CN118383730B - Epileptic seizure early warning method, epileptic seizure early warning system, electronic equipment and storage medium - Google Patents

Abstract

The application discloses an epileptic seizure early warning method, an epileptic seizure early warning system, electronic equipment and a storage medium, and relates to the technical field of risk assessment. In the method, noise interference in the first electroencephalogram signal and the first electrocardiograph signal is removed through a filter, and a second electroencephalogram signal and a second electrocardiograph signal are obtained; carrying out standardization processing on the first motion data to obtain second motion data; extracting electroencephalogram signal features from the second electroencephalogram signal, electrocardiographic signal features from the second electrocardiographic signal, and motion data features from the second motion data; obtaining seizure prediction scores based on the electroencephalogram signal features, the electrocardiographic signal features, and the motion data features; judging whether the seizure prediction score is larger than or equal to a preset early warning threshold value; and when the seizure prediction score is greater than or equal to a preset early warning threshold value, determining that seizure risk exists. By implementing the technical scheme of the application, the accuracy of epileptic seizure probability prediction can be improved.

Description

Epileptic seizure early warning method, epileptic seizure early warning system, electronic equipment and storage medium

Technical Field

The application relates to the technical field of risk assessment, in particular to a seizure early warning method, a seizure early warning system, electronic equipment and a storage medium.

Background

Epilepsy is a neurological disorder that severely affects the quality of life of a patient and is characterized by unpredictable recurrent attacks. The existing epileptic seizure prediction method mainly comprises the steps of analyzing acquired physiological data of a user to be monitored by constructing a prediction model, and judging the occurrence probability of epileptic seizure of the user to be monitored. However, when the probability of occurrence of a disease is actually predicted, since physiological data of a user to be monitored is often single-mode data, abnormal waveform characteristics indicating seizures are usually identified by monitoring an electroencephalogram signal. However, the analysis of a single physiological signal is not comprehensive, so that the accuracy of the obtained epileptic seizure probability is low when the epileptic seizure probability is predicted by a traditional prediction model.

Therefore, how to improve the accuracy of seizure probability prediction is a technical problem to be solved.

Disclosure of Invention

The application provides a seizure early warning method, a seizure early warning system, electronic equipment and a storage medium, which can improve the accuracy of seizure probability prediction.

In a first aspect, the present application provides a seizure early warning method, the method comprising: acquiring a first electroencephalogram signal, a first electrocardiographic signal and first movement data of a user to be monitored; removing noise interference in the first electroencephalogram signal and the first electrocardiogram signal through a filter to obtain a second electroencephalogram signal and a second electrocardiogram signal; performing standardization processing on the first motion data to obtain second motion data; extracting electroencephalogram signal features from the second electroencephalogram signal, extracting electrocardiographic signal features from the second electrocardiographic signal, and extracting motion data features from the second motion data; obtaining seizure prediction scores based on the electroencephalogram signal features, the electrocardiographic signal features, and the motion data features; judging whether the epileptic seizure prediction score is larger than or equal to a preset early warning threshold value; and when the seizure prediction score is larger than or equal to the preset early warning threshold value, determining that the seizure risk exists in the user to be monitored.

By adopting the technical scheme, the physiological data and the motion state of the user are acquired in real time by acquiring the first electroencephalogram signal, the first electrocardiographic signal and the first motion data of the user to be monitored, so that a comprehensive data base is provided. Noise interference in the first electroencephalogram signal and the first electrocardiogram signal is removed through a filter, and a second electroencephalogram signal and a second electrocardiogram signal are obtained, so that the purity and reliability of the signals are improved, the accuracy of subsequent feature extraction and analysis is ensured, and the influence of external noise on a prediction result is reduced. And carrying out standardization processing on the first motion data to obtain second motion data, so that the influence of different orders and units is eliminated, the motion data has consistency and comparability in analysis, and the stability and the accuracy of the model are enhanced. Electroencephalogram signal features are extracted from the second electroencephalogram signals, electrocardiographic signal features are extracted from the second electrocardiographic signals, motion data features are extracted from the second motion data, and therefore multidimensional feature information is obtained, the features can comprehensively reflect physiological states and motion modes of users, abundant input data are provided for epileptic seizure prediction, and comprehensiveness and accuracy of prediction are improved. Based on the electroencephalogram signal characteristics, the electrocardiographic signal characteristics and the motion data characteristics, the epileptic seizure prediction score is obtained, so that various characteristic information is integrated to generate an integrated prediction score. Whether the seizure prediction score is larger than or equal to a preset early warning threshold value is judged, so that the current seizure risk level of a user is determined, and the balance between high sensitivity and high specificity can be achieved by setting a reasonable threshold value, so that early warning accuracy is improved, and false alarm is reduced. When the epileptic seizure prediction score is larger than or equal to a preset early warning threshold, determining that the user to be monitored has epileptic seizure risk, and accordingly effectively improving accuracy of epileptic seizure probability prediction.

Optionally, the acquiring the first electroencephalogram signal, the first electrocardiographic signal and the first motion data of the user to be monitored specifically includes: sending a first control signal to a first wearable device of the user to be monitored, so that the first wearable device collects the first electroencephalogram signal at a preset frequency; sending a second control signal to a second wearable device of the user to be monitored, so that the second wearable device acquires the first electrocardiogram signal at the preset frequency; and sending a third control signal to third wearable equipment of the user to be monitored, so that the third wearable equipment collects the first motion data at the preset frequency.

Optionally, the filter includes a bandpass filter and a notch filter; the removing, by a filter, noise interference in the first electroencephalogram signal and the first electrocardiograph signal to obtain a second electroencephalogram signal and a second electrocardiograph signal specifically includes: determining a first noise type corresponding to the first electroencephalogram signal and a second noise type corresponding to the first electrocardiograph signal; wherein the first noise type comprises first power line interference, electromyographic signal interference, and eye movement artifacts; the second noise type includes second power line interference, baseline wander, and high frequency noise; filtering the electromyographic signal interference and the noise interference of the eye movement artifact through the band-pass filter, and filtering the noise interference of the first power line interference through the notch filter to obtain the second electroencephalogram signal; and filtering noise interference of the baseline drift and the high-frequency noise through the band-pass filter, and filtering noise interference of the second power line interference through a notch filter to obtain the second electrocardiogram signal.

By adopting the technical scheme, the first noise type corresponding to the first electroencephalogram signal and the second noise type corresponding to the first electrocardiograph signal are determined, so that corresponding filtering measures can be adopted for different types of noise, the pertinence and the effectiveness of signal processing are improved, and the accuracy of subsequent signal analysis is ensured. By determining that the first noise type comprises first power line interference, electromyographic signal interference and eye movement artifacts, common interference sources in the electroencephalogram signals can be identified and processed, the noise is effectively removed, and the quality of the electroencephalogram signals is improved. By determining that the second noise type comprises second power line interference, baseline drift and high-frequency noise, common interference sources in the electrocardiogram signals can be identified and processed, the noise is effectively removed, and the quality of the electrocardiogram signals is improved. The electromyogram signal interference and the noise interference of the eye movement artifact are filtered through the band-pass filter, and the noise interference of the first power line interference is filtered through the notch filter, so that a second electroencephalogram signal is obtained, various noises in the electroencephalogram signal can be effectively removed, and the purity and reliability of the signal are improved. The noise interference of baseline drift and high-frequency noise is filtered through the band-pass filter, and the noise interference of the second power line interference is filtered through the notch filter, so that a second electrocardiogram signal is obtained, various noises in the electrocardiogram signal can be effectively removed, and the purity and reliability of the signal are improved.

Optionally, the first motion data includes first X-axis acceleration data, first Y-axis acceleration data, and first Z-axis acceleration data; the second motion data includes second X-axis acceleration data, second Y-axis acceleration data, and second Z-axis acceleration data; the step of carrying out standardization processing on the first motion data to obtain second motion data specifically comprises the following steps: calculating a first mean value and a first standard deviation corresponding to the first X-axis acceleration data, calculating a second mean value and a second standard deviation corresponding to the first Y-axis acceleration data, and calculating a third mean value and a third standard deviation corresponding to the first Z-axis acceleration data; normalizing the first X-axis acceleration data based on the first mean value and the first standard deviation to obtain the second X-axis acceleration data; normalizing the first Y-axis acceleration data based on the second mean and the second standard deviation to obtain second Y-axis acceleration data; and normalizing the first Z-axis acceleration data based on the third mean value and the third standard deviation to obtain the second Z-axis acceleration data.

By adopting the technical scheme, the central trend and the dispersion degree of each axial movement data can be known by calculating the first mean value and the first standard deviation corresponding to the first X-axis acceleration data, calculating the second mean value and the second standard deviation corresponding to the first Y-axis acceleration data and calculating the third mean value and the third standard deviation corresponding to the first Z-axis acceleration data, and necessary statistical parameters are provided for subsequent data standardization processing. The first X-axis acceleration data is standardized based on the first mean value and the first standard deviation to obtain the second X-axis acceleration data, so that the magnitude difference of the data is eliminated, the data is converted into standard normal distribution with the mean value of zero and the standard deviation of one, and consistency and comparability of different axial data in analysis are ensured. The first Y-axis acceleration data is standardized based on the second mean value and the second standard deviation to obtain second Y-axis acceleration data, so that the comparison and analysis of motion data in different axial directions on the same scale are further ensured, the magnitude difference between the different axial directions is eliminated, and the availability of the data and the accuracy of analysis are improved. And the second Z-axis acceleration data is obtained by normalizing the first Z-axis acceleration data based on the third mean value and the third standard deviation, so that all axial motion data are ensured to be converted to the same scale, the consistency and the comparability of the data are improved, and high-quality basic data are provided for subsequent feature extraction and seizure prediction.

