Multi-dimensional conditional mutual information with application on the EEG signal analysis for spatial cognitive ability evaluation
Introduction
Spatial cognition refers to the information processing process of the dimension, shape, direction, and distance of people in the physical space or mental space of three-dimensional objects (Xu, Shushan, et al., 2018). Cognitive impairment, such as degeneration of visual–spatial structure and spatiotemporal disorientation (Johnson, 2010), is one of the most difficult problems in the diagnosis and treatment of brain diseases (Hornero et al., 2009, Whitwell et al., 2007). In the measurement and evaluation of spatial cognitive ability (Morris, 1984), the results of the common methods focus more on subjective judgments, and subjects are easily affected by environmental factors, resulting in errors and a lack of validity and reliability of the measurement results. Therefore, it is of great importance to explore spatial cognitive training and assessment methods for the prevention of spatial cognitive decline in healthy subjects. It is a problem to be solved in the field of brain science and cognitive science to analyze and study spatial cognitive ability and find a quick and effective method to evaluate spatial cognitive ability.
Coupling strength calculation is the mainstream method to quantify the change of spatial cognitive activity in EEG signal analysis. At present, only the method based on causality (Granger causality analysis (GCA) (Protopapa, Mylonas, Spiliotis, Siettos, Smyrnis, & Evdokimidis, 2011) and Sugihara causality) and mutual information (Li and Ouyang, 2010, Toppi et al., 2017, Wen et al., 2016, Wen et al., 2019) considers the coupling direction between EEG signals of different electrodes. Granger causality analysis has quantitatively analyzed the degree of linear dependence between different electrode signals (Bischof and Boulanger, 2003, Mcbride et al., 2015, Tarnanas et al., 2014). However, it has a high requirement for the selection of time window width. Sugihara causality describes the deterministic causality that is not easily detected but exists in the system (Tarnanas, Laskaris, Tsolaki, et al., 2015). CMI method considers the coupling strength relationship and coupling direction. Many studies have shown that CMI is superior to Granger causality when considering the coupling direction of neurons (Bashivan et al., 2016, Jaiswal et al., 2010, Kober and Neuper, 2011, Olton et al., 1979). However, CMI is an information theory method of binary time series, which only considers the one-way coupling between two neurons, but does not consider the influence of other related channels on the coupling relationship between two channels (Wen et al., 2016, Wen et al., 2019).
The above methods identify EEG signals by analyzing the characteristics of time domain, frequency domain, coupling strength, coherence strength, and coupling direction of different electrodes, but the spatial characteristics of EEG signals have not been considered. Therefore, Bashivan proposed to transform EEG signals into multispectral images with spatial attributes, which retains the spatial, temporal, and frequency attributes of EEG signals. Considering the influence of all other related channels on the coupling relationship between two channels, a feature extraction method of EEG signals based on multi-dimensional conditional mutual information (MCMI) was proposed to extract the coupled features of EEG signals before and after spatial cognitive training. Considering the advantages of multi-spectral images in feature extraction of EEG signals (Baskaran et al., 2013, Chen et al., 2018, Ieracitano et al., 2019), a convolutional neural network (CNN) model is used for classification after the coupled characteristics of the multi-frequency combination were transformed into multi-spectral images. To explore the performance of MCMI, this study not only compares the performance of MCMI with GCA and CMI but also compares MCMI with the existing coupling methods such as PCMI (Li & Ouyang, 2010) and MPCMI (Wen et al., 2016, Wen et al., 2019) on six band combinations: Theta–Alpha2–Gamma, Alpha2–Beta2–Gamma, Beta1–Beta2–Gamma, Theta–Beta2–Gamma, Theta–Alpha1–Gamma, and Alpha1–Alpha2–Gamma.
