Method for adjusting model parameters based on electroencephalographic signals, epilepsy treatment device
By adjusting the parameters of the EEG signal model in different time phases, the problem of poor data feature matching was solved, the accuracy of epileptic seizure detection was improved and the false alarm rate was reduced, and more efficient epilepsy treatment was achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BRIGHTON INTELLIGENT TECHNOLOGY (CHANGZHOU) CO LTD
- Filing Date
- 2026-04-15
- Publication Date
- 2026-06-05
AI Technical Summary
In existing technologies, the data feature matching degree of EEG signal models in different time phases is poor, resulting in high false alarm and false negative rates, making it difficult to effectively train model parameters suitable for each time phase.
By dividing the training data into data from the attack period, the interictal period, and the pre-attack period, loss factors such as the false negative rate, false alarm rate, and alarm time are obtained respectively, and the model is trained twice to adjust the parameters of the detection model, so as to ensure that the model parameters are more matched with the data features.
It improves the accuracy of epileptic seizure detection and reduces the false alarm rate, avoids unnecessary electrical stimulation, and enhances the model's adaptability in different temporal states.
Smart Images
Figure CN122141120A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of physiological electrical signal technology, specifically relating to a method for adjusting model parameters based on electroencephalogram (EEG) signals and an epilepsy treatment device. Background Technology
[0002] Neuromodulation is an effective treatment for drug-resistant epilepsy. Reactive neurostimulation uses an EEG-based closed-loop detection algorithm to monitor seizures in real time. Applying electrical stimulation during a seizure can interrupt the propagation of abnormal discharges during the seizure, thereby terminating the seizure. This approach minimizes false alarms from the closed-loop detection algorithm and avoids burns or seizure induction caused by continuous electrical stimulation (DBS).
[0003] In related technologies, models are generally trained based on sensitivity and false alarm rate. Sensitivity refers to the ratio of the number of epileptic seizures successfully predicted by the system to the total number of actual seizures, used to evaluate the ability of the warning model to successfully predict epileptic seizures. False alarm rate refers to the ratio of the number of false alarms issued by the system to the total testing time, usually expressed as the number of false alarms per hour, used to measure the frequency of false alarms issued by the warning model. The temporal states of EEG signals generally include the interictal period, the preictal period, and the ictal period. The different durations and levels of attention of each temporal state result in different data sample sizes for training the model, and the different contributions of the data features of each temporal state to the training model lead to different optimal model parameters. For example, CN120296656A discloses a method and system for epilepsy early warning based on electroencephalogram (EEG) signals. The specification states that the collected EEG signal data contains a total of N epileptic seizure events, and the time point of each seizure is recorded. The patient's continuous EEG signal data is divided into several seizure data sets for training and testing. If N≤5, the first N-1 seizure data are used for model training; if N>5, the first N-2 seizure data are used for model training; the remaining data is used for model testing. The method utilizes an adaptive overlapping data slicing method using a sliding window and a focus loss function, dynamically adjusting sample weights by introducing a modulation factor to reduce the contribution of multiple sample classes to the loss function while enhancing the weight of fewer sample classes, thus eliminating class imbalance. Summary of the Invention
[0004] This invention provides a method for adjusting model parameters based on electroencephalogram (EEG) signals and an epilepsy treatment device to solve the problem of matching phase states with model parameters.
[0005] To address the aforementioned technical problems, this invention provides a method for adjusting model parameters based on electroencephalogram (EEG) signals, comprising: dividing training data into continuous episodic data, interictal data, and pre-episodic data according to temporal phase states; obtaining loss factors based on temporal phase states, namely, obtaining the false negative rate and detection delay time based on episodic data, obtaining the false alarm count and false alarm rate based on interictal data, and obtaining the alarm time based on pre-episodic data; obtaining alternative parameters based on the loss factors; and adjusting the parameters of the detection model according to the alternative parameters.
[0006] Furthermore, the training data includes first training data and second training data; the step of obtaining candidate parameters based on the loss factor includes: a first model training, i.e., obtaining a first model based on the first loss factor corresponding to the first training data; and a second model training, i.e., obtaining candidate parameters based on the second loss factor corresponding to the second training data.
[0007] Furthermore, the single model training includes: setting the first loss factor as the false alarm rate. Detection delay time False negative rate Set the first loss value ,in , These represent the weighting coefficients of each first loss factor; the preset parameters of the initial model are set, i.e., the parameter grid of the initial model is established based on the EEG energy and duration window number; all preset parameters in the parameter grid of the initial model are traversed, and the corresponding first loss values are obtained. ; the first loss value The preset parameters corresponding to the minimum value are used as initial parameters to obtain the first model.
