EEG machine learning-based autism spectrum disorder prediction method

Abstract

The invention relates to the technical field of neuroscience and artificial intelligence, in particular to an autism spectrum disorder prediction method based on EEG machine learning, which comprises the following steps: step 1, collecting EEG data for children with autism spectrum disorder and healthy control children; step 2, preprocessing the collected EEG data, then extracting sleep spindle wave signals in the EEG data, and analyzing the sleep spindle wave signals to obtain feature information data; step 3, performing standardization processing on the obtained feature information data to obtain a data set, and then dividing the data set into a training set and a test set; 4, training the training set through different machine learning algorithms to obtain corresponding training models, and selecting an optimal model from the training models through a five-fold cross validation strategy; and 5, inputting the test set into each optimal model for evaluation, and comparing evaluation results to obtain a target model. The method provides a scientific basis for early intervention of ASD.

Description

Technical Field

[0001] The present invention relates to the fields of neuroscience and artificial intelligence technology, and in particular to a method for predicting autism spectrum disorder (ASD) based on electroencephalogram (EEG) data and machine learning algorithms. The method aims to identify neural activity patterns associated with ASD through non-invasive EEG signals, providing a scientific basis for early intervention. Background Art

[0002] Autism spectrum disorder (ASD) is a neurodevelopmental disorder characterized by difficulties with social interaction, language communication, and repetitive behaviors. Although the pathogenesis of ASD is not fully understood, early identification and intervention have been shown to significantly improve patients' quality of life. Currently, ASD screening relies primarily on behavioral observations and standardized questionnaires, lacking objective biomarker support. Furthermore, the screening process often relies on expert experience, making accurate early screening difficult.

[0003] Electroencephalography (EEG) is a noninvasive method for recording brain electrical activity. It can monitor changes in EEG activity in real time and has been shown to have potential application value in ASD research. Particularly during sleep, significant differences in EEG activity are observed between children with ASD and healthy controls. Sleep spindles, as an important marker of sleep, can reflect the brain's neural function. Combining machine learning techniques to analyze EEG signals, particularly the characteristics of sleep spindles, can provide a new approach for early screening of ASD.

[0004] In view of this, the present invention is proposed. Summary of the Invention

[0005] The purpose of the present invention is to overcome the shortcomings of the above-mentioned prior art and propose a prediction method for autism spectrum disorder based on EEG machine learning. This prediction method analyzes the sleep spindle characteristics in the EEG data (hereinafter also referred to as EEG signals) of ASD and healthy control children (TD), combines machine learning algorithms for classification prediction, and aims to achieve early screening of ASD.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] The present invention provides a method for predicting autism spectrum disorder based on EEG machine learning, comprising the following steps:

[0008] Step 1: Data collection: EEG data were collected from children with autism spectrum disorder and healthy controls.

[0009] Step 2, feature extraction: preprocessing the EEG data collected in step 1, then extracting the sleep spindle signal in the EEG data, and analyzing the sleep spindle signal to obtain feature information data;

[0010] Step 3: Data processing: Standardize the feature information data obtained in step 2 to obtain a data set, and then divide the data set into a training set and a test set;

[0011] Step 4: Model training: Train the training set using different machine learning algorithms to obtain the corresponding training model, and select the optimal model from each training model using a 5-fold cross-validation strategy;

[0012] Step 5: Model testing and evaluation: Input the test set into each optimal model for evaluation, and then obtain the target model by comparing the evaluation results.

[0013] Furthermore, in step 1, EEG data is collected during the time period of 18:00 to 06:00.

[0014] Furthermore, in step 1, the IQ of the children with autism spectrum disorder and the healthy control children is greater than 70.

[0015] Furthermore, in step 2, EDFbrowser is used to filter and remove artifacts from the EEG data.

[0016] Furthermore, in step 2, the characteristic information data includes the density, duration, amplitude, frequency and Chirp index data of fast, slow and average spindle waves of the whole brain.