Optionally, the electroencephalogram signal characteristics comprise power spectral density, average potential value of the electroencephalogram signal, degree of dispersion of the electroencephalogram signal, maximum amplitude of the electroencephalogram signal, amplitude variation range of the electroencephalogram signal, complexity of the electroencephalogram signal, sample entropy and approximate entropy; the electrocardiogram signal features comprise RR interval, QT interval, time domain index of HRV, frequency domain index of HRV, poincare image features and sample entropy; the motion data features include an average of the motion data, a variance of the motion data, a peak of the motion data, an amplitude variation range of the motion data, a gesture variation, and a motion pattern.

Optionally, the electroencephalogram signal features include n first sub-features; the electrocardiogram signal features include m second sub-features; the motion data feature includes p third sub-features; the obtaining the epileptic seizure prediction score based on the electroencephalogram signal characteristic, the electrocardiographic signal characteristic and the motion data characteristic specifically comprises: carrying out standardization processing on the n first sub-features to obtain n standardized first sub-features, carrying out standardization processing on the m second sub-features to obtain m standardized second sub-features, and carrying out standardization processing on the m third sub-features to obtain p standardized third sub-features; calculating an electroencephalogram signal score based on the n normalized first sub-features; calculating an electrocardiogram signal score based on the m normalized second sub-features; calculating a motion data score based on p of the normalized third sub-features; the seizure prediction score is calculated based on the electroencephalogram signal score, the electrocardiographic signal score, and the motion data score.

By adopting the technical scheme, n first sub-features are standardized, n standardized first sub-features are obtained, m second sub-features are standardized, m standardized second sub-features are obtained, p third sub-features are standardized, p standardized third sub-features are obtained, and therefore magnitude and unit differences among different features are eliminated, all the features are compared and analyzed on the same scale, and the accuracy and consistency of subsequent calculation are ensured. The electroencephalogram signal score is calculated based on the n normalized first sub-features, so that a plurality of electroencephalogram features are combined to generate a score representing the overall state of an electroencephalogram, influence of the electroencephalogram signals in epileptic seizure prediction can be accurately estimated, and accuracy of a prediction model is improved. By calculating the electrocardiographic signal score based on the m normalized second sub-features, a plurality of electrocardiographic features are combined to generate a score representing the overall state of the electrocardiograph, which is helpful for accurately evaluating the influence of electrocardiographic signals in seizure prediction and improving the reliability of a prediction model. The motion data score is calculated based on the p normalized third sub-features, so that a plurality of motion features are combined to generate a score representing the overall state of the motion data, the influence of the motion state in seizure prediction can be accurately estimated, and the comprehensiveness of a prediction model is improved. The epileptic seizure prediction score is calculated based on the electroencephalogram signal score, the electrocardiographic signal score and the motion data score, so that multidimensional characteristic information is integrated into a comprehensive prediction score, different physiological signals and motion states can be comprehensively considered by the model, and the accuracy and reliability of epileptic seizure prediction are improved.

Optionally, the calculating the electroencephalogram signal score based on the n normalized first sub-features specifically includes:

the electroencephalogram signal score is calculated by the following formula:

；

wherein, For the electroencephalogram signal fraction,For the ith said normalized first sub-feature,The weight corresponding to the first sub-feature after the standardization is the ith weight;

the calculating the electrocardiographic signal score based on m second sub-features after normalization specifically comprises:

the electrocardiographic signal fraction is calculated by the following formula:

；

wherein, For the fraction of the electrocardiographic signal,For the j-th said normalized second sub-feature,The weight corresponding to the second sub-feature after the standardization is j;

Calculating a motion data score based on the p normalized third sub-features, wherein the motion data score specifically comprises:

the motion data score is calculated by the following formula:

；

wherein, For the score of the motion data,For the kth said normalized third sub-feature,The weight corresponding to the third sub-feature after the standardization is the kth;

The calculating the seizure prediction score based on the electroencephalogram signal score, the electrocardiographic signal score, and the motion data score specifically includes:

The seizure prediction score is calculated by the following formula:

；

wherein, For the seizure predictive score,For the electroencephalogram signal fraction,For the fraction of the electrocardiographic signal,For the motion data score, α is a first weight, β is a second weight, γ is a third weight, and b is a bias term.

In a second aspect of the application there is provided a seizure early warning system, the system comprising: the system comprises an acquisition module, a processing module and an early warning module; the acquisition module is used for acquiring a first electroencephalogram signal, a first electrocardiographic signal and first motion data of a user to be monitored; the processing module is used for removing noise interference in the first electroencephalogram signal and the first electrocardiograph signal through a filter to obtain a second electroencephalogram signal and a second electrocardiograph signal; the processing module is further used for carrying out standardized processing on the first motion data to obtain second motion data; the processing module is further used for extracting electroencephalogram signal characteristics from the second electroencephalogram signals, extracting electrocardiographic signal characteristics from the second electrocardiographic signals and extracting motion data characteristics from the second motion data; the processing module is further used for obtaining epileptic seizure prediction scores based on the electroencephalogram signal characteristics, the electrocardiographic signal characteristics and the motion data characteristics; the processing module is further used for judging whether the seizure prediction score is greater than or equal to a preset early warning threshold value; and the early warning module is used for determining that the user to be monitored has seizure risk when the seizure prediction score is greater than or equal to the preset early warning threshold.

In a third aspect the application provides an electronic device comprising a processor, a memory for storing instructions, a user interface and a network interface for communicating to other devices, the processor being arranged to execute the instructions stored in the memory to cause the electronic device to perform a method according to any of the first aspects of the application.

In a fourth aspect of the application a computer readable storage medium is provided, storing a computer program capable of being loaded by a processor and performing a method according to any of the first aspects of the application.

In summary, one or more technical solutions provided in the embodiments of the present application at least have the following technical effects or advantages:

The physiological data and the motion state of the user are obtained in real time by obtaining the first electroencephalogram signal, the first electrocardiographic signal and the first motion data of the user to be monitored, so that a comprehensive data base is provided. Noise interference in the first electroencephalogram signal and the first electrocardiogram signal is removed through a filter, and a second electroencephalogram signal and a second electrocardiogram signal are obtained, so that the purity and reliability of the signals are improved, the accuracy of subsequent feature extraction and analysis is ensured, and the influence of external noise on a prediction result is reduced. And carrying out standardization processing on the first motion data to obtain second motion data, so that the influence of different orders and units is eliminated, the motion data has consistency and comparability in analysis, and the stability and the accuracy of the model are enhanced. Electroencephalogram signal features are extracted from the second electroencephalogram signals, electrocardiographic signal features are extracted from the second electrocardiographic signals, motion data features are extracted from the second motion data, and therefore multidimensional feature information is obtained, the features can comprehensively reflect physiological states and motion modes of users, abundant input data are provided for epileptic seizure prediction, and comprehensiveness and accuracy of prediction are improved. Based on the electroencephalogram signal characteristics, the electrocardiographic signal characteristics and the motion data characteristics, the epileptic seizure prediction score is obtained, so that various characteristic information is integrated to generate an integrated prediction score. Whether the seizure prediction score is larger than or equal to a preset early warning threshold value is judged, so that the current seizure risk level of a user is determined, and the balance between high sensitivity and high specificity can be achieved by setting a reasonable threshold value, so that early warning accuracy is improved, and false alarm is reduced. When the epileptic seizure prediction score is larger than or equal to a preset early warning threshold, determining that the user to be monitored has epileptic seizure risk, and accordingly effectively improving accuracy of epileptic seizure probability prediction.

Drawings

Fig. 1 is a schematic flow chart of a seizure early warning method according to an embodiment of the present application;

FIG. 2 is a second flow chart of a seizure early warning method according to the embodiment of the application;

fig. 3 is a third flow chart of a seizure early warning method according to an embodiment of the present application;

Fig. 4 is a schematic structural diagram of an epileptic seizure early-warning system according to an embodiment of the present application;

fig. 5 is a schematic structural diagram of an electronic device according to an embodiment of the present application.

Reference numerals illustrate: 1. an acquisition module; 2. a processing module; 3. an early warning module; 500. an electronic device; 501. a processor; 502. a communication bus; 503. a user interface; 504. a network interface; 505. a memory.

Detailed Description

In order that those skilled in the art will better understand the technical solutions in the present specification, the technical solutions in the embodiments of the present specification will be clearly and completely described below with reference to the drawings in the embodiments of the present specification, and it is apparent that the described embodiments are only some embodiments of the present application, not all embodiments.