[0008] Furthermore, the secondary model training includes: setting the second loss factor to the number of false alarms. Detection delay time False negative rate Alarm time Set the preset parameters of the first model, that is, establish the parameter grid of the first model based on the EEG energy and duration window number; traverse all the preset parameters of the first model in the parameter grid to obtain the corresponding local candidate parameters; take the intersection of each local parameter as the candidate parameter; wherein obtaining the corresponding local candidate parameter includes: obtaining the first local parameter based on the number of false alarms; obtaining the second local parameter by combining the detection delay time and the false alarm rate; obtaining the third local parameter based on the alarm time.
[0009] Furthermore, obtaining the first local parameter based on the number of false alarms includes: obtaining the number of false alarms corresponding to all preset parameters of the first model. Set a threshold for the number of false alarms based on the expected number of stimuli. ;set up The first local parameters are obtained by filtering the preset parameters of the first model; the second local parameters are obtained by combining the detection delay time and the false negative rate, including setting the detection effect loss value. ,in These represent the weighting coefficients of each of the second loss factors; the detection performance loss value corresponding to the initial parameters of the first model is obtained. Obtain the detection performance loss values corresponding to all preset parameters of the first model. ;set up The first model's preset parameters are used as the second local parameters; the third local parameters are obtained based on the alarm time, including obtaining the alarm time corresponding to all preset parameters of the first model. According to the alarm time Average number of alarms obtained from interval data According to the alarm time Average number of alarms obtained from pre-attack data ;set up The preset parameters of the first model are used as the third local parameters.
[0010] Furthermore, the parameters of the detection model are adjusted based on the alternative parameters, and the parameter with the number of false alarms closest to the expected number of stimuli is selected as the parameter of the first model.
[0011] The present invention also provides an epilepsy treatment device, comprising: a processor for running the method to obtain parameters of a detection model; an electrode sensor for acquiring online data; the detection model for detecting the temporal state of the online data; and an electrical stimulation module for performing electrical stimulation according to the temporal state.
[0012] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method.
[0013] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method.
[0014] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the method.
[0015] The beneficial effects of this invention are that the method and epilepsy treatment device based on EEG signals for adjusting model parameters first obtain a loss factor based on the temporal state of the training data, namely, obtaining the missed detection rate and detection delay time based on the seizure period data, obtaining the number of false alarms based on the interictal period data, and obtaining the alarm time based on the pre-seizure data. Then, candidate parameters are obtained based on the loss factor. Finally, the parameters of the detection model are adjusted based on the candidate parameters, combining the loss factor with the temporal state of the data to train the model parameters. This allows for the targeted selection of the temporal state of the data samples and the appropriate detection model according to the detection requirements, making the model parameters more closely match the data features, and eliminating the need to consider the weight of sample categories.
[0016] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. To make the foregoing objects, features, and advantages of the invention more apparent and understandable, preferred embodiments are described in detail below with reference to the accompanying drawings. Attached Figure Description
[0017] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0018] Figure 1 This is a flowchart illustrating the method for adjusting model parameters.
[0019] Figure 2 This is a schematic diagram of selecting the first local parameter in the parameter mesh of the first model.
[0020] Figure 3 This is a schematic diagram of selecting the second local parameter in the parameter mesh of the first model.
[0021] Figure 4 This is a schematic diagram of selecting the third local parameter in the parameter mesh of the first model.
[0022] Figure 5 This is a block diagram illustrating the principle of an epilepsy treatment device. Detailed Implementation
[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0024] See Figures 1-4 At least one embodiment provides a method for adjusting model parameters based on electroencephalogram (EEG) signals, comprising: dividing training data into continuous episodic data, interictal data, and pre-episodic data according to the temporal state; obtaining loss factors according to the temporal state, namely, obtaining the false negative rate and detection delay time according to the episodic data, obtaining the false alarm number and false alarm rate according to the interictal data, and obtaining the alarm time according to the pre-episodic data; obtaining alternative parameters based on the loss factors; and adjusting the parameters of the detection model according to the alternative parameters.
