliminary classification of patients with suspected pancreatitis, and the classification effect is good.

In order to solve the above-mentioned technical problems, the technical solution adopted in the present invention is: a pathological feature mining method for patients with pancreatitis, characterized in that the method includes:

Obtaining a labeled sample pulse signal curve, the label is used to distinguish whether the individual corresponding to the sample pulse signal curve is a non-pancreatitis patient or a pancreatitis patient;

Divide the sample pulse signal curve by period to obtain each single-period curve;

Extract the basic time-domain features of each single-period curve;

The stability index and double-peak index of each single-cycle curve were extracted as the case characteristics of pancreatitis corresponding to the single-cycle curve. The double-peak index indicated two peaks and obvious concave arc structure in the single-cycle curve; the stability index was used to represent rate of change in the second half of the single-period curve;

The basic time-domain features of the pulse signal curves of all samples and the case features of pancreatitis form the input feature vector, which is input to the classification model and supervised with the labels corresponding to all samples for training to obtain a trained classification model.

Preferably, said tags are divided into patient tags and non-patient tags;

The trained classification model also includes:

Obtain the pulse signal curve of the person to be diagnosed;

In the pulse signal curve of the person to be diagnosed, the basic time-domain features of all single-cycle curves and pancreatitis case features are input into the trained classification model to obtain a diagnostic label set; each label in the diagnostic label set is related to One-to-one correspondence of single-cycle curves;

If the occurrence frequency of the patient label in the diagnostic label set is greater than the preset threshold, it is determined that the person to be diagnosed is a suspected pancreatitis patient.

Preferably, extracting the stability index of each single-period curve is specifically: by selecting a plurality of sampling points in the single-period curve, and calculating the variance of each sampling point through the first formula, the variance of all sampling points For representing the stability index of the pulse single-cycle signal of each single-cycle curve, the first formula is as follows:

In the first formula, n represents the number of sampling points in the single-cycle curve, N represents the number of sampling points in the entire cycle of the sample pulse signal curve, S _F represents the sampling frequency, t represents the initial position of the sampling point, and D _i represents the variance of the sampling point at i, where i takes a positive integer from 0 to N, the parameter _Xi represents the absolute coordinate value of the sampling point at the abscissa i, x _i represents the abscissa of the sampling point at i, and y _i represents the The vertical coordinate of the sampling point.

Preferably, the sampling frequency is twice the beating frequency of the human pulse.

Preferably, extracting the double-peak index of each single-period curve is specifically as follows: selecting multiple sampling points in the single-period curve, and obtaining the mean value of the ordinate values of each sampling point of non-pancreatitis patients through calculation, and calculating it through the second formula The sum of the difference between the ordinate value of the sampling point between the two peaks in the single-cycle curve and the mean value of the ordinate value of the same sampling point in non-pancreatitis patients, the sum is used as the double peak index; the second formula is as follows:

In the second formula, Y represents the data set of the ordinate of the sampling point, N represents the number of sampling points in the entire cycle of the sample pulse signal curve, n represents the number of sampling points in the single-cycle curve, and t represents the single-cycle pulse signal The point with the highest ordinate value in the curve, T represents the point with the second highest ordinate value in the single-cycle curve, S _F represents the sampling frequency, A represents the average value of the ordinate value of each sampling point in a normal person, _{and Bi} represents pancreatitis patient i The sum of the difference between the ordinate value of the sampling point and A.

Preferably, the basic time-domain characteristics of each single-cycle pulse signal curve include pulse first peak height, second peak height, third peak height, time interval from first peak to second peak, second peak to third peak The time interval, the time interval from the pulse start point to the first trough, the time interval from the first trough to the second trough, and the time interval from the second trough to the pulse start point.

Preferably, the pulse signal curve of the person to be diagnosed is a pulse signal curve of at least one single cycle.

Preferably, the classification model is trained using the SVM kernel method.

