Studies have reported that electroencephalogram signals in Alzheimer’s disease patients usually have less synchronization than those of healthy subjects. Changes in electroencephalogram signals start at early stage but, clinically, these changes are not easily detected. To detect this perturbation, three neural synchrony measurement techniques: phase synchrony, magnitude squared coherence, and cross correlation are applied to three different databases of mild Alzheimer’s disease patients and healthy subjects. We have compared the right and left temporal lobes of the brain with the rest of the brain areas (frontal, central, and occipital) as temporal regions are relatively the first ones to be affected by Alzheimer’s disease. Moreover, electroencephalogram signals are further classified into five different frequency bands (delta, theta, alpha beta, and gamma) because each frequency band has its own physiological significance in terms of signal evaluation. A new approach using principal component analysis before applying neural synchrony measurement techniques has been presented and compared with Average technique. The simulation results indicated that applying principal component analysis before synchrony measurement techniques shows significantly better results as compared to the lateral one. At the end, all the aforementioned techniques are assessed by a statistical test (MannWhitney
Mild cognitive impairment (MCI) is characterized by impaired memory state of brain probably leading towards mild Alzheimer’s disease (MiAD) or Alzheimer’s disease (AD). This prodromal stage of AD has been under a great influence of research for a long time [
Loss of functional connectivity between cortical and hippocampus has long been an important focus of researches to examine the cause of cognitive dysfunction in AD [
Electroencephalogram (EEG) signals are considered functional examples to evaluate cognitive disturbances and a diagnostic tool, especially when a diagnostic doubt exists even after the initial clinical procedures [
Topographically analyzing the EEG signals, Hogan et al. [
The studies, so far, have provided a very limited regional comparison of brain; for instance, less synchronization has been reported between temporal and central regions [
Synchronization, precisely speaking, is a coordination of “rhythmic oscillators” [
Various synchrony measurement techniques have already been discussed to detect any perturbation in the EEG signals of AD patients [
Despite the considerable success of the above mentioned techniques to analyze disruption in the EEG signals of Alzheimer’s patients, further investigations are still required to fulfill the clinical requirements. For instance, in order to detect Alzheimer’s disease at its earlier stages, we need to focus on those areas where Alzheimer’s disease attacks at first and then we need to check its synchronization with the rest of the brain regions. Furthermore, additional novel and comprehensive methods are still required to check the validity of aforementioned techniques on EEG signals.
The above overview suggests that, first, spatialspectral analysis of EEG signals can provide a measure of memory visualization. Second, neural synchrony measurement techniques have a potential to discriminate between AD patients and healthy subjects. What is still missing or ambiguous in the literature survey is the simultaneous comparison of all parts of brain with the right and left temporal lobes (the most affected parts of brain) to analyze synchronization and also the implementation of new methods to apply synchrony measurement techniques. In this research work, the following novel contributions are considered.
We have filtered a dataset of MiAD patients into five different frequency bands (delta, theta, alpha, beta, and gamma). For each frequency band, we have computed neural synchronization to compare all parts of brain (frontal, occipital, and central) with left and right temporal lobes.
Furthermore, three different sets of MiAD patients are compared to check the validity of our methodology. A high intersubject variability has been seen in the EEG signals of AD patients, especially with different level of severity and comorbidities [
In order to remove the ambiguity of biased results due to “features redundancy,” we have applied PCA (principal component analysis) technique before applying synchrony measurement techniques. Reducing features vector dimension, commonly known as feature reduction, will help to get accuracy results and avoid overfitting classification [
Besthorn et al. [
Given the exploratory nature of the study, our priori hypothesis is that the proposed methodology would provide a better insight to investigate the decline in the neural synchronization of AD patients. It would provide a better topographical and spectral analysis of the brain regions eliminating the probability of biased result due to feature redundancy.
The rest of this paper is structured as follows. Section
In this section, we briefly review the synchrony measurement techniques that we have implemented in our datasets which include phase synchrony, cross correlation, and coherence. For this research work, we have selected three synchrony measures from the literature that provides comparatively better results when implemented in EEG signals for the diagnosis of Alzheimer’s disease [
The 21 channels used for EEG recording [
The oscillation of two or more cyclic signals where they tend to keep a repeating sequence of relative phase angles is called phase synchronization. Synchronization of two periodic nonidentical oscillators refers to the adjustment of their rhythmicity, that is, the phase locking between the two signals [
Cross correlation is a mathematical operation used to measure the extent of similarity between two signals. If a signal is correlated to itself, it is called autocorrelated. If we suppose that
