CN109949932A - A method for judging the severity of essential tremor based on machine learning - Google Patents

Abstract

The invention relates to the technical field of computer data processing, and more particularly, to a method for judging the disease degree of essential tremor based on machine learning. The regression model for identifying the degree of essential tremor based on machine learning, for the trained regression model, the tester only needs to write some Chinese characters on the electronic tablet to calculate the degree of the tester suffering from essential tremor.

Description

A method for judging the severity of essential tremor based on machine learning

technical field

The invention relates to the technical field of computer data processing, and more particularly, to a method for judging the disease degree of essential tremor based on machine learning.

Background technique

Essential Tremor (ET) is a common movement disorder, also known as essential tremor and benign tremor. About 60% of patients with essential tremor have a family history, so it is also called hereditary tremor or familial tremor. A number of studies have shown that the average age of onset of the disease is 45 years old, the incidence rate is 0.14% to 5.10%, and the incidence rate of people over 70 years old is as high as 12.6%. Its clinical symptoms are tremor, which is manifested as postural or action tremor. Postural tremor is the most obvious tremor when maintaining a certain posture, and patients with action tremor rarely experience tremor at rest. The tremor usually begins on one hand and gradually spreads to the entire upper extremity and the contralateral upper extremity, up to the muscles of the head and throat. The tremor frequency is generally 4 Hz to 12 Hz. Although essential tremor is regarded as a benign tremor, severe tremor in some patients can hinder fine movements of the hand, affect pronunciation when the laryngeal muscle is involved, and even lose the ability to live independently. Parkinson's disease (PD) and essential tremor have similar features in some cases, so it is sometimes difficult to determine at the time of diagnosis, in which case SPECT-DAT scan test can be performed on patients, but this test is very expensive. In this context, systems that collect patient behavioral information for analysis will be the first choice, as most patients find speech analysis, handwriting analysis, or drawing analysis less stressful, and these techniques are low-cost and less demanding on collection equipment . In addition, there is no quantitative indicator for judging the degree of illness of a patient. At present, only the doctor's judgment is used to give a vague description.

The traditional note analysis is done offline, because only the strokes on the paper can be analyzed. The electronic tablet can collect handwriting data and time information at the same time, and can also collect the pressure on the surface of the device, the angle between the stylus and the tablet. , and even capture the trajectory of the stylus moving in the air. The integrated analysis of these data can be very good for the diagnosis of essential tremor cases. For essential tremor, medical professionals typically use handwriting tasks to diagnose.

SUMMARY OF THE INVENTION

To this end, it is necessary to provide a method for judging the disease degree of essential tremor based on machine learning, and a regression model for identifying the disease degree of essential tremor based on machine learning. Writing some Chinese characters on the electronic tablet can calculate the degree of essential tremor of the test subject.

To achieve the above object, the present invention adopts the following technical solutions:

A method for judging the severity of essential tremor based on machine learning, which includes the following steps:

Step 1. Select N Chinese characters from the Chinese character library as the test words that the tester needs to write when evaluating the essential tremor disease, wherein N≥1;

Step 2: Collect the data of healthy people and patients with essential tremor when they use the electronic pen to write the test words on the electronic handwriting board as sample data;

Step 3: Preprocessing the collected sample data;

Step 4: Organize the sample features into the same structure for training;

Step 5, train the SVR regression model, and the regression result is the disease degree coefficient; save the regression model data after training;

Step 6: Collect the data of the test words written by the tester on the electronic handwriting board, process the collected data according to Steps 3 and 4, and input it into the SVR regression model trained in Step 5 to obtain the disease degree coefficient.

This technical solution is further optimized. The sample data collected in step 2 includes the time when the electronic pen does not touch the tablet on the electronic tablet, the time when the electronic pen contacts the tablet on the electronic tablet, the coordinates X and Y of the pen tip, and the distance between the pen tip and the tablet. The angle O of the horizontal plane, the angle A of the pen tip and the vertical plane, and the pressure of the pen tip on the electronic tablet.