[0017] Furthermore, in step 3, the data ratio of the test set and the training set is 70% to 80%: 20% to 30%; wherein, the ratio of children with autism spectrum disorders and healthy control children in both the training set and the test set is 1:1.

[0018] Furthermore, in step 4, the machine learning algorithm includes support vector machine (SVM), k-nearest neighbor (KNN) and logistic regression (LR).

[0019] Compared with the prior art, the present invention has the following beneficial effects:

[0020] The prediction method provided by the present invention provides an effective prediction method by analyzing the sleep spindle signal characteristics of ASD children and TD children and combining it with machine learning technology, which can provide strong technical support for the early screening of ASD. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The accompanying drawings are incorporated in and constitute a part of this specification and, together with the description, serve to explain the principles of the invention.

[0022] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0023] Figure 1 Schematic diagram of EEG data collection for children with autism spectrum disorders and healthy control children according to the present invention;

[0024] Figure 2 This is a flowchart of the sleep spindle signal extraction and feature selection process of the present invention;

[0025] Figure 3 This is a flow chart of the training and evaluation of the machine learning model of the present invention;

[0026] Figure 4 This is a performance evaluation diagram of different algorithm models for high-function ASD in the embodiment;

[0027] Figure 5 Graph showing the performance evaluation of different algorithm models for low-function ASD in the embodiments. DETAILED DESCRIPTION

[0028] Here, exemplary embodiments will be described in detail, and the embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are only examples consistent with some aspects of the present invention described in detail in the appended claims.

[0029] See also Figures 1 to 3 The present invention provides a method for predicting autism spectrum disorder based on EEG machine learning, comprising the following steps:

[0030] Step 1: Data Collection

[0031] Fourteen high-functioning children with ASD and 13 healthy controls (IQ > 70) were recruited. All subjects were clinically assessed with a Wechsler scale to ensure they met applicable criteria. For those who met the criteria, sleep data was recorded for one night (6:00 PM to 6:00 AM) using a 32-lead EEG device with a sampling rate of 500 Hz and a duration of 12 hours. (Subjects were kept in a natural sleep state during the data collection period, without disrupting their normal sleep schedule.)

[0032] Specifically, the inclusion criteria for ASD are as follows:

[0033] (1) Ages 6 to 12, male or female;

[0034] (2) A diagnosis of ASD was confirmed based on the DSM-V diagnostic criteria and clinical manifestations;

[0035] (3) Agree to participate in this study and sign a written informed consent form.

[0036] ASD exclusion criteria: Subjects who meet any of the following criteria will not be included:

[0037] (1) Patients with other mental and neurological diseases (such as tic disorder, anxiety disorder, major depressive disorder, conduct disorder, epilepsy or other mental and neurological diseases);

[0038] (2) IQ score below 70;

[0039] (3) Currently taking medication for mental illness.

[0040] Diagnostic criteria: The diagnosis is made by clinical physicians based on the DSM-V diagnostic criteria and the child's manifestations.

[0041] Step 2: Feature extraction

[0042] The EEG data collected from the subjects were filtered using EDFbrowser (high-pass filter to remove low-frequency noise, low-pass filter to remove high-frequency noise) and artifacts were removed to ensure signal quality. Subsequently, the open source software Luna (website: https: / / zzz.bwh.harvard.edu / luna / ) was used to extract NREM stage signals and identify the time window of sleep spindle signals from the high-quality signals.

[0043] Step 3: Data processing

[0044] Luna was used to extract the density, duration, amplitude, frequency, and chirp index data of fast, slow, and average spindles in the whole brain from each subject's sleep spindle signal. All data were normalized (z-score normalization was optional) to obtain a dataset, which was then randomly divided into a 70% training set and a 30% test set.

[0045] Among them, the ratio of children with autism spectrum disorder and healthy control children in both the training set and the test set was 1:1.