Compared with the prior art, the present invention has the following advantages:

1. Based on the principles of traditional Chinese medicine, the present invention has discovered the unique pathological features of the pulse images of patients with pancreatitis compared with normal people. In order to visualize these features and differences, two algorithms have been designed to combine the two pathological features of pancreatitis. The characteristics are reflected by the data. The two pathological characteristics are the stability index and the double peak index, combined with the basic time domain characteristics of the single-cycle pulse signal curve to train the classification model, and the trained classification model is used to initially diagnose whether the person to be diagnosed is sick or not. .

2. It is the first time that the present invention uses a computer to detect pancreatitis based on the principle of traditional Chinese medicine. The pulse signal curve of the person to be diagnosed is input into the trained classification model to determine whether the person to be diagnosed is a suspected pancreatitis patient, and the classification is accurate The rate is as high as over 95%, and the accuracy for young patients is even close to 100%. Elderly subjects are older, the elasticity of blood vessels will decrease, and the peak structure of typical pulse image features will become less obvious, so the diagnostic accuracy is lower than that of young patients. This experimental model is suitable for pancreatitis and other chronic diseases. Disease detection has a very good experimental effect, and the accuracy rate is extremely high.

The technical scheme of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

Description of drawings

Fig. 1 is a flowchart of training a classification model according to an embodiment of the present invention.

Figure 2 is a single cycle pulse signal curve of a normal person.

Figure 3 is a pulse signal curve of a patient with pancreatitis.

Fig. 4 is the basic time-domain characteristics of the single-period pulse signal curve of the present invention.

Detailed ways

This embodiment discloses a pulse pathological feature mining method for patients with pancreatitis, the method is:

S101. Obtain a tagged sample pulse signal curve. The label is used to distinguish the pulse signal curve as the pulse signal curve of a non-pancreatitis patient and a pancreatitis patient. Inflammation patient label is "0";

The specific steps to obtain the labeled sample pulse signal curve are as follows: collect the pulse data of multiple non-pancreatitis patients and multiple pancreatitis patients, preprocess the collected pulse data respectively, and then take the collection time as the abscissa, and take the pulse The pulse amplitude is the ordinate, and the pulse signal curve is drawn to obtain the labeled sample pulse signal curve; the preprocessing of the pulse data is based on the existing technology, such as using the Butterworth algorithm for signal filtering, and the filtered The baseline drift is removed from the signal, and the signal is processed again with a smoothing function to make the pulse signal curve smooth;

S102. Division of pulse data cycles: human pulse is a quasi-periodic signal, and each pulse cycle has a similar cycle, amplitude and waveform, but these characteristics vary from person to person, so it is used for non-pancreatitis patients and pancreatitis patients The drawn pulse signal curve is divided into periods to obtain each single-cycle curve, and the pulse curve characteristics representing the curve collector can be obtained by studying each single-cycle pulse signal curve;

S103. Extract the basic time-domain features of each single-period curve; the basic time-domain features include the first peak height, the second peak height, the third peak height, and the time interval from the first peak to the second peak of the single-period curve , the time interval from the second peak to the third peak, the time interval from the pulse onset to the first trough, the time interval from the first trough to the second trough, and the time interval from the second trough to the pulse onset;

S104, extracting the stability index of each single-cycle curve, the stability index is used to represent the rate of change in the second half of the single-cycle curve, and has classification significance for distinguishing pancreatitis patients from non-pancreatitis patients;

S105, extracting the double-peak index of each single-period curve, the double-peak index indicates two peaks and obvious concave arc structure in the single-cycle curve; since the double-peak index of normal people is greater than the double-peak index of patients with pancreatitis,

Then extract the stability index and double peak index of each single-cycle curve as the case characteristics of pancreatitis corresponding to the single-cycle curve;

S106. The basic time-domain features of the pulse signal curves of all samples and the case features of pancreatitis form an input feature vector, input it into the classification model and supervise it with the labels corresponding to all samples, use the SVM kernel method to train the classification model, and obtain a well-trained classification model.

In this embodiment, the non-pancreatitis patient is a healthy person without any disease; the collection of the pulse data is to measure the pulse of the non-pancreatitis patient or the pancreatitis patient through a similar instrument such as a pulse sensor, and collect Collect, and then upload the collected pulse data to the computer for disease detection research. During the collection, 57 non-pancreatitis patients and 7 pancreatitis patients were selected. The pulse data were measured at several time periods every day, and measured for several consecutive days. The pulse data were provided by the 211th Hospital of the Chinese People's Liberation Army.