[0025] Optionally, the training data can be understood as continuously collected historical EEG signal data. The start and end times of the seizure are manually marked, and the data from the first preset time before the start time to the second preset time before the start time is marked as interval data, the data from the second preset time before the start time to the start time is marked as pre-seizure data, and the data from the start time to the end time is marked as seizure data. For example, the first preset time is generally set to 30 minutes or more, that is, the data from 30 minutes to 5 minutes before the start time is defined as interval data, and the second preset time is generally set to 5 minutes or more, that is, the data from 5 minutes before the start time to the start time is defined as pre-seizure data.
[0026] In some embodiments, to design targeted stimulation protocols, doctors typically focus on data near the seizure time in the early stages of epilepsy detection. This results in varying historical sample sizes for the same training model across different time phases, and insufficient interictal data to obtain false alarm counts for model training. Therefore, in this case, the training data can be defined as first training data and second training data. First training data can be understood as manually labeled historical data used when designing the initial stimulation protocol. This data is characterized by a large amount of pre-seizure and seizure-period data, and a smaller amount of interictal data. It is generally used to obtain initial parameters of the model or to train the initial model. It primarily utilizes pre-seizure and seizure-period data to train the model, reducing the false negative rate and detection delay time, while also considering the false alarm rate to a small extent. Therefore, the matching degree of the model parameters is comprehensively evaluated using the first loss values of each loss factor to select the initial parameters of the first model. Second training data can be understood as manually labeled historical data used when adjusting the initial stimulation protocol. This data is characterized by an increased accumulation of interictal data, leading to an increase in the statistics of alarm times and false alarm counts. More data can be used to obtain local candidate parameters corresponding to each second loss factor, and then the parameters of the second model are selected from these local candidate parameters. Specifically, obtaining candidate parameters based on the loss factor includes: a first model training, i.e., obtaining a first model based on the first loss factor corresponding to the first training data; and a second model training, i.e., obtaining candidate parameters based on the second loss factor corresponding to the second training data.
[0027] Optionally, the first model training includes: setting the first loss factor as the false alarm rate. Detection delay time False negative rate Set the first loss value ,in , These represent the weighting coefficients of each first loss factor; the preset parameters of the initial model are set, i.e., the parameter grid of the initial model is established based on the EEG energy and duration window number; all preset parameters in the parameter grid of the initial model are traversed, and the corresponding first loss values are obtained. ; the first loss value The preset parameters corresponding to the minimum value are used as initial parameters to obtain the first model.
[0028] Specifically, weighting coefficients , These represent the different importance of the first loss factor in evaluating the initial model, and are required to be normalized. Then, the initial parameters of the first model are confirmed through grid search. During the actual training of a model, the weighting coefficients are adjusted within a certain range according to the importance placed on different first loss factors. The adjustment range can be determined based on the constraints of each first loss factor. Specifically, as follows.
[0029] False alarm rate =Number of stimuli / Number of non-stimuli. Based on commonly used parameters, the maximum ratio of stimuli to non-stimuli is set to 1:5, i.e. Limited to when At that time, by setting weighting coefficients In order to ensure ( ) 1, i.e., the first loss value Set the maximum value to 1 to meet the limit.
[0030] Detection delay time (Unit: seconds): The range of weighting coefficients is determined based on the maximum tolerable detection delay. For example, when the maximum detection delay time... When the time is 10 seconds, it can be limited to when At this time, the weighting coefficient can be increased and set. In order to ensure ( ) 1, i.e., the first loss value Set the maximum value to 1 to meet the limit.
[0031] False negative rate =1 - (Number of stimuli during an attack / Number of samples during an attack). The range of the coefficient is determined based on the maximum tolerable false negative rate. For example, when the maximum false negative rate is 0.8, the range can be limited to... At that time, set In order to ensure ( ) 1, i.e., the first loss value Set the maximum value to 1 to meet the limit.
[0032] Optionally, the secondary model training includes: setting the second loss factor to the number of false alarms. Detection delay time False negative rate Alarm time Set the preset parameters of the first model, that is, establish the parameter grid of the first model based on the EEG energy and duration window number. For example, set the EEG energy to 51%~150% of the energy parameter of the first model, and the window number threshold to 1-10. Traverse all the preset parameters of the first model in the parameter grid to obtain the corresponding local candidate parameters. Take the intersection of each local parameter as the candidate parameter. The process of obtaining the corresponding local candidate parameters includes: obtaining the first local parameter based on the number of false alarms; obtaining the second local parameter by combining the detection delay time and the false negative rate; and obtaining the third local parameter based on the alarm time.