[0046] Step 4: Model training

[0047] The SVM, KNN, and LR algorithms were used to train the training set to obtain three types of training models. The corresponding optimal model (the model with the highest prediction accuracy) was selected from the three types of training models through a 5-fold cross-validation strategy, and the model with the highest prediction accuracy was selected for evaluation. Specifically, the prediction accuracy, AUC, recall rate, and F-score of different optimal models were tested on the test set, and the advantages and disadvantages of each optimal model were compared to obtain the target model.

[0048] In order to illustrate the effect of the present invention, the present invention has done the following test:

[0049] The characteristic information data of 14 high-functioning ASD children (IQ>70) and 13 healthy control children were input into the optimal model obtained in step 4 to test the prediction performance of different algorithms in high-functioning ASD children.

[0050] The prediction performance of the model was verified, and the results were as follows Figure 4 As shown by Figure 4 It can be seen that the SVM model shows higher predictive efficacy for high-function ASD than the LR and KNN models, with significantly higher area under the curve (AUC), prediction accuracy, recall rate, and F1-score.

[0051] The data of 6 low-functioning ASD children (IQ < 70) and 14 healthy control children were input into the optimal model obtained in step 4 to test the predictive performance of different algorithms in low-functioning ASD children and evaluate the universality of the model.

[0052] The prediction performance of the model was verified, and the results were as follows Figure 5 As shown by Figure 5 It can be seen that the SVM model and the LR model showed similar effectiveness in predicting low-functioning ASD children.

[0053] Combining the performance of high-functioning and low-functioning ASD children, the SVM model can be used as a solution for clinical prediction of ASD children.

[0054] in, Figure 4 The original data are shown in Table 1. Figure 5 The original data are shown in Table 2.

[0055] Table 1 Characteristic information data of high-functioning ASD children (IQ>70) and TD children

[0056]

[0057] Table 2 Characteristic information data of low-functioning ASD children (IQ < 70) and TD children

[0058]

[0059]

[0060] The foregoing description is intended only to provide specific embodiments of the present invention, which will enable those skilled in the art to understand and implement the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention.

[0061] It should be understood that the present invention is not limited to the above description and that various modifications and changes may be made without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. A method for predicting autism spectrum disorder based on EEG machine learning, characterized in that: The following steps are involved: Step 1: Data collection: EEG data were collected from children with autism spectrum disorder and healthy controls. Step 2, feature extraction: preprocessing the EEG data collected in step 1, then extracting the sleep spindle signal in the EEG data, and analyzing the sleep spindle signal to obtain feature information data; Step 3: Data processing: Standardize the feature information data obtained in step 2 to obtain a data set, and then divide the data set into a training set and a test set; Step 4: Model training: Train the training set using different machine learning algorithms to obtain the corresponding training model, and select the optimal model from each training model using a 5-fold cross-validation strategy; Step 5: Model testing and evaluation: Input the test set into each optimal model for evaluation, and then obtain the target model by comparing the evaluation results.

2. The prediction method according to claim 1, characterized in that In step 1, EEG data are collected during the time period from 18:00 to 06:

00.

3. The prediction method according to claim 1, wherein: In step 1, the IQ of the children with autism spectrum disorder and the healthy control children is greater than 70.

4. The prediction method according to claim 1, wherein: In step 2, EDFbrowser is used to filter and remove artifacts from the EEG data.

5. The prediction method according to claim 1, wherein: In step 2, the characteristic information data includes the density, duration, amplitude, frequency and Chirp index data of fast, slow and average spindle waves of the whole brain.

6. The prediction method according to claim 1, characterized in that In step 3, the data ratio of the test set and the training set is 70% to 80%: 20% to 30%; wherein, the ratio of children with autism spectrum disorders and healthy control children in both the training set and the test set is 1:

1.

7. The prediction method according to claim 1, wherein: In step 4, the machine learning algorithms include support vector machine, k-nearest neighbor and logistic regression.

Images