In describing embodiments of the present application, words such as "for example" or "for example" are used to mean serving as examples, illustrations, or descriptions. Any embodiment or design described herein as "such as" or "for example" in embodiments of the application should not be construed as preferred or advantageous over other embodiments or designs. Rather, the use of words such as "or" for example "is intended to present related concepts in a concrete fashion.

In the description of embodiments of the application, the term "plurality" means two or more. For example, a plurality of systems means two or more systems, and a plurality of screen terminals means two or more screen terminals. Furthermore, the terms "first," "second," and the like, are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating an indicated technical feature. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include one or more such feature. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless expressly specified otherwise.

The application provides a seizure early-warning method, and referring to fig. 1, one of flow diagrams of the seizure early-warning method provided by the embodiment of the application is shown. The method comprises the steps S1-S7, wherein the steps are as follows:

step S1: a first electroencephalogram signal, a first electrocardiographic signal, and first movement data of a user to be monitored are acquired.

In one possible implementation, step S1 specifically includes the following steps:

Transmitting a first control signal to a first wearable device of a user to be monitored, so that the first wearable device acquires a first electroencephalogram signal at a preset frequency; transmitting a second control signal to a second wearable device of a user to be monitored, so that the second wearable device acquires a first electrocardiogram signal at a preset frequency; and sending a third control signal to a third wearable device of the user to be monitored, so that the third wearable device collects the first motion data at a preset frequency.

Specifically, in order to acquire a first electroencephalogram signal, a first electrocardiographic signal, and first motion data of a user to be monitored, a system is designed that causes a plurality of wearable devices worn by the user to perform data acquisition at a preset frequency by transmitting control signals to the devices. The purpose of this is to be able to monitor the physiological status of the user accurately in real time, thereby providing high quality data for subsequent seizure prediction.

Step S1 includes the following parts. First, a first control signal is sent to a first wearable device worn by a user to be monitored. The device is an intelligent headband capable of acquiring electroencephalogram signals. By means of the first control signal, it is set to acquire a first electroencephalogram signal of the user at a preset frequency (for example 256 times per second). The high-frequency sampling can capture the tiny change in the brain electrical activity and improve the sensitivity of epileptic seizure detection.

Next, a second control signal is sent to a second wearable device worn by the user, the device being an electrocardiogram recorder. By means of the second control signal, it is set to acquire the first electrocardiogram signal at a preset frequency, for example 512 times per second. High frequency sampling of the electrocardiogram signal helps to accurately capture details of heart activity, particularly abnormal heart rhythm changes that may occur during a seizure.

Finally, a third control signal is sent to a third wearable device worn by the user, such as a smart wristband or smart watch, to collect the first motion data of the user at a preset frequency (e.g., 100 times per second). The device is capable of recording three-axis acceleration data, namely X-axis, Y-axis and Z-axis motion information. These motion data are critical for detecting abnormal motion patterns in seizures.

By the method, the electroencephalogram signal, the electrocardiographic signal and the motion data of the user can be acquired simultaneously. These data not only cover the electrical activity of the user's brain and heart, but also include the user's motor status, providing comprehensive physiological information. The multidimensional data acquisition method can improve the accuracy and reliability of seizure prediction.

In summary, step S1 is performed by sending a control signal to a plurality of wearable devices to acquire a first electroencephalogram signal, a first electrocardiographic signal, and first motion data at a preset frequency. The design can ensure that the physiological state of the user is accurately monitored in real time, and high-quality basic data is provided for subsequent data processing and seizure prediction.

Step S2: and removing noise interference in the first electroencephalogram signal and the first electrocardiogram signal through a filter to obtain a second electroencephalogram signal and a second electrocardiogram signal.

Specifically, in order to obtain a more accurate and clear second electroencephalogram signal and second electrocardiogram signal, step S2 removes noise interference in the first electroencephalogram signal and first electrocardiogram signal by means of a filter. The purpose of this is to eliminate various noise that may affect signal quality, thereby improving data reliability, providing high quality signal input for subsequent feature extraction and seizure prediction.

In one possible implementation, the filter includes a bandpass filter and a notch filter. Referring to fig. 2, a second flow chart of a seizure early-warning method according to an embodiment of the application is shown. The step S2 specifically comprises the steps S21-S23:

step S21: determining a first noise type corresponding to the first electroencephalogram signal and a second noise type corresponding to the first electrocardiograph signal; wherein the first noise type comprises a first power line disturbance, an electromyographic signal disturbance, and an eye movement artifact; the second noise type includes second power line interference, baseline wander, and high frequency noise.

Specifically, first, for the first electroencephalogram signal, the main noise type thereof is determined. Electroencephalogram signals are susceptible to a variety of noise disturbances including power line disturbances, electromyographic signal disturbances, and eye movement artifacts. Power line disturbances are typically caused by power frequency noise of the surrounding power system, with a common frequency of 50Hz or 60Hz. Myoelectric signal interference is high frequency noise due to scalp muscle activity, while eye movement artifacts are low frequency signals due to eye movement.

Also, for the first electrocardiogram signal, the main noise type thereof needs to be determined. Common noise disturbances in electrocardiographic signals include power line disturbances, baseline wander, and high frequency noise. The source of the power line interference is the same as the electroencephalogram signal, and is due to the power frequency noise of the power system. Baseline drift is typically low frequency drift due to respiratory motion or poor electrode contact, while high frequency noise may originate from external electromagnetic interference or physiological activity.

By determining these noise types, an explicit target can be provided for the subsequent filtering step. The recognition of the noise type is a precondition for the filtering process, and can ensure that proper filter parameters are selected, so that specific noise interference is effectively removed.

Step S22: and filtering the electromyographic signal interference and the noise interference of the eye movement artifact through a band-pass filter, and filtering the noise interference of the first power line interference through a notch filter to obtain a second electroencephalogram signal.

Specifically, first, a band-pass filter is used to remove electromyographic signal interference and noise interference of eye movement artifacts. The frequency range of the band pass filter is set to 0.5Hz to 40Hz. The device can effectively reserve the main frequency band in the electroencephalogram signal and remove high-frequency myoelectric noise and low-frequency eye movement artifacts. Myoelectric signal interference is typically high frequency noise due to scalp muscle activity, while eye movement artifacts are low frequency interference caused by eye movement. By means of the band-pass filter, it is ensured that these interfering signals are filtered out, so that purer electroencephalogram signals are obtained.

Next, a notch filter is used to remove noise interference from the first power line interference. The mains interference is usually derived from mains frequency noise of the surrounding power system, at 50Hz or 60Hz. The notch filter is specially designed for eliminating noise with specific frequency, and can accurately remove power frequency interference. By setting the center frequency of the notch filter to be 50Hz (or 60 Hz), the interference of the power line can be effectively eliminated, and the electroencephalogram signal is ensured not to be influenced by power frequency noise.

Step S23: and filtering the noise interference of baseline drift and high-frequency noise through a band-pass filter, and filtering the noise interference of the second power line interference through a notch filter to obtain a second electrocardiogram signal.

Specifically, first, a band-pass filter is used to remove the baseline wander and the interference of high-frequency noise. The frequency range of the band pass filter is set to 0.5Hz to 40Hz. Baseline wander is often a low frequency disturbance due to respiration, posture changes, or poor electrode contact that affects the baseline stability of the electrocardiogram. By setting the lower limit frequency of the band-pass filter to 0.5Hz, these low frequency components can be effectively filtered out, thereby stabilizing the baseline of the signal. The high-frequency noise may be derived from external electromagnetic interference or muscle activity, and the upper limit frequency of the band-pass filter is set to 40Hz, so that the high-frequency noise can be effectively removed, and the useful frequency band in the electrocardiogram signal is reserved.

Next, a notch filter is used to remove noise interference from the second power line interference. The power line interference is usually derived from power frequency noise in the environment, with frequencies of 50Hz or 60Hz. The notch filter is specially designed to eliminate noise of a specific frequency, and by setting the center frequency of the notch filter to be 50Hz (or 60 Hz), the power frequency noise can be accurately filtered. By the aid of the method, influence of power line interference on electrocardiogram signals can be avoided, and purity of the signals is guaranteed.

Step S3: and carrying out standardization processing on the first motion data to obtain second motion data.

In particular, in order to normalize the first motion data to obtain the second motion data, the motion data of different axes need to be converted to the same scale. The purpose of this is to eliminate the magnitude differences between the different axial data, making them consistent and comparable in subsequent analysis, thereby improving the accuracy of feature extraction and seizure prediction.

In one possible embodiment, the first motion data includes first X-axis acceleration data, first Y-axis acceleration data, and first Z-axis acceleration data; the second motion data includes second X-axis acceleration data, second Y-axis acceleration data, and second Z-axis acceleration data. The step S3 specifically comprises the following steps:

and calculating a first mean value and a first standard deviation corresponding to the first X-axis acceleration data, calculating a second mean value and a second standard deviation corresponding to the first Y-axis acceleration data, and calculating a third mean value and a third standard deviation corresponding to the first Z-axis acceleration data.