This technical solution is further optimized, and the sample data preprocessing in step 3 specifically includes checking whether the collected data has errors, and if so, discarding the erroneous data. For example, if there is an interruption in the tester's writing, or there is obvious data that does not conform to common sense when recording, if there is any, the group of data is directly discarded. To normalize the temporal features and map similar data to decimals between -1 and 1, the data can be normalized using the following formula:

Perform Fourier transform on the pen tip X-axis data, Y-axis data, the angle O between the pen tip and the horizontal plane, and the angle A between the pen tip and the vertical plane to obtain the first 50hz frequency domain data; calculate the maximum and minimum values of the recorded pressure data , mean, median, and variance.

This technical solution is further optimized, and the fourth step is specifically, the structure of the preprocessed features is as follows:

X=[T _up T _down f _X f _Y f _O f _A F _max F _min F _ave F _mid σ _F ² ] ^T

where T _up represents the time that the electronic pen does not contact the tablet after normalization, and T _down represents the time that the electronic pen touches the tablet after normalization, f _X and f _Y represents the frequency vector of the pen tip X-axis and Y-axis data, f _O and f _A represent the frequency vector of the angle O between the pen tip and the horizontal plane, and the angle A between the pen tip and the vertical plane, F _max , F _min , F _ave , F _mid , σ _F ² represent the maximum, minimum, mean, median and variance of the pressure data, respectively. In addition, the doctor also needs to give the degree of the patient's disease.

This technical solution is further optimized, and the SVR regression model in step 5 is:

The optimization objective of establishing the SVR regression model is:

ξ _i ^∨ ≥0,ξ _i ^∧ ≥0

Different from the prior art, the above-mentioned technical scheme has the following advantages: for the traditional detection method, the test of the tester needs to be judged by an experienced doctor for the Chinese characters written by him, and the present invention performs the regression model by collecting a sample set. Training, and finally using the regression model to predict the tester, without the need of an experienced doctor for guidance, and at the same time, the evaluation of the patient's condition can be obtained, the corresponding economic expenses will also be greatly reduced, and the present invention diagnoses a tester's condition. The speed will also be much higher than manual diagnosis.

Description of drawings

FIG. 1 is a flow chart showing the execution of a method for judging the disease degree of essential tremor based on machine learning.

Detailed ways

In order to describe in detail the technical content, structural features, achieved objectives and effects of the technical solution, the following detailed description is given in conjunction with specific embodiments and accompanying drawings.

Please refer to FIG. 1, a preferred embodiment of the present invention is a method for judging the disease degree of essential tremor based on machine learning, including the following steps:

Step 1: Select 10 Chinese characters from the Chinese character library as the test characters that the tester needs to write when evaluating essential tremor disease.

Step 2: Collect the data of healthy people and patients with essential tremor when writing the Chinese characters selected in Step 1 on the electronic tablet. These data include: the time when the electronic pen does not touch the tablet on the electronic tablet, the time when the electronic pen touches the tablet on the electronic tablet, the coordinates X and Y of the pen tip, the angle O between the pen tip and the horizontal plane, and the angle between the pen tip and the vertical plane. Included angle A, the pressure of the pen tip on the electronic tablet.

Step 3: Preprocess the collected data. Check whether there are obvious errors in the collected data, such as interruptions in the tester's writing, or obvious data that does not conform to common sense when recording, and if there is any, the group of data is directly discarded. Normalize temporal features to map homogeneous data to decimals between 0 and 1. Data can be normalized using the following formula:

Wherein T _up/down represents the time when the electronic pen does not touch the tablet or contacts the tablet on the electronic tablet, and T _up/down ' represents the normalized time data.