In this embodiment, extracting the stability index of each single-period curve is specifically: by selecting a plurality of sampling points in the single-period curve, and calculating the variance of each sampling point through the first formula, all the sampling points The variance is used to represent the stability index of the pulse single-cycle signal of each single-cycle curve, and the first formula is as follows:

In the first formula, n represents the number of sampling points in the single-cycle curve, N represents the number of sampling points in the entire cycle of the sample pulse signal curve, S _F represents the sampling frequency, t represents the initial position of the sampling point, and D _i represents the variance of the sampling point at i, where i takes a positive integer from 0 to N, the parameter _Xi represents the absolute coordinate value of the sampling point at the abscissa i, x _i represents the abscissa of the sampling point at i, and y _i represents the The vertical coordinate of the sampling point.

As shown in Figure 2 and Figure 3, compared with the pulse of normal people, the pulse condition (within the considered area) of patients with pancreatitis has two peaks and obvious concave arc structure, which can well describe the pulse monotony. The stability of the periodic signal, and can describe the number of peaks and the degree of concave arc in the pulse. Therefore, in this embodiment, extracting the double-peak index of each single-period curve is specifically: select multiple sampling points in the single-period curve, and obtain the mean value of the ordinate value of each sampling point in non-pancreatitis patients by calculation, and obtain The second formula calculates the sum of the difference between the sampling point ordinate value between two peaks in the single-cycle curve and the mean value of the same sampling point ordinate value of non-pancreatitis patients, and this sum is used as a double peak index; the second formula as follows:

In the second formula, Y represents the data set of the ordinate of the sampling point, N represents the number of sampling points in the entire cycle of the sample pulse signal curve, n represents the number of sampling points in the single-cycle curve, and t represents the single-cycle pulse signal The point with the highest ordinate value in the curve, T represents the point with the second highest ordinate value in the single-cycle curve, S _F represents the sampling frequency, A represents the average value of the ordinate value of each sampling point in a normal person, _{and Bi} represents pancreatitis patient i The sum of the difference between the ordinate value of the sampling point and A.

As shown in Figure 4, the basic time-domain characteristics of a single-cycle pulse signal include: from the starting point of the pulse, the first peak height h ₁ , the second peak height h ₂ , the third peak height h ₃ , the first peak to the second The time interval t _a of the peak, the time interval t _b from the second peak to the third peak, the time interval t ₂ -t ₁ from the pulse starting point to the first trough, the time interval t ₃ -t from the first trough to the second trough ₂ , the time interval t ₃ -t ₁ from the second trough to the starting point of the pulse.

In this embodiment, the trained classification model also includes:

S201. Obtain the pulse signal curve of the person to be diagnosed;

S202. In the pulse signal curve of the person to be diagnosed, input the basic time-domain features of all single-cycle curves and pancreatitis case features into the trained classification model to obtain a diagnostic label set; One-to-one correspondence between labels and single-period curves;

S203. If the occurrence frequency of the patient label in the diagnostic label set is greater than the preset threshold, determine that the person to be diagnosed is a suspected pancreatitis patient.

In this embodiment, the pulse signal curve of the person to be diagnosed is a pulse signal curve of at least one single cycle.

After selecting diagnosed patients and non-patients, the trained classification model is verified, and the verification results show that the accuracy rate of the final diagnosis of pancreatitis patients is more than 95%, especially for young patients. The accuracy is even close to 100%. Elderly subjects are older, the elasticity of blood vessels will decrease, and the peak structure of the typical pulse image features will become less obvious, so the diagnostic accuracy is lower than that of young patients. This experimental model has a very good effect on pancreatitis. Experimental results, and the accuracy is extremely high.

The above descriptions are only preferred embodiments of the present invention, and do not limit the present invention in any way. All simple modifications, changes and equivalent changes made to the above embodiments according to the technical essence of the invention still belong to the protection scope of the technical solution of the invention.

