CN116458898A - A feature extraction method and system for Parkinson's disease depression - Google Patents

Abstract

The invention proposes a feature extraction method for Parkinson's disease depression, which belongs to the field of biomedical technology. The specific steps include: obtaining EEG signals, processing the EEG signals to obtain local field potentials, performing preprocessing on the local field potentials to obtain local field potentials to be processed; performing continuous wavelet transformation on the local field potentials to be processed, retaining alpha, lowbeta, highbeta and beta frequency bands; extracting the blasting characteristics of the above-mentioned frequency bands. The invention uses the continuous wavelet transform to process the local field potential, extracts the blasting characteristics of the alpha, lowbeta, high beta and beta frequency bands, and uses the Parkinson's disease depression feature extraction system to realize high-precision classification of the Parkinson's depression degree based on the extracted blasting features.

Description

A feature extraction method and system for Parkinson's disease depression

technical field

The invention relates to the technical field of biomedicine, and more specifically relates to a feature extraction method and system for Parkinson's disease depression.

Background technique

Parkinson's disease is the second most common neurodegenerative disorder, affecting 1%-3% of people over the age of 60 worldwide. It is characterized by nonmotor (sleep, sensory, cognitive, and autonomic disturbances) and motor symptoms (tremor, bradykinesia, stiffness, and gait disturbance) that have a profound impact on patient autonomy and quality of life. Depression is a prominent non-motor symptom in Parkinson's disease. Depression affects 20%-40% of Parkinson's patients, which is twice the rate in the general population. More importantly, doctors often ignore the impact of depression on Parkinson's patients, resulting in a greatly increased incidence of Parkinson's patients and a sharp decline in the quality of life.

In the prior art, methods based on biochemical reagents and EEG are usually used to detect depression in Parkinson's disease, or depression is detected on the basis of feature extraction and classification. In terms of feature extraction, early research on speech-based depression mainly focused on temporal features, such as pause time, recording time, feedback time to questions, speech rate, etc. Later, it was found that a single feature could not cover enough recognizable information to assist clinical diagnosis. With the in-depth study of speech signals, a large number of other speech signal features were constructed. However, the existing feature methods lack topic-scene-related verbal information and are not expressive enough in the field of depression detection, which limits the performance of the final Parkinson's depression detection system; therefore, it is necessary to improve the Parkinson's depression feature extraction method.

Therefore, it is an urgent problem to be solved by those skilled in the art to propose a method and system for feature extraction of Parkinson's disease depression, to perform objective and effective feature extraction, and then to realize accurate classification of Parkinson's depression degree.

Contents of the invention

In view of this, the present invention provides a Parkinson's disease depression feature extraction method and system, based on wavelet transform processing to obtain local field potential features, used for the system to classify the degree of depression.

In order to achieve the above object, the present invention adopts the following technical solutions:

On the one hand, the present invention provides a kind of Parkinson's disease depression feature extraction method, comprises the following steps:

S1. Obtain an EEG signal, process the EEG signal to obtain a local field potential, and perform preprocessing on the local field potential to obtain a local field potential to be processed;

S2, performing continuous wavelet transform on the local field potential to be processed, and retaining alpha, low beta, high beta and beta frequency bands;

S3. Extracting blasting features of each frequency band in S2.

Preferably, said S2 includes:

S21. Comparing the initial part of the local field potential signal f(t) to be processed with the wavelet ψ(t), and calculating wavelet coefficients;

S22. Move the wavelet ψ(t) to the right to obtain wavelet ψ(t-d), where d is the distance to move to the right, and continue to compare the local field potential signal f(t-d) to be processed with the wavelet ψ(t-d) to obtain the wavelet coefficient of the current part;

S23. Repeating S21 and S22, continuously shifting the wavelet to the right until the input of the local field potential signal to be processed ends;

S24, expand the wavelet to obtain the current wavelet function ψ(t/ns), n is the expansion multiple; repeat the above S21, S22, S23, retain the alpha, low beta, high beta and beta frequency bands.

Preferably, the formula for calculating the wavelet coefficients is as follows:

Where a represents the positioning frequency, and b represents the positioning time.

Preferably, the blasting features of each frequency band in S2 are extracted in the S3, including:

The wavelet coefficients on the alpha, lowbeta, highbeta and beta frequency bands are respectively averaged on their corresponding frequency bands to obtain the envelope curve fitting the local field potential signal to be processed, and the 75% quantile passing through the envelope curve is used as a threshold to obtain the blasting characteristics.