Section snippets
The data set
In this paper, we studied the 16 channel EEG signals from 7 subjects without any spiritual system or nerve disease male volunteers, normal eyesight or corrected visual acuity, no color blindness, normal right-handed, and did not take part in similar experiments training, the age distribution in 21–26 before the experiment signed informed consent (Wen, Yuan, Zhou, et al., 2020). The data set included two kinds of EEG signals from 7 subjects participating in the game of Virtual City Walking
Theta-Alpha2-Gamma band combination
The average verification loss of the multi-spectral image data built based on GCA, CMI, PCMI, MCMI, and MPCMI in the CNN model under the combined condition of The Frequency band of Theta–Alpha2–Gamma is shown in Fig. 3. According to Fig. 3, the following conclusions can be drawn: the curve representing the CNN model based on MCMI is the lowest, reaches the lowest point, and is stable after 190 195 iterations, and the average validation loss is stable at 0.09 0.10. The curve of THE CNN model
The value of the MCMI approach
In recent years, the analysis of coupling strength has provided a deeper insight into the mechanism of interaction between different brain regions and has become the focus of EEG signal research. MPCMI method is proposed in cognitive impairment diagnosis (Wen et al., 2016, Wen et al., 2019), and this method can effectively calculate kinetic characteristics of the coupling between different brain areas. It takes into account the effects of all other related channels on the coupling relationship
Conclusion
The MCMI-based EEG signal analysis method proposed in this study can better represent the coupling dynamic relationship between channels, and combining this method with the CNN model can more effectively analyze the spatial characteristics of the coupling relationship. The conclusion shows that the classification model constructed by the MCMI method combined with CNN under the frequency bands of Alpha2–Beta2–Gamma, Theta- Beta2-Gamma, and Beta1–Beta2–Gamma can most effectively evaluate the
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
This research was funded by National Natural Science Foundation of China (61876165, 61503326), Natural Science Foundation of Hebei Province in China (F2016203343), China Postdoctoral Science Foundation (2015M581317). The authors have no any potential conflicts.
References (40)
- et al.
Brain neural synchronization and functional coupling inAlzheimer’s disease as revealed by resting state EEG rhythms
International Journal of Psychophysiology OfficialJournal of the International Organization of Psychophysiology
(2016) - et al.
The MVGC multivariate granger causality toolbox: A new approach togranger-causal inference
Journal of Neuroence Methods
(2014) - et al.
A convolutional neural network approach for classificationof dementia stages based on 2D-spectral representation of EEG recordings
Neurocomputing
(2019) - et al.
Encoding of visual-spatial information in working memory requiresmore cerebral efforts than retrieval: Evidence from an EEG and virtual reality study
Brain Research
(2010) EEG-alpha rhythms and memory processes
International Journal of Psychophysiology
(1997)- et al.
Sex differences in human EEG theta oscillations during spatial navigation invirtual reality
International Journal of Psychophysiology
(2011) - et al.
Estimating coupling direction between neuronal populations with permutationconditional mutual information
Neuroimage
(2010) - et al.
Sugihara causality analysis of scalp EEG for detection of earlyAlzheimer’s disease
NeuroImage: Clinical
(2015) Developments of a water-maze procedure for studying spatial learning in the rat
Journal of Neuroscience Methods
(1984)- et al.
Synaptic devices based neuromorphic computing applications in artificialintelligence
Materials Today Physics
(2021)
Synaptic devices based neuromorphic computing applications in artificialintelligence
Materials Today Physics
Can a novel computerized cognitive screening test provideadditional information for early detection of Alzheimer’s disease?
Alzheimers & Dementia
Resting-state EEG coupling analysis of amnestic mild cognitive impairmentwith type 2 diabetes mellitus by using permutation conditional mutual information
Clinical Neurophysiology
Estimating coupling strength between multivariate neural series withmultivariate permutation conditional mutual information
Neural Networks
Investigation of emotional changes using features of EEG-Gamma band and different classifiers
A review of electroencephalographic changes in diabetes mellitus inrelation to major depressive disorder
Neuropsychiatric Disease and Treatment
Spatial navigation in virtual reality environments: An EEG analysis
Cyberpsychology, Behavior, and Social Networking
Hexadirectional modulation of theta power in human entorhinal cortexduring spatial navigation
Current Biology
Cited by (10)
Dementia classification using a graph neural network on imaging of effective brain connectivity
2024, Computers in Biology and MedicineA bidirectional interaction-based hybrid network architecture for EEG cognitive recognition
2023, Computer Methods and Programs in BiomedicineScalp EEG functional connection and brain network in infants with West syndrome
2022, Neural NetworksCitation Excerpt :Similarly, McCrimmon et al. (2021) used the scalp high-frequency oscillations to accurately identify West subjects from healthy. Although, the characteristics of EEGs are very useful for clinical diagnosis of West as biomarkers (Cao, Zhu, Hu, & Kummert, 2020; Hu, Cao, Lai, Liu et al., 2020; Hussein, Lee, Ward, & McKeown, 2021; Raghu, Sriraam, Temel, Rao, & Kubben, 2020; Wen et al., 2021; Xu et al., 2021), these objective patterns are usually not conducive to understand the underlying neural mechanism of West. Graph theory has been recently applied in the brain imaging, electrophysiology and epilepsy analysis for its ability in quantifying the characteristics of complex system topologies (Bernhardt, Bonilha, & Gross, 2015; Stacey et al., 2020).
Real-time vibration monitoring and analysis of agricultural tractor drivers using an IoT-based system
2023, Journal of Field Robotics