Claims (6)

1. A pathological feature mining method for patients with pancreatitis, characterized in that the method comprises: Obtaining a labeled sample pulse signal curve, the label is used to distinguish whether the individual corresponding to the sample pulse signal curve is a non-pancreatitis patient or a pancreatitis patient; Divide the sample pulse signal curve by period to obtain each single-period curve; Extract the basic time-domain features of each single-period curve; The stability index and double-peak index of each single-cycle curve were extracted as the case characteristics of pancreatitis corresponding to the single-cycle curve. The double-peak index indicated two peaks and obvious concave arc structure in the single-cycle curve; the stability index was used to represent rate of change in the second half of the single-period curve; The basic time-domain features of the pulse signal curves of all samples and the case features of pancreatitis are used to form the input feature vector, which is input to the classification model and supervised with the labels corresponding to all samples for training to obtain a trained classification model; Extracting the stability index of each single-period curve is specifically: by selecting multiple sampling points in the single-period curve, and calculating the variance of each sampling point through the first formula, the variance of all the sampling points is used to represent The stability index of the pulse single-cycle signal of each single-cycle curve, the first formula is as follows: ; In the first formula, n represents the number of sampling points in the single-cycle curve, N represents the number of sampling points in the entire cycle of the sample pulse signal curve, S _F represents the sampling frequency, and t represents the initial position of the sampling point, D _i represents the variance of the sampling point at i, where i takes a positive integer from 0 to N, the parameter Xi represents the absolute coordinate value of the sampling point at the abscissa i, x _i represents the abscissa of the sampling point _at i, and y _i represents the The vertical coordinate of the sampling point; Extracting the double peak index of each single-cycle curve is specifically: select multiple sampling points in the single-cycle curve, calculate the mean value of the ordinate value of each sampling point in non-pancreatitis patients, and calculate the single-cycle curve through the second formula The sum of the difference between the sampling point ordinate value and the mean value of the same sampling point ordinate value of the non-pancreatitis patient between the two peaks, the sum is used as the double peak index; the second formula is as follows: ; In the second formula, Y represents the data set of the ordinate of the sampling point, N represents the number of sampling points in the entire cycle of the sample pulse signal curve, n represents the number of sampling points in the single-cycle curve, and t represents the single-cycle pulse signal The point with the highest ordinate value in the curve, T represents the point with the second highest ordinate value in the single-cycle curve, S _F represents the sampling frequency, A represents the average value of the ordinate value of each sampling point in a normal person, B _i represents pancreatitis patient i The sum of the difference between the ordinate value of the sampling point and A. 2. A method for mining pathological features of pancreatitis patients according to claim 1, wherein the tags are divided into patient tags and non-patient tags; The trained classification model also includes: Obtain the pulse signal curve of the person to be diagnosed; In the pulse signal curve of the person to be diagnosed, the basic time-domain features of all single-cycle curves and pancreatitis case features are input into the trained classification model to obtain a diagnostic label set; each label in the diagnostic label set is related to One-to-one correspondence of single-cycle curves; If the occurrence frequency of the patient label in the diagnostic label set is greater than the preset threshold, it is determined that the person to be diagnosed is a suspected pancreatitis patient. 3. A pathological feature mining method for pancreatitis patients according to claim 1 or 2, wherein the sampling frequency is twice the beating frequency of the human pulse. 4. A kind of pathological feature mining method for pancreatitis patients according to claim 1 or 2, characterized in that, the basic time-domain features of each single-cycle pulse signal curve include pulse first peak height, second peak height , the height of the third peak, the time interval from the first peak to the second peak, the time interval from the second peak to the third peak, the time interval from the pulse onset to the first trough, the time interval from the first trough to the second trough, and The time interval from the second trough to the pulse start point. 5 . The pathological feature mining method for pancreatitis patients according to claim 2 , wherein the pulse signal curve of the person to be diagnosed is at least one single-period pulse signal curve. 6. A method for mining pathological features of pancreatitis patients according to claim 1, wherein the classification model is trained using the SVM kernel method.