Epilepsy prediction model algorithm based on artificial intelligence technology

Abstract

The invention relates to the technical field of epilepsy detection, and discloses an epilepsy prediction model algorithm based on an artificial intelligence technology, which comprises the following steps: an epilepsy electroencephalogram data set is collected, the data set comprises five subsets, each subset comprises 100 single-channel electroencephalogram segments, the length of each segment is 23.6 seconds, two of the subsets are from healthy volunteers, and each subset is from the healthy volunteers; the other three subsets are from epilepsy patients. According to the epilepsy prediction model algorithm based on the artificial intelligence technology, frequency spectrum components of signals are extracted through fast Fourier transform, and significant features are mined through feature selection based on LASSO. Manhattan distance is used to assess importance and weight LASSO penalty terms of features. Due to the form of a linear function, the Manhattan distance is insensitive to noise, so that the Manhattan distance can keep stable performance in a noisy environment.

Description

Technical Field

[0001] This invention relates to the field of epilepsy detection technology, specifically to an epilepsy prediction model algorithm based on artificial intelligence technology. Background Technology

[0002] Epilepsy is a chronic brain disorder characterized by recurrent seizures. It is caused by abnormal electrical activity in the brain's neurons. Seizures are recurrent and brief, and can be detected by electroencephalography (EEG), a non-invasive method that records the brain's electrical activity to detect seizures. During a seizure, the EEG typically shows abnormal patterns of electrical activity.

[0003] Currently, automated detection of neurological diseases such as epilepsy based on electroencephalogram (EEG) signals has proven effective with the application of powerful feature learning methods. However, practical applications often encounter challenges caused by noise contamination in EEG signals. To address the problems related to noise interference and feature redundancy, this invention proposes an epilepsy prediction model algorithm based on artificial intelligence technology. Summary of the Invention

[0004] To address the aforementioned shortcomings of existing technologies, this invention provides an epilepsy prediction model algorithm based on artificial intelligence technology.

[0005] This invention provides the following technical solution: an epilepsy prediction model algorithm based on artificial intelligence technology, comprising the following steps:

[0006] S1. Collect an epilepsy EEG dataset, which contains five subsets. Each subset contains 100 single-channel EEG segments, each segment being 23.6 seconds long. Two subsets are from healthy volunteers, and the other three subsets are from epilepsy patients.

[0007] S2. Through FFT, we can convert the EEG signal from the time domain to the frequency domain and obtain the energy distribution of each frequency component;

[0008] S3. Employ a LASSO-based feature selection method to mine salient features distributed in high-dimensional data, and then process the frequency domain.

[0009] The LASSO objective function can be expressed as:

[0010]

[0011] The dataset is represented in matrix form as X = (x1, x2, ..., xp) ∈ Rn × p, where each xi = (x1i, x2i, ..., xni) T ∈ Rn × 1 is a feature vector, y = (y1, y2, ..., yn) T ∈ Rn is the corresponding response of the sample, and β = (β1, β2, ..., βp) T ∈ Rp is an unknown regression coefficient vector, where di represents the Manhattan distance between the i-th feature and the i-th class label;

[0012] S4. Use a support vector machine with radial basis function (RBF) kernel as a classifier to extract features related to epileptic seizures.

[0013] Preferably, two subsets from healthy volunteers were recorded while awake and relaxed, with eyes open and closed, respectively, using the international 10-20 system for scalp electrode distribution; while the other three subsets from epilepsy patients included EEG signals recorded from the epileptic region and from the contralateral hippocampus, as well as EEG signals during seizures.

[0014] Preferably, when the change in a feature is more relevant to the class label, di approaches 0; such features are considered to have higher importance and therefore receive a lower penalty. Conversely, when the change in a feature is unrelated to the class label, di approaches 1.

[0015] Compared with the prior art, the present invention has the following beneficial effects:

[0016] This AI-based epilepsy prediction model algorithm employs Fast Fourier Transform (FFT) to extract the spectral components of the signal and applies LASSO-based feature selection to mine salient features. Manhattan distance is used to evaluate feature importance and weight the LASSO penalty term. Due to its linear function form, Manhattan distance is insensitive to noise, enabling it to maintain stable performance in noisy environments. Through adaptive weighting, the model achieves stable and noise-resistant feature selection capabilities, avoiding the instability and redundant features inherent in LASSO-based feature selection. Subsequently, a support vector machine with radial basis function kernels is used to detect epilepsy using the obtained feature vectors. Detailed Implementation

[0017] In order to make the objectives, technical solutions and advantages of the embodiments of this disclosure clearer, the technical solutions of the embodiments of this disclosure will be clearly and completely described. In order to keep the following description of the embodiments of this disclosure clear and concise, detailed descriptions of known functions and known components are omitted to avoid unnecessarily obscuring the concept of the present invention.

[0018] An epilepsy prediction model algorithm based on artificial intelligence technology includes the following steps:

[0019] S1. Collect an epilepsy EEG dataset, which contains five subsets. Each subset contains 100 single-channel EEG segments, each segment being 23.6 seconds long. Two subsets are from healthy volunteers, and the other three subsets are from epilepsy patients.