Specifically, in order to perform normalization processing on the first motion data, the mean value and standard deviation of the first X-axis acceleration data, the first Y-axis acceleration data, and the first Z-axis acceleration data need to be calculated first. The purpose of this is to understand the central trend and degree of dispersion of each axis of data, so that the data is converted into a standard normal distribution with zero mean and one standard deviation in the subsequent normalization process. This processing step is an important precondition to ensure data consistency and comparability.

First, all data points of the first X-axis acceleration data are acquired. By calculating the average of these data points, i.e., the first average μ _X, the central tendency of the first X-axis acceleration data can be known. The formula is as follows: ; where N is the total number of data points and f _X,i is the value of the ith data point in the first X-axis acceleration data. Next, a first standard deviation σ _X of the first X-axis acceleration data is calculated to measure the degree of dispersion of the data. The standard deviation is calculated as follows: 。

The same procedure, all data points of the first Y-axis acceleration data are acquired, and the mean values thereof, namely the second mean value mu _Y and the second standard deviation sigma _Y, are calculated. All data points of the first Z-axis acceleration data are acquired in a similar way, and a third mean mu _Z and a third standard deviation sigma _Z of the mean are calculated.

Through the steps, the mean value and the standard deviation of the first X-axis acceleration data, the first Y-axis acceleration data and the first Z-axis acceleration data are respectively calculated. These statistics can accurately describe the central trend and degree of dispersion of each axis of acceleration data.

And normalizing the first X-axis acceleration data based on the first mean value and the first standard deviation to obtain second X-axis acceleration data.

Specifically, the first X-axis acceleration data is subjected to normalization processing. The normalized formula is as follows: ; wherein, As the second X-axis acceleration data,For the first X-axis acceleration data, mu _X is the first mean and sigma _X is the first standard deviation. This conversion shifts the center of the raw data to zero and adjusts its dispersion to unit standard deviation. Through the normalization process, the magnitude and distribution characteristics of the first X-axis acceleration data are unified on the same scale. The normalized data mean value is zero, and the standard deviation is one, so that the normalized data mean value can be directly compared with other axial data in subsequent analysis and comprehensively processed.

And normalizing the first Y-axis acceleration data based on the second mean value and the second standard deviation to obtain second Y-axis acceleration data.

Specifically, as in the above embodiment, the first Y-axis acceleration data is normalized based on the second mean value and the second standard deviation to obtain the second Y-axis acceleration data, so that redundant description is omitted herein.

And normalizing the first Z-axis acceleration data based on the third mean value and the third standard deviation to obtain second Z-axis acceleration data.

Specifically, as in the above embodiment, the first Z-axis acceleration data is normalized based on the third mean value and the third standard deviation to obtain the second Z-axis acceleration data, so that redundant description is omitted herein.

Step S4: electroencephalogram signal features are extracted from the second electroencephalogram signal, electrocardiographic signal features are extracted from the second electrocardiographic signal, and athletic data features are extracted from the second athletic data.

In one possible embodiment, the electroencephalogram signal characteristics include power spectral density, average potential value of the electroencephalogram signal, degree of dispersion of the electroencephalogram signal, maximum amplitude of the electroencephalogram signal, amplitude variation range of the electroencephalogram signal, complexity of the electroencephalogram signal, sample entropy, and approximate entropy; the electrocardiogram signal features comprise RR interval, QT interval, time domain index of HRV, frequency domain index of HRV, poincare image features and sample entropy; the motion data characteristics include an average value of the motion data, a variance of the motion data, a peak value of the motion data, an amplitude variation range of the motion data, a posture variation, and a motion pattern.

Specifically, in order to extract useful features from the second electroencephalogram signal, the second electrocardiographic signal, and the second motion data, the purpose of step S4 is to obtain key information capable of reflecting the physiological state of the user by a feature extraction method, which provides a basis for subsequent seizure prediction. Doing so can improve the accuracy and reliability of the predictions.

First, an electroencephalogram signal feature is extracted from a second electroencephalogram signal. Electroencephalogram signal characteristics include power spectral density, average potential value of an electroencephalogram signal, degree of dispersion, maximum amplitude, amplitude variation range, complexity, sample entropy, and approximate entropy. The power spectral density calculates the power distribution of the signal in each frequency band by a frequency domain analysis method, such as Fast Fourier Transform (FFT), and reflects the spectral characteristics of the brain electrical activity. The average potential value reflects the overall potential level by calculating the average of the signal over a period of time. The degree of dispersion reflects the fluctuation of the signal by calculating the standard deviation or variance of the signal. The maximum amplitude and amplitude variation range reflects the amplitude characteristics of the signal by finding the extremum and amplitude difference of the signal. The complexity can be measured using fractal dimension, etc., and the sample entropy and the approximate entropy are used to measure the complexity and randomness of the signal.

Next, an electrocardiogram signal feature is extracted from the second electrocardiogram signal. The electrocardiogram signal features comprise RR interval, QT interval, time domain index of HRV, frequency domain index of HRV, poincare graph features and sample entropy. RR interval QT interval is the time from the start of the Q wave to the end of the T wave by detecting the time interval between adjacent R waves. Time domain indicators of HRV, such as SDNN and RMSSD, reflect heart rate variability. Frequency domain indicators of HRV, such as Low Frequency (LF) and High Frequency (HF) power, reflect the state of activity of the autonomic nervous system. The Poincare plot features calculate short-term and long-term variability by plotting scatter plots of adjacent RR intervals. Sample entropy is used to measure the complexity of the electrocardiographic signal.

Finally, motion data features are extracted from the second motion data. The motion data features include mean, variance, peak, amplitude variation range, attitude variation, and motion pattern of the motion data. The mean and variance reflect the overall level and volatility of the motion state by calculating the mean and variance of each axial data. The peak value and amplitude variation range reflects the motion strength by finding the extreme value and the amplitude difference of the data. The attitude change is detected by analyzing gyroscope data. The exercise mode identifies different exercise states such as walking, running, falling and the like through a machine learning algorithm.

Step S5: based on the electroencephalogram signal characteristics, the electrocardiographic signal characteristics and the motion data characteristics, the epileptic seizure prediction score is obtained.

Specifically, in order to obtain the seizure prediction score based on the electroencephalogram signal feature, the electrocardiographic signal feature, and the motion data feature, the purpose of step S5 is to construct a comprehensive prediction model by fusing various physiological and motion features, so as to accurately evaluate the risk of seizures. By doing so, the accuracy and reliability of prediction can be improved, and timely early warning information can be provided.

In one possible implementation, the electroencephalographic signal features include n first sub-features; the electrocardiogram signal features include m second sub-features; the motion data feature includes p third sub-features. Referring to fig. 3, a third flow chart of a seizure early-warning method according to an embodiment of the application is shown. Step S5 specifically includes steps S51-S55:

Step S51: and carrying out standardization processing on the n first sub-features to obtain n standardized first sub-features, carrying out standardization processing on the m second sub-features to obtain m standardized second sub-features, and carrying out standardization processing on the m third sub-features to obtain p standardized third sub-features.

Specifically, to ensure consistency and comparability of the n first sub-features, the m second sub-features, and the p third sub-features in the seizure prediction model, step S51 normalizes these features. The purpose of this is to transform features of different magnitudes and units onto the same scale, so that each feature can have consistent influence and weight in subsequent calculation, and the accuracy and reliability of the prediction model are improved.

First, raw data of each feature is acquired, and the mean and standard deviation thereof are calculated. For the n first sub-features, it is assumed that these features are f _EEG1,f_EEG2,…,f_EEGn, respectively. The mean μ _EEGi and standard deviation σ _EEGi for each feature were calculated as follows: ; where N is the number of samples, Is the a data point in the i first sub-feature.

Next, using the same method, the mean μ _ECGj and standard deviation σ _ECGj of the m second sub-features f _ECG1,f_ECG2,…,f_ECGn, and the mean μ _Motionk and standard deviation σ of the p third sub-features f _Motion1,f_Motion2,…,f_Motionn are calculated _Motionk.

The purpose of calculating the mean and standard deviation is to know the central trend and the degree of dispersion of each feature, so that the data can be converted into standard normal distribution with zero mean and one standard deviation in the subsequent standardization process.

Then, normalization processing is performed on each sub-feature. The normalized formula is as follows:

；

wherein, For the first sub-feature after the i-th normalization,For the second sub-feature normalized by j,Is the third sub-feature after the kth normalization.

The normalization process converts each feature into a standard normal distribution with zero mean and one standard deviation by subtracting the mean from the original feature value and dividing by the standard deviation. This transformation can eliminate the magnitude differences between different features, and make their influence and weights consistent in the model so that a particular feature is not biased in subsequent calculations.

Step S52: an electroencephalogram signal score is calculated based on the n normalized first sub-features.

Specifically, in order to accurately calculate the electroencephalogram signal score, step S52 performs processing by weighting and summing the normalized first sub-features, and applying a nonlinear transformation function. The purpose of this is to integrate the information of the different features and capture the complex relationship between the features using a nonlinear transformation, thereby generating a score that accurately reflects the overall state of the electroencephalogram signal. The following embodiments will explain in detail a specific manner of calculating the electroencephalogram signal score.

Step S53: an electrocardiogram signal score is calculated based on the m normalized second sub-features.

