Sugondo Hadiyoso*
| Inung Wijayanto | Annisa Humairani
© 2023 IIETA. This article is published by IIETA and is licensed under the CC BY 4.0 license (http://creativecommons.org/licenses/by/4.0/).
OPEN ACCESS
The rapid progression of diseases in the elderly, such as Alzheimer's Dementia (AD), necessitates effective early detection mechanisms to ensure appropriate healthcare provision. Given the consistently increasing prevalence of AD, the potential for emerging socio-economic challenges is significant. This underlines the importance of developing early detection strategies to mitigate the progression of this disease. Electroencephalograms (EEG) present a promising avenue for the early diagnosis of AD. EEG signals harbor crucial information pertaining to neuronal death triggered by amyloid plaque accumulation, a characteristic feature of AD. Spectral analysis reveals a deceleration in signal activity in AD patients when compared to healthy elderly individuals. However, this method is frequently compromised by low-frequency noise, necessitating the exploration of alternative approaches for analyzing EEG signal features for early AD detection. Considering the complex nature of EEG signals, it is hypothesized that pathological conditions, such as AD, may induce alterations in signal complexity. In this study, an early detection model for AD was simulated utilizing an approach that focused on EEG signal complexity. Complexity analysis, incorporating Spectral Entropy (SpecEn) and fractal dimensions, was calculated across 19 EEG channels from a total of 34 subjects (16 normal and 18 with Mild Cognitive Impairment (MCI)). Performance validation of the proposed method was achieved through Linear Discriminant Analysis (LDA), yielding an accuracy of 82.4%, specificity of 77.8%, and sensitivity of 87.5%. The findings from this study suggest that EEG analysis can serve as a reliable tool for the early detection of AD.
Alzheimer, early detection, complexity, SpecEn, fractal
Alzheimer's Dementia (AD), the most prevalent form of dementia globally, continues to experience a rise in its incidence [1]. This chronic, neurodegenerative condition, predominantly afflicting the elderly population, is marked by cognitive impairment and memory loss [2, 3]. Despite not being directly lethal, AD's rapid progression significantly diminishes quality of life. The disease often induces considerable dependence on others, impacting social life and imposing substantial costs [4].
As of present, no definitive treatment protocols for AD exist [5]. Delayed intervention can precipitate swift deterioration, underscoring the importance of early detection and subsequent treatment in curbing this potentially aggressive degenerative process [6]. The primary pathogenesis of AD is believed to be attributable to extensive neuronal death and neurotransmitter damage. The accumulation of beta-amyloid plaques is strongly implicated in the disruption of inter-neuronal information transmission [7].
Biochemical modalities and medical imaging serve as standard biomarkers in AD diagnosis [3]. Biochemical modalities such as cerebrospinal fluid (CSF) are sensitive to pathological changes due to AD [8, 9]. However, these medical modalities necessitate invasive procedures and find limited use in clinical practice. Medical imaging techniques, including MRI, CT-Scan, and FDG-PET [10, 11], whilst sensitive to AD, are costly and typically available only in central hospitals [3].
In light of neuronal degradation and brain electrical activity, electroencephalogram (EEG) offers an alternative in AD analysis, encompassing early detection and severity assessment [12]. EEG's advantages of being low-cost, non-invasive, easily installed, and readily available in public health centers render it a preferred choice. EEG has long been relied upon as a primary biomarker or auxiliary tool in AD studies. Early detection of AD can occur well before the onset of pre-clinical symptoms or at the stage of mild cognitive impairment (MCI). The neuronal damage or death pivotal to AD induces changes in EEG signal characteristics. Power spectral analysis of EEG reveals a slowdown of waves [13, 14], indicative of reduced alpha wave activity [15, 16]. Furthermore, coherence and synchronization analyses exhibit a decrease in AD compared to normal aging [17, 18].
