Hemispheric Coherence Analysis of Wide Band EEG Signals for Characterization of Post-Stroke Patients with Dementia

Stroke is one of the non-contagious diseases that causes the highest death in the world [1]. In 2015 it was reported that more than 15 million of the population has died from stroke and others have permanent physical disabilities [2]. Another effect of stroke that is often ignored is cognitive impairment that may develop into dementia [3]. This condition will worsen the physiology of the patient so that he may lose his enthusiasm for life. Intensive care is required to delay this deterioration process. Accurate diagnosis is one of the important things to determine and design the right therapy [4].

Clinical examination through memory testing is the main standard in determining a person's cognitive status. The most popular and widely used memory tests include the Mini-Mental State Examination (MMSE) or Montreal Cognitive Assessment (MoCA) test [5]. Then a supporting examination is also needed in relation to the analysis of structural changes or mechanisms in the brain. One of the tools for dementia analysis that are cheap and efficient in setting up is the electroencephalogram (EEG) [6, 7]. Furthermore, EEG offers safe (low-risk), real-time and has a higher temporal resolution than other techniques such as MRI, CT, or PET imaging [8]. Since the stroke causes the death of a number of cells in the brain, the EEG is thought to be able to represent the brain condition of post-stroke patients with dementia. The electrical activity of the brain tissue will be disrupted in this condition. Thus, it is hoped that EEG can be a supporting tool in diagnosing post-stroke dementia.

EEG has been used for several decades to diagnose and management brain disorders [9]. A closely related disease is epilepsy where there is a direct disruption of electrical activity and an EEG can clearly illustrate how this disorder occurs [10]. Then in the last two decades, EEG has potential to observe brain cortical abnormalities associated with cognitive impairment [11, 12]. Furthermore, since computer-based digital signal processing has been widely developed, EEG signal analysis has become more fruitful. Recently this is well known as the Quantitative EEG (QEEG) [13, 14]. Quantitative EEG can provide information numerically related to the characteristics of the EEG waves if there is an abnormality in the brain. Several studies have focused on using QEEG to study brain dynamics in subjects with dementia.

3.1 EEG recording

This study involved post-stroke patients and normal subjects recruited by neurologist, Dr. Hasan Sadikin General Hospital Bandung, Indonesia. All participants agreed to be included in this study by confirming informed consent. This study has also obtained ethical approval No. LB.02.01/X.6.5/272/2019 from Hasan Sadikin Hospital's ethical committee. EEG recording was done at the diagnostic center of Dr. Hasan Sadikin General Hospital. In this study, sixteen sets of EEG signal recordings consisting of ten normal control subjects and five post-stroke patients (6 - 36 months follow up) with dementia were observed. Validation of stroke patients based on medical records and CT scan imaging. Confirmation of dementia and non-dementia through the MoCA test. Normal subject must conform the criteria include: healthy, never experienced brain injury, not indicated both of neurological and physiological disorders. The MoCA score shows > 25 for normal and < 19 for demented subjects. The demographics of the normal and patient groups are presented in Table 1.

Table 1. The demographics of the normal and patient groups

1.png

Figure 1. Scalp electrode placement in this study [12]

EEG signals are recorded using the Cadwell EasyIII amplification system with electrode placement according to 10-20 systems (Figure 1). Recordings were performed on nineteen unipolar EEG channels representing brain regions including pre-frontal, frontal, central, temporal and occipital. They are Fp1, Fp2, F7, F3, Fz, F4, F8, T3, C3, Cz, C4, T4, T5, P3, Pz, P4, T6, O1, and O2. A notch filter with a frequency of 50 Hz is applied to reject a large amount of power source noise. The raw EEG signal data is then stored at a sampling frequency of 250 Hz. Data is labeled in the European Data Format (EDF).

3.2 Coherence measurement

Coherence is a measure of the dynamic activity and spatial information of the brain in clinical EEG applications. Coherence provides an estimate of functional interactions or cortical connectivity between the nervous systems in all regions of the brain. It shows the spectral correlation between two different EEG electrodes. Coherence is a measure of synchronization between two signals based primarily on phase consistency. High coherence value occurs when the phase difference between channels tends to be constant. Therefore, it is considered an ideal measure related to brain disorders. Coherence analysis has also recently been used to study dementia associated with vascular disease specifically diabetes. Research conducted by Bian et al., reported that cognitive decline in diabetic patients affects EEG signals including decreased alpha and theta coherence in several brain areas compared to normal subjects [31]. The study that has been carried out provides the opportunity for coherence observation in post-stroke dementia cases which is proposed in this study. Coherence can be expressed by dividing the squared magnitude of the cross-spectral density of two channels by the product of the power spectral density of two channels. Where the power spectral density is the power distribution based on the frequency component of the signal. The power spectral density was calculated on the autocorrelation of the signal using a fast Fourier transform. The coherence (Cxy) of the x and y signals is calculated using power spectral density (Pxx and Pyy) and cross-spectral density (Pxy), Eq. (1) shows the calculation of coherence [32].