Preferably, the blasting characteristics include blasting duration, blasting probability, and blasting amplitude.

Preferably, the blasting area of the local field potential signal to be processed is obtained by fitting the time-amplitude diagram of the waveform of the local field potential signal to be processed based on wavelet coefficients of each frequency band, the width of the blasting area represents the duration of the blasting, the amplitude of the envelope of the blasting area is the blasting amplitude, and the blasting probability is the blasting frequency within a period of time.

On the other hand, the present invention proposes a Parkinson's disease depression feature extraction system for realizing the above-mentioned Parkinson's disease depression feature extraction method, the system comprising:

A preprocessing module, configured to acquire EEG signals, process the EEG signals to obtain local field potentials, and perform preprocessing on the local field potentials to obtain local field potentials to be processed;

The CWT processing module is used for performing continuous wavelet transformation on the local field potential to be processed, and retaining alpha, lowbeta, highbeta and beta frequency bands;

The feature extraction module is used to extract the blasting features of each frequency band in the CWT processing module.

It also includes a classification module, which is used to classify the degree of depression in Parkinson's disease according to the bursting features; wherein the classification is based on two labels: normal and depression.

It can be seen from the above-mentioned technical solutions that, compared with the prior art, the feature extraction method and system for Parkinson's disease depression provided by the present invention can effectively classify Parkinson's disease depression through signal analysis and feature extraction of local field potentials. The process is efficient and direct, and the specific beneficial effects are as follows:

(1) The present invention uses the continuous wavelet transform to process the local field potential, extracts the blasting characteristics of the alpha, lowbeta, highbeta and beta frequency bands, and utilizes the Parkinson's disease depression feature extraction system proposed by the present invention to achieve high-precision classification of Parkinson's depression degree.

(2) There is a significant difference in the average burst amplitude in the alpha frequency band between Parkinson's patients with depression and non-depression patients (P=0.001), so we believe that this feature can be used as a marker of PD with depression and provide a new algorithm for the future development of brain-computer interface.

Description of drawings

Fig. 1 is the flow chart of Parkinson's disease depression feature extraction method that the embodiment of the present invention provides;

Fig. 2 is the flow chart of continuous wavelet processing provided by the embodiment of the present invention;

Fig. 3 is the time-amplitude diagram of the local field potential signal to be processed fitted by the beta frequency band wavelet coefficient after the CWT processing provided by the embodiment of the present invention;

Fig. 4 is the time amplitude and spectrum amplitude diagram of the local field potential signal to be processed after CWT processing provided by the embodiment of the present invention;

Fig. 5 is a comparison chart of each item in the depression group and the normal group on the alpha frequency band; wherein A is a comparison chart of blasting probability, B is a comparison chart of blasting duration, C is a comparison chart of blasting amplitude, and D is a comparison chart of blasting length between two groups;

Fig. 6 is the comparison diagram of items on the depression group and the normal group on the beta frequency band; wherein A is a comparison diagram of blasting probability, B is a comparison diagram of blasting duration, C is a comparison diagram of blasting amplitude, and D is a comparison diagram of blasting length between two groups;

Fig. 7 is a comparison diagram of items in the depression group and the normal group on the lowbeta frequency band; wherein A is a comparison diagram of blasting probability, B is a comparison diagram of blasting duration, C is a comparison diagram of blasting amplitude, and D is a comparison diagram of blasting length between two groups;

Figure 8 is a comparison chart of items in the depression group and the normal group on the highbeta frequency band; where A is a comparison chart of blasting probability, B is a comparison chart of blasting duration, C is a comparison chart of blasting amplitude, and D is a comparison chart of blasting length between two groups;

FIG. 9 is a flow chart of integrated learning processing provided by an embodiment of the present invention;

FIG. 10 is a characteristic curve of each machine learning model provided by the embodiment of the present invention.

Detailed ways

The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some, not all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present invention.

On the one hand, the embodiment of the present invention discloses a feature extraction method for Parkinson's disease depression, the flow chart is shown in Figure 1, which specifically includes the following steps:

S1. Acquire EEG signals, process EEG signals to obtain local field potentials, and preprocess local field potentials to obtain local field potentials to be processed; preprocessing includes denoising local field potentials, removing poorly recorded data, and filtering.

S2. Perform continuous wavelet transformation on the local field potential to be processed, and retain alpha, low beta, high beta and beta frequency bands.