Specifically, in order to accurately calculate the electrocardiographic signal fraction, step S53 performs processing by weighting and summing the normalized second sub-features, and applying a nonlinear transformation function. The purpose of this is to integrate the information of the different features and capture the complex relationships between the features using a nonlinear transformation, thereby generating a score that accurately reflects the overall state of the electrocardiogram signal. The following embodiments will describe in detail a specific manner of calculating the electrocardiographic signal score.

Step S54: based on the p normalized third sub-features, a motion data score is calculated.

Specifically, to accurately calculate the motion data score, step S54 performs processing by weighted summing the normalized third sub-feature and applying a nonlinear transformation function. The purpose of this is to integrate the information of the different features and capture the complex relationships between the features using nonlinear transformations, thereby generating a score that accurately reflects the overall state of the motion data. The detailed description of the specific manner in which the athletic data score is calculated will follow.

Step S55: the seizure prediction score is calculated based on the electroencephalogram signal score, the electrocardiographic signal score, and the motion data score.

Specifically, in order to comprehensively evaluate the electroencephalogram signal score, the electrocardiographic signal score, and the motion data score, step S55 calculates the seizure prediction score by means of weighted summation. The purpose of this is to combine the different types of feature scores to generate a composite score to more accurately assess the risk of seizures.

In one possible implementation, step S52 specifically includes the following steps:

the electroencephalogram signal fraction is calculated by the following formula:

；

wherein, For the electroencephalogram signal fraction,For the first sub-feature after the i-th normalization,And the weight corresponding to the first sub-feature after the i-th normalization.

Specifically, in this formula, a weighted sum of the normalized first sub-features is first calculated. Each normalized first sub-feature has a corresponding weight w _i that reflects the importance of that feature in the overall score. Then, a nonlinear transformation function is applied to the result of the weighted summation. The result of the weighted summation is first processed using a ReLU (RECTIFIED LINEAR Unit) function, whose formula is: reLU (x) =max (0, x). The ReLU function is a commonly used activation function, mainly for introducing non-linear properties. In this formula, the function of the ReLU is to zero the negative part of the weighted sum result, leaving only the positive part. Doing so filters out insignificant negative changes, emphasizes significant positive changes, and enhances the responsiveness of the model to positive feature changes. Next, the sigmoid function is applied to further process the output of ReLU, with the formula: sigmoid (x) =1/(1+e ^−x). The sigmoid function limits the output of the ReLU to between 0 and 1 so that the result can be interpreted as a probability or score reflecting the overall state of the electroencephalogram signal.

Further describing this formula, complex nonlinear relationships may exist between the characteristics of the electroencephalogram signals. The linear model cannot effectively capture these relationships, while the ReLU function can introduce nonlinear characteristics, thereby enhancing the expressive power of the model on complex feature relationships. The ReLU function is simple to calculate, only the maximum value between the input and 0 is needed, and the calculation efficiency is high. In addition, the gradient of the ReLU function in the positive value area is 1, so that the training process of the model is accelerated, the problem of gradient disappearance is avoided, and the training speed and the convergence effect of the model are improved. The ReLU function zeroes out the negative part of the input, leaving only the positive part. This filters out insignificant negative changes, emphasizing significant positive changes. This is more important for larger variations in the eigenvalues of the electroencephalogram signal, as these variations may be more reflective of the risk of seizures.

The output range of the Sigmoid function is 0 to 1, and the input value can be mapped to a limited range. By applying a Sigmoid function to the ReLU output, the composite score can be converted to a probability value, which can be interpreted as a risk score for seizures. The Sigmoid function is an S-shaped curve, and can smooth input values and reduce the influence of noise on a final result. For electroencephalographic signals, the smoothed output may better reflect overall trends rather than being disturbed by individual outliers. The output value of the Sigmoid function may represent the clarity of classification when it is close to 0 or 1, and the uncertainty when it is near 0.5. The characteristics enable the result to have clear probability interpretation, and are helpful for reasonably evaluating and judging the seizure risk in practical application.

By weighted summing the normalized electroencephalogram features and then applying the ReLU and Sigmoid functions, the model is able to capture important nonlinear feature relationships in the electroencephalogram signals and emphasize important forward feature changes. Ultimately providing a smooth and probabilistic interpreted output.

In one possible implementation, step S53 specifically includes the following steps:

the electrocardiographic signal fraction is calculated by the following formula:

；

wherein, For the fraction of the ecg signal,For the second sub-feature normalized by j,And the weight corresponding to the second sub-feature after the j-th normalization.

Specifically, in this formula, a weighted sum of the normalized second sub-features is first calculated. Each normalized second sub-feature has a corresponding weight w _j that reflects the importance of that feature in the overall score. Then, a nonlinear transformation function is applied to the result of the weighted summation. The weighted summation result is first processed using a hyperbolic tangent function (tanh), the formula of which is: ; the tanh function is used to capture the two-way variation in the weighted sum result, its output range is [ -1,1], which can handle positive and negative values, and is suitable for the characteristic variation of electrocardiogram signals. Next, the output of tanh is further processed using a logistic function, the formula: ; the logistic function further smoothes the output of the tanh, limiting it to between 0 and 1, so that the result can be interpreted as a probability or score reflecting the overall state of the electrocardiogram signal.

Further describing this formula, the output range of the tanh function is [ -1,1], which can handle not only positive values but also negative values. This is important for electrocardiographic signals, since changes in electrocardiographic characteristics may have positive and negative implications. Because different electrocardiographic parameters (e.g., RR interval, QT interval, time and frequency domain indicators of HRV, etc.) may appear to increase or decrease under different physiological and pathological conditions, these changes can provide important information about cardiac function, especially in the detection of seizures.

Forward change (eigenvalue increase), RR interval increase: RR interval represents the time interval between two adjacent heartbeats. As RR intervals increase, this indicates a slowing of heart rate. This may occur at rest, relaxation or deep sleep, but may also indicate abnormal cardiac function in certain pathological situations, such as problems with the cardiac conduction system. QT interval increases: the QT interval represents the time from the start of the cardiac electrical activity (Q wave) to the end of repolarization (T wave). If the QT interval is prolonged, it may be indicative of abnormal cardiac repolarization, increasing the risk of arrhythmia. HRV (heart rate variability) increases: an increase in HRV generally indicates better autonomic nervous system function and heart health. High HRV is generally associated with good health and lower stress levels.

Negative change (eigenvalue decrease), RR interval decrease: when the RR interval decreases, this indicates an increase in heart rate. An increased heart rate may occur during exercise, stress or excitation, but in pathological situations may also be indicative of heart dysfunction, such as tachycardia. QT interval reduction: too short a QT interval may indicate insufficient cardiac repolarization time, which may be associated with certain types of arrhythmias. HRV (heart rate variability) reduction: a decrease in HRV is generally indicative of impaired autonomic nervous system function and higher stress levels. Low HRV is associated with poor health status and higher risk of cardiovascular disease.

Positive and negative changes in electrocardiogram features provide important physiological status information when detecting seizures. For example:

pre-seizure: before a seizure, the heart rate may increase (RR interval decreases), HRV may change, indicating that the autonomic nervous system is affected.

During the onset: during seizures, there may be dramatic changes in the electrocardiogram characteristics, such as significant heart rate changes (apparent RR interval), possibly accompanied by QT interval changes, indicating that the heart is significantly affected.

After onset: after the episode has ended, heart rate and HRV may gradually return to normal. These changes in recovery are also important to help assess the impact of the episode and recovery.

The tanh function can smoothly map the input values between [ -1,1], capture nonlinear relationships between features, and enhance the expressive power of the model. Features in electrocardiographic signals (such as time and frequency domain indicators of HRV) have complex nonlinear relationships. the tanh function can better capture these complex relationships by introducing nonlinear transformations, providing a sensitive response to feature changes. And the change of the electrocardiogram characteristics can have positive and negative meanings. the tanh function can process the positive and negative input values and map them smoothly to [ -1,1], thus preserving and distinguishing these changes.

The Logistic function is an S-shaped curve with an output value ranging from 0 to 1. By applying a Logistic function to the output of the tanh function, the composite score can be converted to a value between 0 and 1, which can be interpreted as a probability or risk score for seizures. The Logistic function can smooth the input values and make the results more interpretative and practical.

By introducing a nonlinear transformation (tanh) and a probability mapping (Logistic), the model can more fully express the characteristic relation in the electrocardiogram signal, and provide accurate risk assessment.

In one possible implementation, step S54 specifically includes the following steps:

The motion data score is calculated by the following formula:

；

wherein, In order to score the motion data,For the third sub-feature after the kth normalization,And the weight corresponding to the third sub-feature after the k normalization.

Specifically, in this formula, specifically, a weighted sum of the normalized third sub-features is first calculated. Each normalized third sub-feature has a corresponding weight w _k that reflects the importance of that feature in the overall score. Next, the squaring operation squares each term of the weighted sum result, which may enhance sensitivity to large variations. The squaring operation helps to amplify larger eigenvalues, weaken smaller eigenvalues, and thus highlight more important feature variations. Next, the squared result is processed by applying a tanh function, the formula is: ; the tanh function limits the results to the [ -1,1] range, making the output smoother and able to handle positive and negative variations.