The coherence functions estimate the linear correlation of signals in frequency domain [
For discrete signals
The datasets that we are analyzing have been recorded from three different countries of European Union. Specialist at the memory clinic referred all patients to the EEG department of the hospital. All patients passed through a number of recommended tests: minimental state examination (MMSE) [
The EEG
This EEG dataset is composed of 5 MiAD patients (2 males; aged
This dataset consists of 8 MiAD patients (6 males; aged
For current research work, we have obtained a version of the data that is already preprocessed of artifacts by using independent component analysis (ICA), a blind source separation technique (BSS). Details of these procedures can be found in [
EEG time series are classified into five frequency bands. Each frequency band has its own physiological significance [
Delta (
Theta (
Alpha (
Beta (
Gamma (
Bandpass filter is applied to each EEG channel to extract the EEG data in specific frequency band [
In this research work, a novel methodology using PCA and neural synchrony measurement of the brain is proposed. We have compared our proposed method with other methods which takes the average of synchrony measures for all channels in one region of the brain. As mentioned previously, we are comparing the right and left temporal lobes with the frontal, central, and occipital areas; so, there are a total of 7 comparisons of the brain ((left temporalright temporal (LTRT)), (left temporalfrontal (LTF)), (left temporalcentral (LTC)), (left temporaloccipital (LTO)), (right temporalfrontal (RTF)), (right temporalcentral (RTC)), and (right temporaloccipital (RTO))) for all frequency bands (
First, we apply neural synchrony measurement techniques to each channel pair (time series of two channels) of two different regions for all frequency bands and then we take the average of those results. For instance, we apply phase synchrony measure to each channel pair of right and left temporal lobes ((
After getting the results, we compare the neural synchronization of AD patients and healthy subjects, for all three measurement techniques (phase synchronization, cross correlation, and coherence), by MannWhitney
Average and PCA methods.
In this method, instead of applying synchrony measurement techniques directly to the filtered data, first we apply principal component analysis (PCA) technique to all channels of one. This eliminates any redundant information that a region could provide. For instance, we apply PCA to all three channels of left temporal lobe (
The basic purpose of PCA is to reduce the dimensionality of a dataset to convert it to uncorrelated variables providing maximum information about a data while eliminating interrelated variables. In other words, it transforms the highly dimensional dataset (of
In our case, we apply PCA to all channels in one particular region, for instance, the application of PCA for the left temporal lobe as shown in Figure
Application of PCA on left temporal lobe channels signals.
To investigate whether there is a significant difference between the EEG signals of MiAD patients and the control subjects and also to prove the probable significance of our proposed methodology, we apply the Wilcoxon rank sum (MannWhitney) test [
Since we are applying three different synchrony measures to three different sets of data, first we consider our first proposed method (taking average of synchrony values) to compute the synchrony measure. We apply all three measures for all 7 different comparisons of brain for all frequency bands and compute the results by MannWhitney test. Then, we apply the same techniques on all, above mentioned, three datasets using the second proposed method (PCA based synchrony measures). This will enable us to compare our results in two different perspectives as follows.
Investigating three different synchrony measures at a time will help us to compare which measure works better for EEG signals.
Secondly, we are able to compare two different methods for three synchrony measures using three different datasets.
In addition to evaluating the statistical significance of our proposed method, this will also help us to differentiate the MiAD patients from healthy subjects.
The aim of the present study is to find the relationship of EEG synchronization with AD and thus to explore further dimensions in disconnection theorem of cognitive dysfunction in AD and also to investigate a better method to detect any changes in EEG synchrony that can be considered a biomarker for the early detection of AD. Here, we investigate and discuss results in two different angles. First, we discuss the role of synchrony measures to examine a change in EEG synchrony in MiAD patients and later we confer the significance of applying PCA before synchrony measures.
We have observed that all of the synchrony measures, tested in this paper, show a decrease in EEG synchrony for MiAD patients as compared to healthy subjects. However, cross correlation shows a higher number of significant results at the
First, we discuss
Lower
Interestingly, we find a decrease in alpha band synchronization for all three synchrony measures in almost all regions. For instance, for cross correlation,
As aforementioned, mostly the areas that show lower dysfunctional connectivity are right temporalcentral and left temporaloccipital. A lower synchronization in these connections, especially in RTC region, for alpha band indicates a disturbance in the perception and integration of somatosensory information, visuospatial processing, and cognitive disorder. This information is in line with clinical findings presented in [
Synchrony measure  Brain connections  Frequency regions 