[0033] Optionally, obtaining the first local parameter based on the number of false alarms includes: obtaining the number of false alarms corresponding to all preset parameters of the first model. Based on the expected stimulus amount Set a threshold for the number of false alarms. ;set up The preset parameters of the first model are used as the first local parameters, for example, setting... .See Figure 2 In the parameter grid of the first model, the right-hand axis is a color bar, representing the number of false alarms throughout the day at the corresponding threshold. The redder the color, the more stimuli are detected. The number of false alarms during the 24-hour interval between attacks is statistically analyzed, and the expected number of stimuli is set. The preset parameter is 2000, and only the number of false alarms between 1000 and 3000 is selectable. Figure 2 Within the green box area, the window number threshold is between 8 and 10, and all selectable energy thresholds meet the requirements.
[0034] Optionally, obtaining the second local parameter by combining the detection delay time and the false negative rate includes: setting the detection effect loss value. ,in These represent the weighting coefficients of each of the second loss factors; the detection performance loss value corresponding to the initial parameters of the first model is obtained. Obtain the detection performance loss values corresponding to all preset parameters of the first model. ;set up The preset parameters of the first model are used as the second local parameters. See Figure 3 In the parameter grid of the first model, the right-hand axis is a color bar, representing the change in loss at the corresponding threshold. The redder the color, the larger the value. The green box specifies the threshold where the change in loss is less than 0. The change in loss for the detection effect of the statistical interval data is... Only the preset parameter for reducing the loss value is selectable, i.e. Figure 3 Within the green box range, the window number threshold is between 1 and 9, and all selectable energy thresholds meet the requirements.
[0035] Optionally, obtaining the third local parameters based on the alarm time includes: obtaining the alarm time corresponding to all preset parameters of the first model. According to the alarm time Average number of alarms obtained from interictal data (times / minute); based on alarm time Average number of alarms obtained from pre-attack data ;set up The preset parameters of the first model are used as the third local parameters, for example, the requirements are... .See Figure 4 In the parameter grid of the first model, the right-hand axis is a color bar, representing the ratio of the number of alarms before an attack to the number of alarms between attacks at the corresponding threshold. The redder the color, the larger the value. The green box indicates the threshold where the ratio is greater than 1.5. This represents the average number of alarms recorded 0-5 minutes before the onset of the attack, within the 30-minute data period. The average number of alarms between 30 and 5 minutes (interictal period) The ratio is selectable only for preset parameters with a ratio greater than 1.5. Figure 4 The area within the green box. All selectable energy thresholds between window number thresholds of 8-10 meet the requirements; when the window number threshold is 6, energy thresholds of numbers 43-68 meet the requirements; when the window number threshold is 7, energy thresholds of numbers 38-44 meet the requirements.
[0036] Optionally, the parameters of the detection model are adjusted based on the alternative parameters by selecting the parameter whose false alarm count is closest to the expected number of stimuli from the alternative parameters as the parameters of the first model.
[0037] In some embodiments, an epilepsy treatment device is also provided, comprising: a processor for running the method to acquire parameters of a detection model; an electrode sensor for acquiring online data; the detection model for detecting the temporal state of the online data; and an electrical stimulation module for performing electrical stimulation according to the temporal state. Specifically, online electroencephalogram (EEG) signals are acquired and monitored using a first model and its corresponding parameters. When a seizure is detected, the processor controls the electrical stimulation module to perform electrical stimulation.
[0038] See Figure 5 The epilepsy treatment device includes structural components and hardware circuitry. The method operates on a host computer and its processor. Signal input is provided by electrode sensors, amplifiers, and analog-to-digital converters. Details are as follows.
[0039] (1) Electrode sensor, on which the acquisition module is used to acquire the original analog EEG signal, the amplifier module (hardware analog circuit) amplifies the original analog EEG signal, and the analog-to-digital conversion module (hardware) converts the amplified analog EEG signal into a digital signal at a fixed sampling rate.
[0040] (2) Host computer (PC hardware + PC software, MCU / ARM / DSP hardware + embedded software) stores, forwards, and processes digital signals. The host computer is equipped with a processor and a detection model to process digital signals, output the detection results of the time phase, and display them to the user in conjunction with the UI workflow. The processor implements the steps of the method of adjusting model parameters based on EEG signals by executing the computer program.
[0041] (3) Display module: presents the UI interface, which is user-operable, and allows setting loss factors or displaying the parameters of the detection model.