While EEG spectral analysis has been employed to characterize AD and assess its severity, it is notably sensitive to noise, a common issue in EEG signal. Therefore, a supplementary method is required to bolster spectral-based analysis. Given that the EEG signal is a cumulative result of complex, dynamic processes by neuron cells, a signal complexity-based analysis is considered proficient in characterizing EEG. Consequently, this study proposes an early detection methodology for AD based on EEG signals utilizing a signal complexity approach. The study analyzed recorded EEG signals from normal elderly individuals and those with MCI. A complexity approach integrating Spectral Entropy (SpecEn) and fractal dimension was deployed for EEG characterization. Linear discriminant analysis (LDA) was employed for evaluating the proposed method. This study aims to serve as a supportive and complementary method for AD early detection, supplementing spectral-based characterization.
The ensuing sections of the paper are organized as follows: Section 2 outlines the EEG dataset and the methods employed in this study. Section 3 presents the study findings and subsequent discussions. Section 4 concludes the paper, delineating the implications of the study, its conclusions, and prospects for future work.
2.1 System design
Figure 1 shows our proposed system. First, EEG signals entered the pre-processing stage that included Independent Component Analysis (ICA) denoising and Band Pass Filter (BPF) from 1-30 Hz. ICA denoising was applied because it is superior in separating noise from signal in the raw dataset. Not only that, but ICA also improve the data quality [19]. Pre-processing stage resulted filtered signals. Then, four different calculations were performed to extract the features, i.e., SpecEn, Katz, Higuchi, and Sevcik fractal dimension. In the final stage, the features were classified using LDA classifier. Each stage of the proposed system is briefly explained in the following subsections.
2.2 Normal dan MCI EEG dataset
This study used the scalp EEG dataset containing two conditions, namely normal and Mild Cognitive Impairment (MCI). There are 34 selected subjects (16 normal and 18 MCI) who were hospitalized in the units of cardiac catherization of Sina and Nour Hospitals located in Isfahan, Iran. In the current dataset, there are more EEG recordings than the previous release where there are 27 EEG recordings (16 normal and 11 MCI). Patients’ age range is 60-77 with coronary angiography history in recent year. Patients with a track record of major psychiatric disorders, dementia, head trauma, substance misuse, and other serious medical disease were not included. However, all subjects had to complete the neuropsychiatric interview considering Peterson’s criteria for MCI. An assessment tool, neuropsychiatry unit cognitive (NUCOG), has been used in confirming the diagnosis of MCI [20].
The recording session for all EEG signals were done in the mornings while subjects were comfortably rested with closed eyes in a quiet room and using the same EEG amplification system. There are 19 channels (Fp1, Fp2, F7, F3, Fz, F4, F8, T3, C3, Cz, C4, T4, T5, P3, Pz, P4, T6, O1, and O2) which represent 19 electrodes positioned corresponding to the 10-20 international system. The sampling frequency is 256Hz when recording the EEG signal continuously. During the recording process, patients were being checked out repeatedly to keep them aware and awake [20]. The length of the recording was about 30 minutes. In this dataset, EEG signals are stored in the European Data Format (*.edf) as a standard for exchange and storage.
2.3 Spectral entropy
Generally, entropy is a measurement of signal complexity. As the complexity increases, the resulted entropy value is higher. There are various entropy calculations that have been used in lots of EEG signals complexity measurement [21-24]. In this study, we estimate the signal regularity using spectral entropy (SpecEn). SpecEn calculates the power spectral distribution of the Fourier transform using Shannon entropy. SpecEn estimates signal dynamics in the frequency domain. Meanwhile, entropy is a measure of the uncertainty of a data series. So that, high spectral entropy represents the high dynamics of the power spectral at a certain frequency range. While low SpecEn represents spectral power with low dynamics or condensed into a single frequency [24].
SpecEn represents the probabilities of power spectral densities using Shannon’s entropy equation. The spectral entropy is normalized according to the range of frequency $[f 1, f 2]$ with Eq. (1) below.
Spectral entropy $[f 1, f 2]=-\frac{1}{\log [N[f 1, f 2]]} \sum_{f i=f 1}^{f 2} P_n\left(f_i\right) \log \left(P_n\left(f_i\right)\right)$ (1)
where, $[N[f 1, f 2]]$ is the total components of frequency in range $[f 1, f 2]$ and $P_n\left(f_i\right)$ is the probabilities of the total components of frequency [25, 26].
