Diagnosis of autism spectrum disorder using EEMD and multiscale fluctuation based dispersion entropy with Bayesian optimized light GBM

Introduction

 

Autism spectrum disorder (ASD) is a neurological condition that affects the social and communication skills of millions of people around the world. It is usually diagnosed by observing the behavior and development of children, but this can be subjective and time-consuming. Is there a better way to detect autism using objective and reliable measures?

 

A recent study published in Multimedia Tools and Applications suggests that there is. The researchers proposed a novel method to diagnose ASD using electroencephalogram (EEG) signals, which are recordings of the brain’s electrical activity. EEG signals can reveal the differences in the brain’s function and structure between people with and without ASD.

 

The proposed method

 

The proposed method consists of four steps:

  1. Signal decomposition: The EEG signals are decomposed into several components using a technique called ensemble empirical mode decomposition (EEMD). This technique can adapt to the non-stationary and nonlinear nature of the EEG signals and extract the intrinsic features of the brain dynamics.
  2. Feature extraction: The complexity and randomness of each component are measured using a new entropy measure called multiscale fluctuation based dispersion entropy (MFDE). This measure can capture the fluctuations of the signal at different scales and provide a robust and efficient feature representation.
  3. Feature selection: The most relevant and informative features are selected using a filter-based method called minimum redundancy maximum relevance (mRMR). This method can reduce the dimensionality and redundancy of the features and improve the classification performance.
  4. Classification: The selected features are fed into a machine learning model called light gradient boosting machine (LGBM) to classify the EEG signals into ASD or neurotypical groups. LGBM is a fast and accurate model that can handle large and imbalanced datasets. The parameters of the model are optimized using a technique called Bayesian optimization, which can find the best combination of parameters in a limited number of iterations.

 

The results

 

The researchers tested their method on a dataset of EEG signals from 122 subjects, 61 with ASD and 61 without. They compared their method with several existing methods based on different signal decomposition, feature extraction, and classification techniques. They evaluated the methods using various metrics, such as accuracy, specificity, sensitivity, area under the curve (AUC), and Kappa statistic.

See also  PTEN controls alternative splicing of autism spectrum disorder-associated transcripts in primary neurons

 

The results showed that the proposed method outperformed all the other methods in terms of all the metrics. The proposed method achieved an accuracy of 99.59%, a specificity of 99.37%, a sensitivity of 99.18%, an AUC of 0.9998, and a Kappa statistic of 0.9919. Moreover, the proposed method had the lowest computational time and error measures among all the methods.

 

The implications

 

The study demonstrates that EEG signals can be used as a potential biomarker for diagnosing ASD. The proposed method can provide a fast, accurate, and reliable diagnosis of ASD using a simple and non-invasive technique. The proposed method can also be applied to other neurological disorders, such as epilepsy, Alzheimer’s disease, and schizophrenia.

 

The study also contributes to the field of EEG signal analysis and machine learning by introducing a new entropy measure and a new parameter optimization technique. The new entropy measure can capture the complexity and randomness of the EEG signals at different scales, and the new parameter optimization technique can find the optimal parameters of the machine learning model efficiently.

 

The study opens up new avenues for further research and development in the area of ASD diagnosis and treatment. For example, the researchers suggest that future work can explore the use of other signal decomposition and feature extraction techniques, such as wavelet transform and fractal dimension. They also suggest that future work can investigate the use of other machine learning models, such as deep neural networks and support vector machines. Furthermore, they suggest that future work can validate the proposed method on larger and more diverse datasets, as well as on real-time and online applications.

 

The conclusion

 

In conclusion, the study presents a novel method to diagnose ASD using EEG signals and machine learning. The method can provide a fast, accurate, and reliable diagnosis of ASD using a simple and non-invasive technique. The method can also be applied to other neurological disorders, as well as to other domains of EEG signal analysis and machine learning. The study is a significant step towards leveraging the power of EEG signals and machine learning to improve the lives of people with ASD and their families.

 

FAQ

What is EEG and how does it work?

 

EEG stands for electroencephalogram, which is a technique that records the electrical activity of the brain using electrodes attached to the scalp. EEG signals reflect the synchronous activity of millions of neurons that communicate with each other through electrical impulses. EEG signals can reveal the patterns and rhythms of the brain’s function and structure, as well as the changes in the brain’s activity due to various stimuli, tasks, or states.

See also  The emerging role of extracellular vesicles in the diagnosis and treatment of autism spectrum disorders

 

How can EEG signals be recorded and processed?

 

EEG signals can be recorded using a device called an EEG machine, which consists of a set of electrodes attached to the scalp, a signal amplifier, and a computer. The electrodes detect the electrical activity of the brain, which is amplified and converted into digital signals by the signal amplifier. The computer then processes and displays the EEG signals using various software and algorithms. The EEG signals can be analyzed in terms of their frequency, amplitude, phase, and coherence, as well as their spatial and temporal patterns.

