Attention detection using EEG signals and machine learning: A review
image: An overview of attention detection using EEG signals, which includes six steps: an experimental paradigm design, in which the task and the stimuli are defined and presented to the subjects; EEG data acquisition, where EEG data are recorded from the subjects using electrodes; preprocessing, in which artifacts such as eye movements and muscle activity are removed and the EEG signals are filtered and segmented; feature extraction, in which relevant features such as power spectrum, entropy, and coherence are extracted from the EEG signals; classification, in which machine learning algorithms are applied to classify the attention levels based on the features; feedback, in which the classification results are presented to the users.
Credit: Beijing Zhongke Journal Publising Co. Ltd.
Attention refers to the capacity to select, modulate, and sustain a focus on the most relevant information for behavior, and it plays an essential role in all perceptual and cognitive operations. Attention helps individuals allocate cognitive resources to specific tasks and ignore background noise. Furthermore, attention serves as an important reference for the investigation of neuroscience and many other disciplines. Therefore, attention has become a popular research topic in both academia and industry.
Despite its importance, attention has always been an ambiguous concept, making it hard to give a precise definition. According to Knudsen’s attentional model, “attention is considered as the result of three coordinated processes: (A) monitoring and evaluation of ongoing cognitive processes, (B) excitation of task-relevant processes, and (C) inhibition of task-irrelevant processes”. Attention can be divided into several submodalities, including selective attention, spatial attention, visual attention, sustained attention, and vigilance.
Due to the importance of attention in daily life, the analysis of attention has many practical applications. In response to this need, researchers have developed several techniques to measure attention state, including eye blinking, functional magnetic resonance imaging (fMRI), muscle movement, and electroencephalogram (EEG). Among these techniques, EEG is a noninvasive, effective, and convenient method for analyzing attention. EEG records the electrical signals produced by the brain, reflecting changes in the brain activity over time. By analyzing EEG signals, researchers can discriminate between attentive states, monitor variations in visual sustained attention, and decode auditory and spatial attention. Thus, applying EEG signals in attention detection and analysis has a lot of practical significance in attention studies.
There is a long history of research on attention using EEG signals. In the 1980s, several studies found evidence that the gamma band may be related to the attention or intention of movement. Since then, researchers have discovered associations between attention and EEG signals, demonstrating that visual attention, spatial attention, auditory attention, and attention level can all be measured through the analysis of EEG signals. Many researchers have applied this technique to practical applications. For example, estimating the vigilance and attention level of drivers to ensure transportation safety; using EEG sensors to detect auditory attention and applying this information to neuro-steered hearing devices; using overt and covert attention to provide reliable control of the brain-computer interface (BCI) systems; finding that the detection of attentive states can aid in the diagnosis of common diseases such as conduct disorder or attention deficit hyperactivity disorder (ADHD). There is also a trend toward portable and mobile EEG devices. For instance, some researchers demonstrated that it is possible to interpret a subject’s auditory attention by analyzing EEG data from a single trial. Some researchers designed an attention state classification system using an in-ear EEG device.
In recent years, many authors have conducted reviews of EEG analysis from various perspectives, covering areas such as emotion recognition, disease diagnosis, and motor imagery. However, relatively few surveys on the use of EEG signals in attention detection have been conducted. Some surveys have focused on specific aspects of attention, such as auditory attention and its future applications, review of auditory attention detection methods. Other studies have mentioned attention detection under specific conditions, for example, some studies surveyed attention detection in a virtual environment, some reviewed attention detection methods in BCI games. However, there does not exist a comprehensive and systematic review that covers in-depth the application of machine learning algorithms for EEG-based attention detection.
To fill this gap, this survey published in Machine Intelligence Research offers an up-to-date, comprehensive, and insightful overview of attention detection using EEG signals. Rather than focusing on specific aspects of attention, the team of researchers from Nanjing University of Aeronautics and Astronautics provide a broad overview of this field. To achieve this goal, they summarize the methods used at different stages of attention detection, discuss the existing problems and future trends in this field, and provide a comprehensive summary of research suggestions for future researchers. There is an overview of attention detection using EEG signals, which includes six steps: an experimental paradigm design, in which the task and the stimuli are defined and presented to the subjects; EEG data acquisition, where EEG data are recorded from the subjects using electrodes; preprocessing, in which artifacts such as eye movements and muscle activity are removed and the EEG signals are filtered and segmented; feature extraction, in which relevant features such as power spectrum, entropy, and coherence are extracted from the EEG signals; classification, in which machine learning algorithms are applied to classify the attention levels based on the features; feedback, in which the classification results are presented to the users.
Section 2 introduces the fundamental background information about attention categories, including paper selection strategy and attention category.
Section 3 begins by reviewing the common datasets and paradigms used in this field, as well as the EEG devices and electrode configurations employed for recording. Then, researchers describe the preprocessing, feature extraction, and classification methods applied to the EEG data. Researchers keep a primary focus on the utilization of machine learning methods in EEG-based attention detection.
Section 4 offers best practice recommendations, discusses current problems in this area, and suggests possible directions for future research. First, researchers present some practical recommendations based on previous research results from four aspects: Data acquisition, artifacts handling, feature extraction and classification. Second, researchers summarize some problems: Need for high-quality data, lack of standardization and generalization, low spatial resolution and limited electrode locations in portable devices, need for advanced analysis methods, and need for real-world scenarios. Third, researchers summarize the views of these articles and propose the trends which they believe are valuable: More public dataset, traditional machine learning methods, deep learning methods, expanding sensor selection integrated with cognitive theories, real-world scenarios, development of neuro-steered applications, personalized and adaptive EEG-based attention detection, and multimodal data.
In this article, researchers systematically and comprehensively review the research on attention in the field of EEG over the past 5 years. They mainly focus on two issues: AAD and attention level classification. They summarize the mainstream EEG data preprocessing and feature extraction methods and provide a detailed analysis of machine learning methods found in the literature. They also consider the factors that influence the effectiveness and suitability of these models. Furthermore, they offer some suggestions for data acquisition, feature extraction, and classification methods. This review also reveals some existing problems and challenges in this field, such as the lack of standardized datasets, the trade-off between accuracy and efficiency, the ethical and privacy issues of EEG data collection and analysis, etc. Therefore, researchers suggest some possible future research directions such as using multimodal data to improve the accuracy and robustness of attention detection. They hope that their research results can provide an objective and useful practical guide for future researchers.
See the article:
Attention Detection Using EEG Signals and Machine Learning: A Review
http://doi.org/10.1007/s11633-024-1492-6