2.4 Fractal Dimension
One of the EEG characteristics is high complexity with self-similar patterns. Fractal Dimension (FD) is a useful technique to handle the self-similar pattern complexity. Generally, FD measures the waveform complexity in time series analysis [27]. A few methods of FD have been developed to support the EEG signals classification [28-30]. This study proposes an AD early detection method utilizing Higuchi, Katz, and Sevcik FD to estimate the EEG waveform complexity
2.4.1 Higuchi Fractal Dimension
Higuchi Fractal Dimension (HFD) occurs in a time-series $A(t)$ to calculate the value of FD that represents the complexity of EEG waveform. A new time-series $A(t)_k^m$ and the length of the series are defined in Eqs. (2)-(3) [31].
$A(t)_{k}^{m}={{a}_{m}},{{a}_{m+k}},{{a}_{m+2k}},...,{{a}_{m+\left[ \frac{N-m}{k} \right]k}\quad}$ (2)
$L_m(k)=\frac{\left(\sum_{i=1}^{\left[\frac{N-m}{k}\,\,\,\right]}\,\,\,\left|a_{i m+k}-a_{(i-1) k}\right|\,\,\,\right)\,\,\,\frac{N-1}{\left[\frac{N-m}{k}\,\,\,\right] k}}{k}$ (3)
Figure 1. The proposed system
where, m = 1, 2, …, k is integer which state the time delay with the maximum interval time is $k_{\max }$, symbol of $[\ldots]$ is the Gauss notation, and the term of (4) [32], specifies the factor of normalization at the length of FD. Then, the average length of the series is obtained using Eq. (5) below.
$\frac{N-1}{\left[\frac{N-m}{k}\quad\right]k}$ (4)
$\langle L(k)\rangle=\sum_{m=1}^k \frac{L_i(k)}{k}$ (5)
Then, if the results are plotted on a log-log graph, the HFD can be obtained from a slope of the linear regression plot between $\ln (L(k))$ to $\ln (1 / k)$. The obtained curve defines the HFD with $D$ dimension after the value $L(k) \alpha k^{-D}$ is complied.
2.4.2 Katz Fractal Dimension
Katz Fractal Dimension (KFD) is a direct calculation from the time-series $A(t)$ that starts with its transpose array [33, 34]. When N is length of the curve, the measurement of KFD is written as Eq. (6) [35].
$D=\frac{\log \left( n \right)}{\log \left( n \right)+\log \left( \frac{d}{L} \right)}=\frac{\log \left( \frac{L}{a} \right)}{\log \left( \frac{d}{a} \right)}$ (6)
where, L represents the signal length, a defines average distance between the successive points, and n is the total of steps presented in the graphic of curve [32].
2.4.3 Sevcik Fractal Dimension
A newer method to calculate the FD of a waveform in a quicker way is Sevcik Fractal Dimension (SFD) [36]. The waveform dimension, complexity, and randomness can be quickly estimated and measured since the calculation characteristics are simple and fast [37]. SFD approximates the FD D from the sampled time series of N dimension. Based on [36], the estimation of D is obtained from the Hausdorff Dimension $D_h$ of a waveform by plotting the N points of curve L to cells of a unit square $N \times N$ in a metric space that has been normalized, through transformation of double linear. Eq. (7) defines the SFD calculation [35].
${{D}_{h}}=1+\frac{\ln \left( L \right)}{\ln \left[ 2.\left( N-1 \right) \right]}$ (7)
where, L is the length of a curve.
2.5 Linear Discriminant Analysis
Linear Discriminant Analysis (LDA) is a supervised classifier that creates a combination of the original predictors as its new variable [38, 39]. The target of LDA is to achieve a predictor value from the new variable which combines the actual class and the actual predictors [40]. LDA does the grouping process of the objects into groups which are mutually discriminative. If there are two classes in the EEG dataset, the used discriminant function will be one. The required training samples are fewer, yet the results are as good as the other classifiers [41]. The discriminant function can be written mathematically in Eq. (8) [38].
$D=\sum\limits_{j=1}^{p}{{{w}_{j}}{{Z}_{j}}}$ (8)
where, w is the weight factor.