To further describe this formula, the squaring operationEach term of the weighted sum result is squared, so that larger eigenvalues can be amplified and smaller eigenvalues can be weakened. For movement data, larger changes may be of greater significance, as they may be associated with seizures. Squaring helps to highlight important feature variations and thus increases the sensitivity of the model to key features.

The tanh function has an output range of [ -1,1], can handle positive and negative values, and smoothly limits the result to a limited range. By applying the tanh function, nonlinear transformation can be performed on the squared result, so that the output is smoother and more stable, and complex nonlinear relations among features are captured.

Large changes in motion data (e.g., sudden acceleration changes, falls, etc.) may be more reflective of the risk of seizures. The squaring operation amplifies these changes, making the model more sensitive to large changes, thereby improving the accuracy of the predictions.

And the change of the motion data characteristics can have positive and negative meanings. Because different movement patterns and states may appear to increase or decrease in eigenvalues, these changes may represent different meanings in different contexts, especially when detecting seizures, both of which need to be taken into account.

Forward direction change (eigenvalue increase), motion intensity increase: as athletic data characteristics (e.g., acceleration, speed) increase, it may be indicated that the user is performing more vigorous activities. For example, the acceleration during running is greater than during walking. Attitude change: a sudden acceleration increase may indicate a change of the user from a stationary state to a moving state, or a fall. These abrupt positive changes are important in monitoring seizures, which may cause the user to fall suddenly or undergo intense muscle contraction.

Negative change (eigenvalue decrease), motion intensity decrease: when the athletic data characteristics decrease, it may be indicated that the user has changed from a strenuous exercise to a stationary state. For example, the acceleration and speed may drop significantly when the resting state is entered after running. Stable posture: a sustained low acceleration value may indicate that the user is in a stable posture, such as sitting or lying. These negative changes are also important because the user may enter an unconscious resting state after a seizure. In seizure detection, both positive and negative changes in motion data may provide key clues. For example:

pre-seizure: a slight positive change, such as a slight increase in acceleration, may occur prior to the episode, indicating that the user may be experiencing a slight motion in anticipation of a sense of being, such as experiencing discomfort.

During the onset: severe positive changes may occur during the episode, such as sudden falls, strong muscle twitches, etc. These dramatic changes appear as distinct peaks in the acceleration data.

After onset: after the episode has ended, the user may enter a resting state where the acceleration signature may decrease significantly, indicating a decrease in the intensity of the motion.

The tanh function can process the positive and negative input values and map them smoothly to [ -1,1], thus preserving and distinguishing these changes. the tanh function smoothes the squared result further and is limited between [ -1,1] so that the result can be interpreted as a smoothed score. This is important for practical applications because a smooth score can reduce the effects of noise, making the results more stable and reliable. By introducing nonlinear transformation (tanh) and enhanced sensitivity (squaring operations), the model can more fully express the feature relationships in the motion data, providing accurate risk assessment.

In one possible implementation, step S55 specifically includes the following steps:

The seizure prediction score is calculated by the following formula:

；

wherein, For the seizure predictive score,For the electroencephalogram signal fraction,For the fraction of the ecg signal,For the motion data score, α is a first weight, β is a second weight, γ is a third weight, and b is a bias term.

Specifically, weighted summation is performed based on these scores. Each score S _EEG、S_ECG and S _Motion has a corresponding weight α, β, and γ. These weights reflect the importance of the different signals in the overall score.

The offset term b can be adjusted in a translation mode on the basis of the weighted sum, so that the output of the model is more suitable for the distribution of data. For example, in neural networks, the input to the activation function typically requires adjustment via weighted summation and bias term. By introducing the bias term, it can be ensured that the activation function works effectively near the center of the data distribution. The bias term b also enhances the expressive power of the model. Even if the values of all input features are zero, the bias term b still enables the model to produce a non-zero output, thereby increasing the flexibility of the model. At the same time, the introduction of the bias term b can reduce the under fitting phenomenon of the model. Under-fitting refers to a pattern in which the pattern cannot capture data, and is manifested by high training errors and high test errors. By adding the bias term, the model can better fit training data, and overall performance is improved. And there may be systematic deviations in the data, i.e., global shifts in the eigenvalues. For example, all eigenvalues are higher or lower. The bias term b can help correct such systematic deviations so that the model more accurately reflects the actual condition of the data.

The weights α, β, γ and the bias term b in this formula are determined by training a model. The specific training mode is as follows:

First, a large amount of historical data needs to be collected and prepared, which should include: electroencephalogram (EEG) signals and their features, electrocardiogram (ECG) signals and their features, motor data and their features, and seizure tags (i.e., whether a seizure has occurred) will be used to train and validate the model. And extracting the required characteristics from the original data, and carrying out standardization processing on the characteristics so as to enable the characteristics to have the same scale.

The weights α, β and γ and bias term b of the model are initialized. Random initialization methods, such as gaussian or uniform distribution, may be used, or more complex initialization methods (such as Xavier initialization) may be used to enhance training.

An appropriate loss function is then selected to evaluate the prediction error of the model. Common loss functions include Mean Square Error (MSE), cross entropy loss (for classification problems), and the like. Assuming that a mean square error is used, the loss function L may be defined as: ; wherein W is the number of samples, In order to be an actual tag,Is the predicted value of the model.

Optimization algorithms (e.g., gradient descent, adam, etc.) are then used to adjust the weights and biases of the model to minimize the loss function. The training process comprises the following steps: 1. forward propagation: the output of the model, i.e. seizure prediction score S _total, is calculated. 2. Calculating loss: the error between the current predicted value and the actual tag is calculated using the loss function. 3. Back propagation: the loss function is calculated with respect to weights α, β and γ and is the gradient of bias term b. These gradients represent the effect of each parameter on the total loss. 4. Updating weights: the weights and biases are updated using an optimization algorithm. For example, using a gradient descent algorithm, the updated formula for weights and biases is: ; where η is the learning rate used to control the step size of each update.

During training, the model is evaluated using a validation set to prevent overfitting. By monitoring the loss on the verification set, the super parameters (such as learning rate, regularization parameters and the like) of the model are adjusted, so that the generalization capability of the model is ensured. And by iterative training, the above steps are repeatedly performed until the model converges, i.e. the value of the loss function is minimized on both the training set and the validation set and no longer drops significantly.

The weights α, β and γ and the bias term b are obtained by training a model. By using historical data, defining loss functions, forward and backward propagation, gradient descent, and other optimization algorithms, the model can gradually adjust these weights to minimize prediction errors. These weights ultimately reflect the relative importance of the different signals (electroencephalogram, electrocardiogram and motion data) in seizure prediction, enabling the model to accurately assess the risk of seizures.

Step S6: judging whether the seizure prediction score is larger than or equal to a preset early warning threshold.

Specifically, after model training is completed, a preliminary threshold T is selected to evaluate the performance of the model. The preliminary threshold may be selected based on the distribution of the model outputs, e.g., a median or mean of the output scores is selected as the preliminary threshold. The behavior of the model at different thresholds is then evaluated using Receiver Operating Characteristics (ROC) curve analysis methods. The ROC curve drawing method includes calculating a true case rate (TPR) and a false case rate (FPR) at different thresholds, and drawing these values into a curve. TPR (sensitivity) means that in all cases where seizures actually occur, the model predicts correctly the proportion of seizures; FPR (1-specific) means that in all cases where no seizure actually occurred, the model was mispredicted as the proportion of seizures.

And then determining the optimal threshold value by analyzing the ROC curve. The optimal threshold is typically chosen to be the closest point on the ROC curve to the (0, 1) point, the corresponding threshold being capable of achieving the optimal balance between sensitivity and specificity. The specific method comprises the following steps: the TPR and FPR corresponding to each threshold are first calculated. And calculates the Euclidean distance from each point to the (0, 1) point: . And finally, selecting a threshold value corresponding to the point with the smallest distance as an optimal preset early warning threshold value T.

After the optimal threshold is determined, the selected threshold is verified using an independent verification set or cross-validation method, ensuring that it performs well on different data sets. If the validation results show sensitivity or specificity is not ideal, it may be necessary to adjust the threshold or retrain the model and re-evaluate the threshold. In practical applications, with the continuous accumulation of data and the occurrence of new situations, dynamic adjustment of a preset early warning threshold may be required. The threshold can be periodically re-evaluated and adjusted by continuously monitoring the performance of the model and user feedback to ensure long-term effectiveness and reliability of the early warning system.

Step S7: and when the seizure prediction score is larger than or equal to a preset early warning threshold value, determining that the seizure risk exists in the user to be monitored.

Specifically, when the seizure prediction score is greater than or equal to a preset early warning threshold, it indicates that the user has a higher seizure risk and needs to trigger early warning. When triggering early warning, the system should execute the following operations: the alarm notification is sent to the user or the guardian thereof, and the alarm can be sent in a mode of a smart bracelet, a mobile phone application, a short message and the like so as to ensure that the alarm can be received in time. If the system is connected to a medical facility, the relevant healthcare personnel may be automatically notified, scheduled for emergency intervention or further examination. Thus, when the user suffers from epileptic seizure, medical assistance can be obtained rapidly, and risks and injuries caused by seizure are reduced.