Wavelet transform is a method for analyzing signal time-frequency, which has the characteristics of multi-resolution analysis. That is, it has low time domain resolution and high frequency domain resolution for the low frequency part of the signal, and has high time domain resolution and low frequency domain resolution for the high frequency part of the signal. Wavelet analysis has a strong sensitivity to the local characteristics of the signal. Compared with the Fourier transform, it can more clearly reflect the change of non-stationary signal frequency with time.

The continuous wavelet transform flow chart is shown in Figure 2, including the following steps:

S21. The initial part of the local field potential signal f(t) to be processed is compared with the wavelet ψ(t), and the wavelet coefficient is calculated; the coefficient C is used to reflect the degree of approximation between this part and the wavelet. When the value of C is larger, it means that it is more similar. On the contrary, the lower the value, the less similar it is.

S22. Move the wavelet ψ(t) to the right to obtain the wavelet ψ(t-d), where d is the distance to the right, and the current local field potential signal f(t-d) to be processed continues to compare with the wavelet ψ(t-d) to obtain the wavelet coefficient of the current part;

S23. Repeat S21 and S22, and continuously move the wavelet to the right until the input of the local field potential signal to be processed ends;

S24, expand the wavelet to obtain the current wavelet function ψ(t/ns), n is the expansion multiple; repeat S21, S22, S23, and keep alpha, low beta, high beta and beta frequency bands.

In the specific implementation process, the wavelet coefficient calculation formula is as follows:

Where a represents the positioning frequency, and b represents the positioning time.

S3. Extracting blasting features of each frequency band in S2.

The wavelet coefficients in the alpha, low beta, high beta, and beta frequency bands are averaged in their corresponding frequency bands to obtain the envelope curve of the local field potential signal to be processed, and the 75% quantile passing through the envelope line is used as the threshold value to obtain the blasting characteristics (blasting duration, blasting probability, and blasting amplitude).

Based on the wavelet coefficients of each frequency band, the time-amplitude diagram of the local field potential signal waveform to be processed is fitted to obtain the blasting area of the local field potential signal to be processed. The width of the blasting area represents the blasting duration, the envelope amplitude of the blasting area is the blasting amplitude, and the blasting probability is the blasting frequency within a period of time.

Taking the beta frequency band as an example, Fig. 3 shows the time-amplitude diagram of the unprocessed local field potential signal fitted by the beta frequency band wavelet coefficient after CWT processing. After the CWT processing of the local field potential signal to be processed, the time amplitude and spectrum amplitude diagram are shown in Figure 4. The wavelet coefficients on the beta frequency band are averaged on the beta frequency band to obtain the envelope curve of the original signal, and the 75% quantile of the envelope line is used as the threshold to obtain the blasting characteristics (blasting duration, blasting probability, and blasting amplitude). Blasting less than 100 ms is not included in the statistics. As shown in Figure 3, the black line in the figure is the original signal, that is, the local field potential signal to be processed; the red line is the envelope curve fitted by the wavelet coefficient, and the blue line is the 75% quantile; the shaded part in the figure is the region where the signal has blasting, the width of the shaded part represents the duration of the blasting, the amplitude of the envelope of the blasting area is the blasting amplitude, and the blasting probability is the blasting frequency within a period of time (as shown in Figure 3, the blasting probability is: 7/4 (blasting/s)).

Specifically, according to the HAMD-24 scale, Parkinson's patients were divided into depression and normal groups. In the above four frequency bands, three features were extracted from each frequency band, and the three features of each frequency band were compared. The comparison is shown in Figure 5-Figure 8. The analysis found that the average burst amplitude on the alpha frequency band was significantly different between the two groups, so this feature can be used as a marker for PD with depression. Provide new algorithms for the future development of brain-computer interfaces.

Based on the above conclusions, the embodiment of the present invention also proposes a feature extraction system for Parkinson's disease depression, which can implement the above method and further classify the degree of depression in Parkinson's disease, specifically including:

The preprocessing module is used to obtain EEG signals, process the EEG signals to obtain local field potentials, and preprocess the local field potentials to obtain local field potentials to be processed;

The CWT processing module is used to perform continuous wavelet transformation of the local field potential to be processed, and retain alpha, lowbeta, highbeta and beta frequency bands;

The feature extraction module is used to extract the blasting features of each frequency band in the CWT processing module.

A classification module is also included, which is used to classify the degree of depression in Parkinson's disease according to the bursting features.

Specifically, in the classification module of this embodiment, in this embodiment, the 12 features extracted by wavelet transform are classified using the method of integrated learning, which is used to verify that the blasting feature proposed by the present invention has a good representation effect for the classification of Parkinson's depression. The specific process is shown in Figure 9:

Based on the information of the local field potential signal to be processed after CWT processing, blasting features are extracted, and the extracted features are input into the classification prediction model.