 $C_{x y}(f)=\frac{\left|P_{x y}(f)\right|^{2}}{P_{x x}(f) * P_{v v}(f)}$              (1)

where, f is frequency. Eq. (1) will generate a coherence value with a range of 0 to 1, which has high coherence if the value is close to 1. In this study, coherence was measured based on two electrodes at different locations. Observations were carried out on interhemispheric (left and right hemisphere) and intrahemispheric (same region) connections. The mid channel including Fz, Cz and Pz is not calculated. Table 2 shows the electrode pairs for both scenarios. In this study, coherence is calculated in the conventional EEG band with a wide range of 1 - 30 Hz, as a representation of the delta, theta, alpha and beta bands.

3.3 Statistical analysis

In measuring the significant differences between groups which were observed, we used an independent t-test. We set the confidence level at 95% where the coherence between groups is considered significant difference if the p-value <0.05. All t-test calculations were performed on Ms. Excel. By testing the significance of differences, it is expected to know how coherence can be a discrimination feature.

Since EEG contains a large amount of noise, signal filtering is a typical one that must be performed. Muscle artifact with a frequency of more than 30Hz often interfere with EEG signals. Therefore, the raw EEG signal is filtered with frequency response of 1-30Hz. Figure 3 shows a sample denoising the EEG signal for channel "T3". It can be seen that the components of the muscle noise with dominant activity at high frequency are successfully suppressed. Thus the bias of the coherence measurement can be suppressed typically in the measurement of the temporal area where the power of the muscle noise can disrupt the EEG signal.

After the denoise was performed using a conventional filter, the coherence for the electrode pairs is calculated using Equation 1. Comprehensive investigations were carried out including Interhemispheric and Intrahemispheric coherence. Figures 4, 5 and 6 show the mean and standard deviation of coherence between post-stroke dementia and normal control subjects. Indications of the significant difference are also presented. In the anterior-posterior intrahemispheric mean coherence, it appears that there is a decrease in the coherence value in both of the groups observed. This is consistent with studies by Jeong et al. [15]. where the local coherence value is higher than the distal coherence value.

7.png

Figure 7. The electrode pairs in right-left interhemispheric coherence with p <0.05

Interhemispheric observations showed that the mean coherence value in dementia patients was decreased compared to normal control subjects Coh_dem < Coh_control) for all electrode pairs. Significantly found in C3-C4, T3-T4, and T5-T6 pairs (p<0.05) as shown in Figure 7. It represents the central and temporal lobe regions. Our findings confirm the study by Al-Qazzaz et al. [28, 29], which investigated the signal complexity in stroke patients related cognitive impairment, showing that there was a significant reduction in signal in the central and temporal regions. We presumed that stroke-related dementia patients have damaged a number of neurons and synapses in these regions. In contrast to Alzheimer's, a coherence right-left interhemispheric significant reduction is common in the frontal lobe region [15].

In observing intrahemispheric coherence, we also found a reduction in coherence in patients compared to normal control subjects. There were significant differences found in the electrode pairs of local and distal intrahemispheric coherence, includes frontal, central and temporal. These results represent the consistency with the measurement of interhemispheric coherence where the central and temporal regions have a coherence reduction resulted from a failure of functional connectivity. Thus it can be thought that a decrease in coherence values between brain regions correlates strongly with cognitive impairment as reported in the study [33-35].

The coherence investigation in this study showed a pattern of decreased coherence values in post-stroke dementia patients compared with normal elderly subjects. The most likely reason that could be accepted for the lower coherence is the death of a large number of neurons and degeneration of synapses leading to decreased function of cortical connectivity [15, 36]. This can be interpreted as a disconnection syndrome of functional interactions between regions in the brain [25]. In this study we also found that cognitive impairment in post-stroke dementia patients might be associated with decreased coherence in the temporal lobe, which plays an important role in hearing, language and memory [37-39]. The low reduction or insignificant difference in other regions may be because the post-stroke dementia patients involved in this study still had normal vision and were not aphasia, represented in the frontal and occipital regions, respectively.

Although this study provides an important finding, there are limitations that can be of concern. First, the number of subjects involved in the study was small, particularly post-stroke patients. The second limitation is that the analysis was not performed in post-stroke patients with no dementia. This may be very important as a baseline because it is a relevant comparison rather than normal elderly subjects. At least the preliminary findings of this study suggest that coherence analysis of EEG signals can be a method of investigating and distinguishing dementia in post-stroke patients. It is also hoped that it can be used to investigate the severity of post-stroke dementia so that it can be used for early detection. Future studies need to be extended to larger populations in order to comply statistical criteria that can be considered as supporting clinical diagnostics.