Fourier transform is performed on the pen tip X-axis data, Y-axis data, the angle O between the pen tip and the horizontal plane, and the angle A between the pen tip and the vertical plane to obtain the first 50hz frequency domain data. Fourier transform is a method of analyzing signals. It can analyze the components of the signal and represent the data in the frequency domain space. The data can be converted from the time domain space to the frequency domain space according to the following formula:

For the SVR regression model, if the training data directly uses the time domain information, because the time domain information fluctuates greatly, and the difference between similar samples is also large, the trained model has a high probability of overfitting, and the training data After converting into frequency domain information, the data volume becomes smaller and the feature difference becomes obvious, making it easier to fit. Calculate the maximum, minimum, mean, median, and variance of the recorded pressure data. In addition, the doctor also needs to give the degree of disease of the patient, which is used as a label for regression training. The prevalence and corresponding quantitative data are as follows:

Sickness Quantitative data healthy 0.0 slight 0.25 medium 0.5 more serious 0.75 serious 1.0

Step 4: Organize the sample features into the same structure for training. The preprocessed features are structured as follows:

X=[T _up T _down f _X f _Y f _O f _A F _max F _min F _ave F _mid σ _F ² ] ^T

where T _up represents the time that the electronic pen does not contact the tablet after normalization, and T _down represents the time that the electronic pen touches the tablet after normalization, f _X and f _Y represents the frequency vector of the pen tip X-axis and Y-axis data, f _O and f _A represent the frequency vector of the angle O between the pen tip and the horizontal plane, and the angle A between the pen tip and the vertical plane, F _max , F _min , F _ave , F _mid , σ _F ² represent the maximum, minimum, mean, median and variance of the pressure data, respectively. The generated feature vectors will be input to the SVR regression model as training samples, and a good SVR regression model can be learned through these feature vectors.

Step 5: Train the SVR regression model, and the regression result is the disease degree coefficient. Save the trained regression model data. The SVR regression model is:

The optimization objective of establishing the SVR regression model is:

st|y _i -y _predict |≤ε

is the feature vector of the sample obtained in step 3, and y _i is the disease quantification coefficient corresponding to the sample. In order to improve the generalization ability of the model, slack variables are introduced Then the SVR problem can be transformed into the following optimization objective:

Among them, C represents the penalty coefficient, that is, the tolerance of errors. The higher the C, the lower the tolerance of the model to errors, and the easier it is to over-fit. If C is too large or too small, the generalization ability of the model will deteriorate. However, since the constraints of the optimization objective are not convex functions, further transformation is performed:

ξ _i ^∨ ≥0,ξ _i ^∧ ≥0

where ξ _i ^∧ represents the loss above the sensitivity barrier, and ξ _i ^∨ represents the loss below the sensitivity barrier. Considering the constraints, the Lagrangian operators α ^∧ , α ^∨ , β ^∧ , β ^∨ are introduced to transform the optimization problem into a dual problem:

Take the derivative of the above formula, and bring the result into the original formula, and introduce the kernel function Then you can get:

0≤α _i ^∧ ≤C,0≤α _i ^∨ ≤C

The above process needs to meet the KKT condition. According to the complementary relaxation condition in the KKT condition, the equivalence relationship between b and the support vector can be deduced, and then the form of the entire model can be obtained. The complementary relaxation conditions are as follows:

From the complementary relaxation condition, (α _i ^∧ -α _i ^∨ )=0, so the support vector is the sample point where (α _i ^∧ -α _i ^∨ )≠0. Get:

Then select a sample that satisfies 0<α _i ^∨ <C, and solve b by the following formula.

The kernel function of SVR selects the radial basis kernel.

The SVR regression model is trained and saved for future use, and the data collected by the electronic tablet when the user selects 10 characters in the first step of handwriting is collected, and then the data is preprocessed in step 3, and the preprocessed data is input into the SVR regression model. Regression is performed to give the prediction results and the prevalence coefficient for doctors' reference.