Cross correlation  RTC  Delta ( 

Theta ( 


Alpha ( 
0.009  
RTO  Delta ( 


Theta ( 


Alpha ( 
0.0029  
RTF  Delta ( 


Theta ( 


Alpha ( 
0.0062  
LTC  Delta ( 


Theta ( 


Alpha ( 
0.0192  
LTO  Delta ( 


Theta ( 


Alpha ( 
0.0052  
LTF  Delta ( 


Theta ( 


Alpha ( 
0.0091  
LTRT  Delta ( 


Theta ( 


Alpha ( 
0.0253  


Phase synchrony  RTC  Delta ( 
0.0067 
Theta ( 
0.0403  
Alpha ( 
0.05  
RTO  Delta ( 
0.0041  
Alpha ( 
0.0271  


Coherence  RTC  Delta ( 
0.0378 
RTO  Delta ( 
0.0378  
Alpha ( 
0.0192 
Similarly, for
Total number of significant values in case of PCA and Average method.
Synchrony measure  Method 



Cross correlation  PCA  26  35 
Average  22  30  


Phase synchrony  PCA  8  11 
Average  2  8 
Our second hypothesis was to show the significance of using PCA techniques to eliminate the redundant information from the data that can give biased results, before applying synchrony measures. As expected, we found a big difference in results with and without PCA method. We have found that more than 90% of the values are better in case of
For instance, for
The reults are also shown by boxplot in Figure
Boxplots show the results of three synchrony measures for PCA and Average methods.
Similarly, for
As the redundant information is eliminated from the datasets, the results are not biased and are more reliable.
Secondly, it proves that application of PCA generates more significant results as compared to average synchrony measure method.
The aim of the current study was to show the significance of applying PCA method to eliminate redundant information from the datasets to get more reliable results. In this study, three different datasets were selected with different specifications and three different synchrony measures are applied to prove the significance of our approach. Moreover, we have compared our proposed method with Average methods to compute synchronization in MiAD patients as well as in control subjects.
Results revealed that cross correlation measure showed higher difference in synchronization of MiAD and control subjects as compared to phase synchrony, while coherence function did not perform very well. They have also indicated that alpha and theta bands play a major role in identifying the change in synchronization from MiAD and control subjects especially in right temporalcentral region (RTC) and also in left temporaloccipital (LTO) region.
Furthermore, the original contribution of this research work is the comparison of previous methods of applying synchrony measures with PCA based method. Our proposed method proved the importance of eliminating redundant information, from EEG time series, that may come from consecutive electrodes. It should be noted that comparison with previous findings is problematic due to the significant differences in the utilized methodology and the utilization of different kinds of synchrony measures on different kinds of datasets. However, our results are consistent with most of the studies on the loss of average EEG synchrony in different parts of the brain for MiAD patients and are also in accordance with the clinical findings.
Furthermore, we have successfully shown the importance and significance of our proposed method, to detect lower synchronization in MiAD patients, as compared to the Average method for all three datasets.
Future work will involve the study of much significant results of lower synchronization in case of
The authors declare that there is no conflict of interests regarding the publication of this paper.