In this embodiment, through the analysis of six machine learning models, it is finally determined that the ensemble learning with the highest accuracy rate is used as the classification prediction model. The specific results of the classification accuracy of each machine learning model are shown in Table 1, and Figure 10 is the characteristic curve (Receiver Operating Characteristic, ROC) of each machine learning model. Ensemble learning is a predictive model algorithm that combines multiple machine learning techniques to achieve the effect of reducing variance, bias or improving prediction.

Table 1 Machine learning model classification accuracy

According to the HAMD-24 scale, Parkinson's patients were divided into depression group and normal group, which were used to train the classification prediction model. The integrated learning classification prediction model includes: K-NearestNeighbor, Multilayer Perceptron, Random Forests, and Decision Tree. The grid search method is used to find the optimal parameter combination, and the corresponding proportions are [1,1,1,1]. Vote on the classification results predicted by the proximity algorithm, multi-layer perceptron, random forest, and decision tree, and the minority obeys the majority to determine the classification result.

Each embodiment in this specification is described in a progressive manner, each embodiment focuses on the difference from other embodiments, and the same and similar parts of each embodiment can be referred to each other. As for the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and for relevant details, please refer to the description of the method part.

The above description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention will not be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims (8)

1. A method for extracting characteristics of parkinsonism depression, which is characterized by comprising the following steps:

s1, acquiring an electroencephalogram signal, processing the electroencephalogram signal to obtain a local field potential, and preprocessing the local field potential to obtain a local field potential to be processed;

s2, carrying out continuous wavelet transformation on the local field potential to be processed, and reserving alpha, low beta, high beta and beta frequency bands;

s3, extracting blasting characteristics of each frequency band in the S2.

2. The method for extracting features of depression in parkinson' S disease according to claim 1, wherein said S2 comprises:

s21, comparing the initial part of the local field potential signal f (t) to be processed with a wavelet ψ (t), and calculating a wavelet coefficient;

s22, shifting the wavelet psi (t) rightward to obtain a wavelet psi (t-d), wherein d is the distance of the shift rightward, and continuously comparing the current local field potential signal f (t-d) to be processed with the wavelet psi (t-d) to obtain a wavelet coefficient of the current part;

s23, repeating the S21 and the S22, and continuously moving the wavelet to the right until the input of the local field potential signal to be processed is finished;

s24, expanding the wavelet to obtain a current wavelet function psi (t/ns), wherein n is an expansion multiple; and repeating the S21, the S22 and the S23, and reserving the alpha, the low beta, the high beta and the beta frequency bands.

3. The method for extracting features of depression in parkinson's disease according to claim 2, wherein said wavelet coefficients are calculated as follows:

where a represents the positioning frequency and b represents the positioning time.

4. The method for extracting features of depression in parkinson' S disease according to claim 2, wherein said extracting the blasting features of each frequency band in S2 in S3 comprises:

and respectively averaging wavelet coefficients on the alpha frequency band, the low beta frequency band, the high beta frequency band and the beta frequency band to obtain an envelope curve fitting the local field potential signal to be processed, and obtaining the blasting characteristics by taking 75% quantiles passing through the envelope curve as a threshold value.

5. A method of feature extraction for depression in parkinson's disease according to claim 4, wherein said blast features comprise duration of blast, probability of blast, magnitude of blast.

6. The method for extracting the characteristics of the depression of the parkinson's disease according to claim 5, wherein a blasting area of the local field potential signal to be processed is obtained based on a time amplitude graph of a waveform of the local field potential signal to be processed fitted with wavelet coefficients of each frequency band, the width of the blasting area represents the blasting duration, the amplitude of an envelope of the blasting area is the blasting amplitude, and the blasting probability is the blasting frequency in a period of time.

7. A feature extraction system for parkinson's disease depression, comprising:

the preprocessing module is used for acquiring an electroencephalogram signal, processing the electroencephalogram signal to obtain a local field potential, and preprocessing the local field potential to obtain a local field potential to be processed;

the CWT processing module is used for carrying out continuous wavelet transformation on the local field potential to be processed and reserving alpha, lowbeta, highbeta and beta frequency bands;

and the feature extraction module is used for extracting the blasting features of each frequency band in the CWT processing module.

8. The parkinsonism depression feature extraction system of claim 7, wherein said system further comprises:

and the classification module is used for classifying the depression degree of the Parkinson's disease according to the blasting characteristics.