 

What is machine learning and how does it work?

 

Machine learning is a branch of artificial intelligence that enables computers to learn from data and perform tasks without explicit programming. Machine learning algorithms can find patterns, extract features, and make predictions from data using various mathematical and statistical methods. Machine learning algorithms can be divided into two main types: supervised and unsupervised. Supervised learning algorithms learn from labeled data, i.e., data that has a known outcome or target variable. Unsupervised learning algorithms learn from unlabeled data, i.e., data that has no predefined outcome or target variable.

 

What is the difference between entropy and complexity in EEG signals?

 

Entropy and complexity are two measures that quantify the randomness and orderliness of EEG signals, respectively. Entropy measures the uncertainty or unpredictability of EEG signals, i.e., how much information is contained in the signals. Complexity measures the regularity or irregularity of EEG signals, i.e., how much structure or organization is present in the signals. Entropy and complexity are inversely related, i.e., higher entropy means lower complexity and vice versa.

 

What are the advantages of using EEMD and MFDE for EEG signal analysis?

 

EEMD and MFDE are two techniques that can overcome some of the limitations of traditional EEG signal analysis methods. EEMD is a technique that can decompose EEG signals into several components that represent the intrinsic modes of the signals, without introducing any distortion or artifacts. MFDE is a technique that can measure the entropy and complexity of EEG signals at different scales, without being affected by noise or outliers. EEMD and MFDE can capture the nonlinear and non-stationary characteristics of EEG signals, as well as the multiscale fluctuations of the signals.

 

What are the benefits of using LGBM and Bayesian optimization for EEG signal classification?

 

LGBM and Bayesian optimization are two techniques that can improve the accuracy and efficiency of EEG signal classification. LGBM is a machine learning model that can handle large and imbalanced datasets, as well as deal with missing values and categorical features. LGBM can also reduce the computational time and memory usage by using a gradient boosting algorithm that grows trees leaf-wise rather than level-wise. Bayesian optimization is a technique that can find the optimal parameters of the machine learning model, by using a probabilistic model that incorporates prior knowledge and evidence. Bayesian optimization can also reduce the number of iterations and evaluations by using an acquisition function that balances exploration and exploitation.

See also  The relation of internal state language and theory of mind abilities in the narratives of children with autism spectrum disorder and normal developing children

 

How can EEG signals reveal the differences between people with and without ASD?

 

EEG signals can reveal the differences in the brain’s function and structure between people with and without ASD by showing the patterns of connectivity and synchronization among different brain regions. People with ASD tend to have lower connectivity and synchronization in some brain regions, such as the frontal and temporal lobes, which are involved in social and language processing. People with ASD also tend to have higher connectivity and synchronization in other brain regions, such as the occipital and parietal lobes, which are involved in visual and spatial processing.

 

What are the advantages and disadvantages of using machine learning for ASD diagnosis?

 

Machine learning is a powerful technique that can provide fast, accurate, and objective diagnosis of ASD using EEG signals and other data sources. Machine learning can also help to discover new biomarkers and predictors of ASD, as well as to personalize and optimize the diagnosis and treatment of ASD. However, machine learning also has some limitations and challenges, such as the need for large and high-quality datasets, the difficulty of interpreting and explaining the models, and the ethical and social implications of using machine learning for ASD diagnosis.

 

What are the challenges and limitations of using EEG signals and machine learning for ASD diagnosis?

 

Some of the challenges and limitations of using EEG signals and machine learning for ASD diagnosis are:

  • The quality and reliability of the EEG signals depend on various factors, such as the number and placement of the electrodes, the noise and artifacts in the recording environment, and the individual differences and variability among the subjects.
  • The interpretation and generalization of the EEG signals and machine learning models depend on various factors, such as the size and diversity of the dataset, the validity and reliability of the labels, and the robustness and explainability of the models.
  • The ethical and social implications of using EEG signals and machine learning for ASD diagnosis need to be considered, such as the privacy and security of the data, the consent and feedback of the subjects, and the potential bias and stigma of the diagnosis.

 

How can the proposed method be improved or extended in the future?

 

Some of the possible ways to improve or extend the proposed method in the future are:

  • Using more advanced signal decomposition and feature extraction techniques, such as wavelet transform and fractal dimension, to capture the finer details and dynamics of the EEG signals.
  • Using more sophisticated machine learning and deep learning models, such as deep neural networks and support vector machines, to enhance the accuracy and efficiency of the classification.
  • Using more comprehensive and diverse datasets, such as datasets from different age groups, genders, ethnicities, and cultures, to increase the validity and generalizability of the diagnosis.
  • Using more interactive and online applications, such as applications that can provide real-time and personalized feedback and intervention for the subjects.

 

Source:

https://link.springer.com/article/10.1007/s11042-023-18059-x

Leave a Comment