It should be noted that, if the seizure prediction score is smaller than the preset early warning threshold, it indicates that the current seizure risk of the user is lower, and the early warning is not required to be triggered. The system will continue to monitor the physiological signal of the user for the next predictive calculation. The continuous monitoring can ensure that when the state of the user changes, the system can update the prediction result in time and send out early warning when necessary.

Referring to fig. 4, a schematic structural diagram of an epileptic seizure early-warning system provided by an embodiment of the present application is shown, where the system includes an acquisition module 1, a processing module 2, and an early-warning module 3; the acquisition module 1 is used for acquiring a first electroencephalogram signal, a first electrocardiographic signal and first motion data of a user to be monitored; the processing module 2 is used for removing noise interference in the first electroencephalogram signal and the first electrocardiogram signal through a filter to obtain a second electroencephalogram signal and a second electrocardiogram signal; the processing module 2 is further used for carrying out standardized processing on the first motion data to obtain second motion data; the processing module 2 is further used for extracting electroencephalogram signal characteristics from the second electroencephalogram signals, extracting electrocardiographic signal characteristics from the second electrocardiographic signals and extracting exercise data characteristics from the second exercise data; the processing module 2 is also used for obtaining seizure prediction scores based on the electroencephalogram signal characteristics, the electrocardiographic signal characteristics and the motion data characteristics; the processing module 2 is further used for judging whether the seizure prediction score is greater than or equal to a preset early warning threshold value; and the early warning module 3 is used for determining that the user to be monitored has seizure risk when the seizure prediction score is larger than or equal to a preset early warning threshold value.

In a possible implementation manner, the obtaining module 1 is further configured to send a first control signal to a first wearable device of a user to be monitored, so that the first wearable device collects a first electroencephalogram signal at a preset frequency; the acquisition module 1 is further configured to send a second control signal to a second wearable device of a user to be monitored, so that the second wearable device acquires a first electrocardiogram signal at a preset frequency; the acquisition module 1 is further configured to send a third control signal to a third wearable device of the user to be monitored, so that the third wearable device acquires the first motion data at a preset frequency.

In a possible embodiment, the processing module 2 is further configured to determine a first noise type corresponding to the first electroencephalogram signal and a second noise type corresponding to the first electrocardiographic signal; wherein the first noise type comprises a first power line disturbance, an electromyographic signal disturbance, and an eye movement artifact; the second noise type includes second power line interference, baseline wander, and high frequency noise; the processing module 2 is further configured to filter out electromyographic signal interference and noise interference of eye movement artifacts through a band-pass filter, and filter out noise interference of first power line interference through a notch filter, so as to obtain a second electroencephalogram signal; the processing module 2 is further configured to filter noise interference of baseline wander and high-frequency noise through a band-pass filter, and filter noise interference of second power line interference through a notch filter, so as to obtain a second electrocardiogram signal.

In a possible implementation manner, the processing module 2 is further configured to calculate a first mean value and a first standard deviation corresponding to the first X-axis acceleration data, calculate a second mean value and a second standard deviation corresponding to the first Y-axis acceleration data, and calculate a third mean value and a third standard deviation corresponding to the first Z-axis acceleration data; the processing module 2 is further configured to normalize the first X-axis acceleration data based on the first mean value and the first standard deviation, to obtain second X-axis acceleration data; the processing module 2 is further configured to normalize the first Y-axis acceleration data based on the second mean value and the second standard deviation, to obtain second Y-axis acceleration data; the processing module 2 is further configured to normalize the first Z-axis acceleration data based on the third mean value and the third standard deviation, and obtain second Z-axis acceleration data.

In a possible implementation manner, the processing module 2 is further configured to perform normalization processing on the n first sub-features to obtain n normalized first sub-features, perform normalization processing on the m second sub-features to obtain m normalized second sub-features, and perform normalization processing on the m third sub-features to obtain p normalized third sub-features; the processing module 2 is further used for calculating electroencephalogram signal scores based on the n normalized first sub-features; the processing module 2 is further used for calculating an electrocardiogram signal fraction based on the m normalized second sub-features; the processing module 2 is further used for calculating a motion data score based on the p normalized third sub-features; the processing module 2 is further configured to calculate a seizure prediction score based on the electroencephalogram signal score, the electrocardiographic signal score, and the motion data score.

In a possible embodiment, the processing module 2 is further configured to calculate an electroencephalogram signal score according to the following formula: ; wherein, For the electroencephalogram signal fraction,For the first sub-feature after the i-th normalization,The weight corresponding to the first sub-feature after the i-th normalization is given; the processing module 2 is further configured to calculate an electrocardiogram signal fraction according to the following formula: ; wherein, For the fraction of the ecg signal,For the second sub-feature normalized by j,The weight corresponding to the second sub-feature after the j-th normalization is given; the processing module 2 is further configured to calculate a motion data score according to the following formula: ; wherein, In order to score the motion data,For the third sub-feature after the kth normalization,The weight corresponding to the third sub-feature after the k normalization; the processing module 2 is further configured to calculate and obtain a seizure prediction score according to the following formula: ; wherein, For the seizure predictive score,For the electroencephalogram signal fraction,For the fraction of the ecg signal,For the motion data score, α is a first weight, β is a second weight, γ is a third weight, and b is a bias term.

It should be noted that: in the device provided in the above embodiment, when implementing the functions thereof, only the division of the above functional modules is used as an example, in practical application, the above functional allocation may be implemented by different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules, so as to implement all or part of the functions described above. In addition, the embodiments of the apparatus and the method provided in the foregoing embodiments belong to the same concept, and specific implementation processes of the embodiments of the method are detailed in the method embodiments, which are not repeated herein.

The application also discloses electronic equipment. Referring to fig. 5, fig. 5 is a schematic structural diagram of an electronic device according to an embodiment of the present disclosure. The electronic device 500 may include: at least one processor 501, at least one network interface 504, a user interface 503, a memory 505, at least one communication bus 502.

Wherein a communication bus 502 is used to enable connected communications between these components.

The user interface 503 may include a Display screen (Display) and a Camera (Camera), and the optional user interface 503 may further include a standard wired interface and a standard wireless interface.

The network interface 504 may optionally include a standard wired interface, a wireless interface (e.g., WI-FI interface), among others.

Wherein the processor 501 may include one or more processing cores. The processor 501 connects various parts throughout the server using various interfaces and lines, performs various functions of the server and processes data by executing or executing instructions, programs, code sets, or instruction sets stored in the memory 505, and invoking data stored in the memory 505. Alternatively, the processor 501 may be implemented in at least one hardware form of digital signal processing (DigitalSignalProcessing, DSP), field programmable gate array (Field-ProgrammableGateArray, FPGA), programmable logic array (ProgrammableLogicArray, PLA). The processor 501 may integrate one or a combination of several of a central processor (CentralProcessingUnit, CPU), an image processor (GraphicsProcessingUnit, GPU), a modem, etc. The CPU mainly processes an operating system, a user interface, an application program and the like; the GPU is used for rendering and drawing the content required to be displayed by the display screen; the modem is used to handle wireless communications. It will be appreciated that the modem may not be integrated into the processor 501 and may be implemented by a single chip.

The memory 505 may include a random access memory (RandomAccessMemory, RAM) or a Read-only memory (rom). Optionally, the memory 505 comprises a non-transitory computer readable medium (non-transitorycomputer-readablestoragemedium). Memory 505 may be used to store instructions, programs, code sets, or instruction sets. The memory 505 may include a stored program area and a stored data area, wherein the stored program area may store instructions for implementing an operating system, instructions for at least one function (such as a touch function, a sound playing function, an image playing function, etc.), instructions for implementing the above-described various method embodiments, etc.; the storage data area may store data or the like involved in the above respective method embodiments. The memory 505 may also optionally be at least one storage device located remotely from the processor 501. Referring to fig. 5, an operating system, a network communication module, a user interface module, and an application program may be included in the memory 505, which is a computer readable storage medium.

In the electronic device 500 shown in fig. 5, the user interface 503 is mainly used for providing an input interface for a user, and acquiring data input by the user; and the processor 501 may be configured to invoke the memory 505 to store an application program that, when executed by the one or more processors 501, causes the electronic device 500 to perform the method as in one or more of the embodiments described above. It should be noted that, for simplicity of description, the foregoing method embodiments are all described as a series of acts, but it should be understood by those skilled in the art that the present application is not limited by the order of acts described, as some steps may be performed in other orders or concurrently in accordance with the present application. Further, those skilled in the art will also appreciate that the embodiments described in the specification are all of the preferred embodiments, and that the acts and modules referred to are not necessarily required for the present application.

In the foregoing embodiments, the descriptions of the embodiments are emphasized, and for parts of one embodiment that are not described in detail, reference may be made to the related descriptions of other embodiments.

In the several embodiments provided by the present application, it should be understood that the disclosed apparatus may be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative, such as a division of units, merely a division of logic functions, and there may be additional divisions in actual implementation, such as multiple units or components may be combined or integrated into another system, or some features may be omitted, or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be through some service interface, device or unit indirect coupling or communication connection, electrical or otherwise.