[0020] Two subsets of healthy volunteers were recorded while awake and relaxed, with eyes open and closed, respectively, using the international 10-20 system for scalp electrode distribution. The other three subsets of epilepsy patients included EEG signals recorded from the epileptic region and from the contralateral hippocampus, as well as EEG signals during seizures.

[0021] The structure and content of the dataset enable us to effectively apply machine learning algorithms for feature extraction and classification to distinguish between EEG signals during and outside of seizures. Through further data processing and feature selection, we can train a high-performance epilepsy detection model. Furthermore, because the dataset contains EEG data under various conditions and states, we can also investigate the influence of different factors on the occurrence of epilepsy, thereby gaining a deeper understanding of the pathological mechanisms of epilepsy.

[0022] S2. Through FFT, we can convert the EEG signal from the time domain to the frequency domain and obtain the energy distribution of each frequency component;

[0023] Electroencephalogram (EEG) signals are characterized by weak amplitude and low signal-to-noise ratio (SNR), making them highly susceptible to noise interference, such as eye movement and muscle artifacts. Time-domain and time-frequency domain methods struggle to detect epileptic seizures from artifacts with similar time-frequency patterns. However, in the frequency domain, noise interference is primarily concentrated within specific frequency bands. Converting the time-domain signal to the frequency-domain signal can partially avoid noise contamination of the features.

[0024] S3. Employ a LASSO-based feature selection method to mine salient features distributed in high-dimensional data, and then process the frequency domain.

[0025] The LASSO objective function can be expressed as:

[0026]

[0027] The dataset is represented in matrix form as X = (x1, x2, ..., xp) ∈ Rn × p, where each xi = (x1i, x2i, ..., xni) T ∈ Rn × 1 is a feature vector, y = (y1, y2, ..., yn) T ∈ Rn is the corresponding response of the sample, and β = (β1, β2, ..., βp) T ∈ Rp is an unknown regression coefficient vector, where di represents the Manhattan distance between the i-th feature and the i-th class label;

[0028] When a feature's change is more relevant to the class label, di approaches 0; such features are considered more important and therefore receive a lower penalty. Conversely, when a feature's change is unrelated to the class label, di approaches 1.

[0029] S4. Use a support vector machine with radial basis function (RBF) kernel as a classifier to extract features related to epileptic seizures.

[0030] Support Vector Machines (SVMs) are linear classifiers that construct one or more hyperplanes for classification, regression, or other tasks. The strategy of SVMs is to move points close to the hyperplane as far away from it as possible. Compared to logistic regression and neural networks, SVMs offer a clearer and more powerful way to learn complex nonlinear equations. In this work, considering detection performance and computational cost, SVMs using radial basis function kernels are considered to achieve the best performance.

[0031] The above embodiments are merely exemplary embodiments of the present invention and are not intended to limit the present invention. The scope of protection of the present invention is defined by the claims. Those skilled in the art can make various modifications or equivalent substitutions to the present invention within its spirit and scope of protection, and such modifications or equivalent substitutions should also be considered to fall within the scope of protection of the present invention.

Claims

1. An epilepsy prediction model algorithm based on artificial intelligence technology, characterized in that, Includes the following steps: S1. Collect an epilepsy EEG dataset, which contains five subsets. Each subset contains 100 single-channel EEG segments, each segment being 23.6 seconds long. Two subsets are from healthy volunteers, and the other three subsets are from epilepsy patients. S2. Through FFT, we can convert the EEG signal from the time domain to the frequency domain and obtain the energy distribution of each frequency component; S3. Employ a LASSO-based feature selection method to mine salient features distributed in high-dimensional data, and then process the frequency domain. The LASSO objective function can be expressed as: The dataset is represented in matrix form as X = (x1, x2, ..., xp) ∈ Rn × p, where each xi = (x1i, x2i, ..., xni) T ∈ Rn × 1 is a feature vector, y = (y1, y2, ..., yn) T ∈ Rn is the corresponding response of the sample, and β = (β1, β2, ..., βp) T ∈ Rp is an unknown regression coefficient vector, where di represents the Manhattan distance between the i-th feature and the i-th class label; S4. Use a support vector machine with radial basis function (RBF) kernel as a classifier to extract features related to epileptic seizures.

2. The epilepsy prediction model algorithm based on artificial intelligence technology according to claim 1, characterized in that: Two subsets of healthy volunteers were recorded while awake and relaxed, with eyes open and closed, using the international 10-20 system for scalp electrode distribution; while three other subsets of patients with epilepsy included EEG signals recorded from the epileptic region and from the contralateral hippocampus, as well as EEG signals during seizures.

3. The epilepsy prediction model algorithm based on artificial intelligence technology according to claim 1, characterized in that: When a feature's change is more relevant to the class label, di approaches 0; such features are considered more important and therefore receive a lower penalty. Conversely, when a feature's change is unrelated to the class label, di approaches 1.

Images