[0042] In some embodiments, a computer device is also provided, including a memory, a processor, and a computer program stored on the memory. The processor executes the computer program to implement the steps of the method. The processor may be a central processing unit (CPU), an ASIC, or one or more integrated circuits configured to implement embodiments of the present invention. In specific implementations, if the memory and processor are implemented independently, they can be interconnected via a bus to communicate with each other. The bus may be an Industry Standard Architecture (ISA) bus, a PCI bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. If the memory and processor are integrated on a single chip, they can communicate with each other through an internal interface.
[0043] In some embodiments, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed by a processor, implements the steps of the method. The storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.
[0044] In some embodiments, a computer program product is also provided, including a computer program that, when executed by a processor, implements the steps of the method. The computer program may include program code, which includes computer operation instructions and may be stored in a computer-readable storage medium. Based on this understanding, when the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be implemented in the form of a software product or sold or used as an independent product, the computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the method described in the various embodiments of the present invention.
[0045] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and there may be other division methods in actual implementation. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed.
[0046] Based on the above-described preferred embodiments of the present invention, and through the above description, those skilled in the art can make various changes and modifications without departing from the technical concept of the present invention. That is, the technical scope of the present invention is not limited to the contents of the specification.
Claims
1. A method for adjusting model parameters based on electroencephalogram (EEG) signals, characterized in that, include: The training data is divided into continuous episodic data, interictal data, and pre-epidemic data based on the temporal state. Loss factors are obtained based on the temporal state, namely, the missed detection rate and detection delay time are obtained based on the data during the attack period, the number of false alarms and the false alarm rate are obtained based on the data during the interval, and the alarm time is obtained based on the data before the attack. Obtain alternative parameters based on the loss factor; Adjust the parameters of the detection model based on the alternative parameters.
2. The method according to claim 1, characterized in that, The training data includes first training data and second training data; The method of obtaining alternative parameters based on the loss factor includes: One model training session is to obtain the first model based on the first loss factor corresponding to the first training data. Secondary model training involves obtaining alternative parameters based on the second loss factor corresponding to the second training data.
3. The method according to claim 2, characterized in that, The single model training includes: Set the first loss factor to the false alarm rate. Detection delay time False negative rate ; Set the first loss value ,in , These represent the weighting coefficients of each first loss factor; Set the preset parameters for the initial model, that is, establish the parameter grid of the initial model based on the EEG energy and duration window number; Iterate through all the preset parameters in the parameter grid of the initial model and obtain the corresponding first loss value. ; The first loss value The preset parameters corresponding to the minimum value are used as initial parameters to obtain the first model.
4. The method according to claim 3, characterized in that, The secondary model training includes: Set the second loss factor to the number of false alarms. Detection delay time False negative rate Alarm time ; Set the preset parameters for the first model, that is, establish the parameter grid of the first model based on the EEG energy and duration window number; Traverse all preset parameters of the first model in the parameter grid to obtain the corresponding local alternative parameters; The intersection of all local parameters is used as the candidate parameter; where The process of obtaining the corresponding local candidate parameters includes: The first local parameter is obtained based on the number of false alarms; The second local parameter is obtained by combining the detection delay time and the false detection rate; The third local parameter is obtained based on the alarm time.
5. The method according to claim 4, characterized in that, The first local parameters obtained based on the number of false alarms include: Obtain the number of false alarms corresponding to all preset parameters of the first model. ; Set a threshold for the number of false alarms based on the expected number of stimuli. ; set up The preset parameters of the first model are used as the first local parameters. The second local parameters obtained by combining detection delay time and false negative rate include: Set the detection effect loss value ,in These represent the weighting coefficients of each second loss factor; Obtain the detection performance loss value corresponding to the initial parameters of the first model. ; Obtain the detection performance loss values corresponding to all preset parameters of the first model. ; set up The preset parameters of the first model are used as the second local parameters; The third local parameters obtained based on the alarm time include: Obtain the alarm times corresponding to all preset parameters of the first model. ; According to the alarm time Average number of alarms obtained from interval data ; According to the alarm time Average number of alarms obtained from pre-attack data ; set up The preset parameters of the first model are used as the third local parameters.
6. The method according to claim 5, characterized in that, The parameters of the detection model were adjusted based on the alternative parameters, and the parameter with the closest false alarm count to the expected stimulus count was selected as the parameter of the first model.
7. An epilepsy treatment device, characterized in that, include: The processor runs the method as described in any one of claims 1-6 to obtain parameters of the detection model; Electrode sensors are used to collect online data; The detection model is used to detect the temporal state of online data; The electrical stimulation module performs electrical stimulation according to the phase state.
8. A computer device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method as claimed in any one of claims 1-6.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1-6.
10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1-6.