The units described as separate units may or may not be physically separate, and units shown as units may or may not be physical units, may be located in one place, or may be distributed over a plurality of network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional unit in the embodiments of the present application may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit. The integrated units may be implemented in hardware or in software functional units.

The integrated units, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a computer readable memory. Based on this understanding, the technical solution of the present application may be embodied essentially or in a part contributing to the prior art or in whole or in part in the form of a software product stored in a memory, comprising several instructions for causing a computer device (which may be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method of the various embodiments of the present application. And the aforementioned memory includes: various media capable of storing program codes, such as a U disk, a mobile hard disk, a magnetic disk or an optical disk.

The above are merely exemplary embodiments of the present disclosure and are not intended to limit the scope of the present disclosure. That is, equivalent changes and modifications are contemplated by the teachings of this disclosure, which fall within the scope of the present disclosure. Other embodiments of the disclosure will be apparent to those skilled in the art from consideration of the specification and practice of the disclosure.

This application is intended to cover any variations, uses, or adaptations of the disclosure following, in general, the principles of the disclosure and including such departures from the present disclosure as come within known or customary practice within the art to which the disclosure pertains. It is intended that the specification and examples be considered as exemplary only, with a scope and spirit of the disclosure being indicated by the claims.

Claims (8)

1. A seizure pre-warning method, the method comprising:

Acquiring a first electroencephalogram signal, a first electrocardiographic signal and first movement data of a user to be monitored;

removing noise interference in the first electroencephalogram signal and the first electrocardiogram signal through a filter to obtain a second electroencephalogram signal and a second electrocardiogram signal;

Performing standardization processing on the first motion data to obtain second motion data;

extracting electroencephalogram signal features from the second electroencephalogram signal, extracting electrocardiographic signal features from the second electrocardiographic signal, and extracting motion data features from the second motion data; the electroencephalogram signal features include n first sub-features; the electrocardiogram signal features include m second sub-features; the motion data feature includes p third sub-features; carrying out standardization processing on the n first sub-features to obtain n standardized first sub-features, carrying out standardization processing on the m second sub-features to obtain m standardized second sub-features, and carrying out standardization processing on the m third sub-features to obtain p standardized third sub-features;

Calculating an electroencephalogram signal score based on the n normalized first sub-features; the electroencephalogram signal score is calculated by the following formula: wherein S _EEG is the electroencephalogram signal fraction, For the ith said normalized first sub-feature, W _i is the weight corresponding to the ith normalized first sub-feature;

calculating an electrocardiogram signal score based on the m normalized second sub-features; the electrocardiographic signal fraction is calculated by the following formula: wherein S _ECG is the electrocardiographic signal fraction, For the j-th said normalized second sub-feature, W _j is the weight corresponding to the j-th normalized second sub-feature; calculating a motion data score based on p of the normalized third sub-features; the motion data score is calculated by the following formula: wherein S _Motion is the motion data score, For the kth said normalized third sub-feature, W _k is the weight corresponding to the kth normalized third sub-feature;

Calculating the seizure prediction score based on the electroencephalogram signal score, the electrocardiographic signal score, and the athletic data score; the seizure prediction score is calculated by the following formula: s _total＝a·S_EEG+β·S_ECG+γ·S_Motion +b; wherein S _total is the seizure prediction score, S _EEG is the electroencephalogram signal score, S _ECG is the electrocardiographic signal score, S _Motion is the motion data score, α is a first weight, β is a second weight, γ is a third weight, and b is a bias term;

Judging whether the epileptic seizure prediction score is larger than or equal to a preset early warning threshold value; the preset early warning threshold is determined by analyzing an ROC curve, and the preset early warning threshold is a point closest to a (0, 1) point on the ROC curve, namely, a real case rate (TPR) and a false case rate (FPR) corresponding to each threshold are calculated, and Euclidean distance from each point to the (0, 1) point is calculated: selecting a threshold value corresponding to a point with the smallest distance as the preset early warning threshold value;

And when the seizure prediction score is larger than or equal to the preset early warning threshold value, determining that the seizure risk exists in the user to be monitored.

2. The method according to claim 1, wherein the acquiring the first electroencephalogram signal, the first electrocardiographic signal, and the first motion data of the user to be monitored, specifically comprises:

Sending a first control signal to a first wearable device of the user to be monitored, so that the first wearable device collects the first electroencephalogram signal at a preset frequency;

Sending a second control signal to a second wearable device of the user to be monitored, so that the second wearable device acquires the first electrocardiogram signal at the preset frequency;

and sending a third control signal to third wearable equipment of the user to be monitored, so that the third wearable equipment collects the first motion data at the preset frequency.

3. The method of claim 1, wherein the filter comprises a bandpass filter and a notch filter; the removing, by a filter, noise interference in the first electroencephalogram signal and the first electrocardiograph signal to obtain a second electroencephalogram signal and a second electrocardiograph signal specifically includes:

Determining a first noise type corresponding to the first electroencephalogram signal and a second noise type corresponding to the first electrocardiograph signal; wherein the first noise type comprises first power line interference, electromyographic signal interference, and eye movement artifacts; the second noise type includes second power line interference, baseline wander, and high frequency noise;

Filtering the electromyographic signal interference and the noise interference of the eye movement artifact through the band-pass filter, and filtering the noise interference of the first power line interference through the notch filter to obtain the second electroencephalogram signal;

And filtering noise interference of the baseline drift and the high-frequency noise through the band-pass filter, and filtering noise interference of the second power line interference through a notch filter to obtain the second electrocardiogram signal.

4. The method of claim 1, wherein the first motion data comprises first X-axis acceleration data, first Y-axis acceleration data, and first Z-axis acceleration data; the second motion data includes second X-axis acceleration data, second Y-axis acceleration data, and second Z-axis acceleration data; the step of carrying out standardization processing on the first motion data to obtain second motion data specifically comprises the following steps:

Calculating a first mean value and a first standard deviation corresponding to the first X-axis acceleration data, calculating a second mean value and a second standard deviation corresponding to the first Y-axis acceleration data, and calculating a third mean value and a third standard deviation corresponding to the first Y-axis acceleration data;

normalizing the first X-axis acceleration data based on the first mean value and the first standard deviation to obtain the second X-axis acceleration data;

normalizing the first Y-axis acceleration data based on the second mean and the second standard deviation to obtain second Y-axis acceleration data;

and normalizing the first Z-axis acceleration data based on the third mean value and the third standard deviation to obtain the second Z-axis acceleration data.

5. The method of claim 1, wherein the electroencephalogram signal characteristics include power spectral density, average potential value of an electroencephalogram signal, degree of dispersion of an electroencephalogram signal, maximum amplitude of an electroencephalogram signal, amplitude variation range of an electroencephalogram signal, degree of complexity of an electroencephalogram signal, sample entropy, and approximate entropy; the electrocardiogram signal features comprise RR interval, QT interval, time domain index of HRV, frequency domain index of HRV, poincare image features and sample entropy; the motion data features include an average of the motion data, a variance of the motion data, a peak of the motion data, an amplitude variation range of the motion data, a gesture variation, and a motion pattern.

6. A seizure early-warning system for performing the seizure early-warning method of claim 1, the system comprising: the system comprises an acquisition module, a processing module and an early warning module;

The acquisition module is used for acquiring a first electroencephalogram signal, a first electrocardiographic signal and first motion data of a user to be monitored; the processing module is used for removing noise interference in the first electroencephalogram signal and the first electrocardiograph signal through a filter to obtain a second electroencephalogram signal and a second electrocardiograph signal;

the processing module is further used for carrying out standardized processing on the first motion data to obtain second motion data;

The processing module is further used for extracting electroencephalogram signal characteristics from the second electroencephalogram signals, extracting electrocardiographic signal characteristics from the second electrocardiographic signals and extracting motion data characteristics from the second motion data; the electroencephalogram signal features include n first sub-features; the electrocardiogram signal features include m second sub-features; the motion data feature includes p third sub-features;

The processing module is further configured to perform normalization processing on the n first sub-features to obtain n normalized first sub-features, perform normalization processing on the m second sub-features to obtain m normalized second sub-features, and perform normalization processing on the m third sub-features to obtain p normalized third sub-features;

the processing module is further used for calculating electroencephalogram signal scores based on n normalized first sub-features;

the processing module is further used for calculating an electrocardiogram signal fraction based on m normalized second sub-features;

The processing module is further used for calculating a motion data score based on p normalized third sub-features; the processing module is further used for calculating an epileptic seizure prediction score based on the electroencephalogram signal score, the electrocardiographic signal score and the motion data score;

the processing module is further used for judging whether the seizure prediction score is greater than or equal to a preset early warning threshold value;

And the early warning module is used for determining that the user to be monitored has seizure risk when the seizure prediction score is greater than or equal to the preset early warning threshold.

7. An electronic device comprising a processor (501), a memory (505), a user interface (503) and a network interface (504), the memory (505) for storing instructions, the user interface (503) and the network interface (504) for communicating to other devices, the processor (501) for executing the instructions stored in the memory (505) to cause the electronic device (500) to perform the method according to any of claims 1-5.

8. A computer readable storage medium storing instructions which, when executed, perform the method of any one of claims 1-5.