It should be noted that, in this document, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any relationship between these entities or operations. any such actual relationship or sequence exists. Moreover, the terms "comprising", "comprising" or any other variation thereof are intended to encompass non-exclusive inclusion, such that a process, method, article or terminal device comprising a list of elements includes not only those elements, but also a non-exclusive list of elements. other elements, or also include elements inherent to such a process, method, article or terminal equipment. Without further limitation, an element defined by the phrase "includes..." or "comprises..." does not preclude the presence of additional elements in the process, method, article, or terminal device that includes the element. In addition, in this document, "greater than", "less than", "exceeds", etc. are understood to exclude the number; "above", "below", "within" and the like are understood to include the number.

Although the above embodiments have been described, those skilled in the art can make additional changes and modifications to these embodiments once they know the basic inventive concept, so the above is only the implementation of the present invention For example, it does not limit the scope of patent protection of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the description and drawings of the present invention, or directly or indirectly used in other related technical fields, are similarly included in this document. The invention is within the scope of patent protection.

Claims (5)

1. a method for judging the degree of morbidity of essential tremor based on machine learning, is characterized in that: it comprises the following steps, Step 1. Select N Chinese characters from the Chinese character library as the test words that the tester needs to write when evaluating the essential tremor disease, wherein N≥1; Step 2: Collect the data of healthy people and patients with essential tremor when they use the electronic pen to write the test words on the electronic handwriting board as sample data; Step 3: Preprocessing the collected sample data; Step 4: Organize the sample features into the same structure for training; Step 5, train the SVR regression model, and the regression result is the disease degree coefficient; save the regression model data after training; Step 6: Collect the data of the test words written by the tester on the electronic handwriting board, process the collected data according to Steps 3 and 4, and input it into the SVR regression model trained in Step 5 to obtain the disease degree coefficient. 2. the method for judging the degree of disease of essential tremor based on machine learning as claimed in claim 1, is characterized in that: the sample data collected in the described step 2 comprises that the electronic pen does not contact the handwriting board on the electronic handwriting board. Time, the time the electronic pen touches the tablet on the electronic tablet, the coordinates X and Y of the pen tip, the angle O between the pen tip and the horizontal plane, the angle A between the pen tip and the vertical plane, and the pressure of the pen tip on the electronic tablet. 3. the method for judging the disease degree of essential tremor based on machine learning as claimed in claim 1, it is characterized in that: in described step 3, sample data preprocessing, specifically comprises checking whether the collected data is wrong, if If there is, discard the erroneous data; normalize the temporal features, map the same data to decimals between -1 and 1, and use the following formula to normalize the data: Perform Fourier transform on the pen tip X-axis data, Y-axis data, the angle O between the pen tip and the horizontal plane, and the angle A between the pen tip and the vertical plane to obtain the first 50hz frequency domain data; calculate the maximum and minimum values of the recorded pressure data , mean, median, and variance. 4. The method for judging the degree of morbidity of essential tremor based on machine learning as claimed in claim 1, wherein the step 4 is specifically structured as follows to the preprocessed feature: X=[T _up T _down f _X f _Y f _O f _A F _max F _min F _ave F _mid σ _F ² ] ^T where T _up represents the time that the electronic pen does not contact the tablet after normalization, and T _down represents the time that the electronic pen touches the tablet after normalization, f _X and f _Y represents the frequency vector of the pen tip X-axis and Y-axis data, f _O and f _A represent the frequency vector of the angle O between the pen tip and the horizontal plane, and the angle A between the pen tip and the vertical plane, F _max , F _min , F _ave , F _mid , σ _F ² represent the maximum, minimum, mean, median and variance of the pressure data, respectively. 5. the judging method of the disease degree of essential tremor based on machine learning as claimed in claim 1, is characterized in that: in described step 5, SVR regression model is: The optimization objective of establishing the SVR regression model is: st-ε-ξ _i ^∨ ≤ωgx _i +by _i ≤ξ _i ^∧ +ε ξ _i ^∨ ≥0, ξ _i ^∧ ≥0.