Essential features are calculated using SpecEn, HFD, KFD, and SFD for each EEG channel. So that each channel will produce four features. These features then become predictors in the classification stage. The proposed method was validated using LDA. Seeing that the number of datasets is relatively small, the distribution of training and test data uses the cross validation (CV) method where 5-CV is used in this study. Several scenarios were simulated to observe which method generated the highest accuracy. The scenario includes grouping features by area, combining the resulting feature vectors from all feature extraction methods, and observing accuracy when applying one feature extraction method. The average of features calculated in each area includes frontal (Fp1, Fp2, F3, F4), right temporal (F8, T4, T4), left temporal (F7, T3, T5), central (C3, C4, Fz, Cz, Pz), and occipital (P3, P4, O1, O2) [20]. Table 1 shows the details of each scenario.
Table 1. Scenarios used in simulation
|
Scenario |
Classification Feature |
|
A |
Average area (SpecEn, HFD, KFD, SFD) |
|
B |
All channels (SpecEn, HFD, KFD, SFD) |
|
C |
SpecEn (Average area) |
|
D |
HFD (Average area) |
|
E |
KFD (Average area) |
|
F |
SFD (Average area) |
This study implements the measurement accuracy, specificity and sensitivity to validate the performance of the system which defined in Eqs. (9) (10) and (11), respectively.
$Acc=\frac{TP+TN}{TN+FP+TP+FN}\quad\times 100$ (9)
$Spec=\frac{TN}{TN+FP}\times 100$ (10)
$Sen=\frac{TP}{TP+FN}\times 100$ (11)
where, Acc is accuracy, Spec is specificity, Sen is sensitivity, TP is true positive, TN is true negative, FP is false positive, and FN is false negative.
In this study, the duration of the EEG signal processed was five minutes, which was taken from the beginning of the recording. This is indeed an assumption, to avoid EEG signals containing drowsiness as a consequence of a long recording. The EEG signal resulting from noise rejection by ICA and BPF is then calculated for its spectral entropy and fractal. Figure 2 and Figure 3 show the estimated signal complexity calculated using spectral entropy and fractal dimension.
Figure 2 shows that the SpecEn value of the MCI group is lower than that of the normal group. With the T-Test (confidence level = 95%), significant differences were found at the electrodes Fp1, Fp2, T6 and O1. These results indicate that there is a decrease in the complexity of the EEG signal in MCI patients, particularly in the frontal and occipital areas. The same trend is also shown in Fig. 3 where the signal complexity of the MCI group is lower than the normal group. These characteristics will then be used for MCI and normal discrimination in the next stage. The results of the performance validation using LDA and 5-CV for each scenario are shown in Table 1.
Table 2 shows that the highest accuracy is 82.4% with specificity and sensitivity 77.78% and 87.5%, respectively. The confusion matrix is presented in Table 3. The highest accuracy was obtained in scenario B, where all features from SpecEn and FDs are used. Meanwhile, if only one feature extraction method was applied (scenario C-F), the highest accuracy was achieved by the scenario F with an accuracy of 76.5%. In scenario C, SpecEn generated the lowest accuracy among the other methods. This occurred because the SpecEn is very distorted to the natural noise of the EEG so that it delivered greater bias between normal and MCI than the fractal method. Additionally, SpecEn estimates the complexity of the signal based on spectral analysis, which is very vulnerable to the influence of noise. Scenario A was done by grouping of features based on brain area and the accuracy was 70.6% which indicates lower than when all features were used as predictors, as in scenario B.
The brain represents a complex process of resultant large amounts of neuronal activity. Pathological changes due to a disease can affect the electrical activity of neurons including changes in the degree of signal complexity. The decline in complex biological mechanisms can be an indicator of decreased ability to respond to stimuli. Decreased cognitive function in the elderly with AD occurs due to damage to neurons or neurotransmitters in the brain, causing disruption of communication between cells. This affects changes in the EEG signal and thus may underlie the early detection of AD proposed in this study.
Figure 2. Mean of the SpecEn values of each electrode in the normal and MCI groups
Figure 3. Mean of the KFD values of each electrode in the normal and MCI groups
Table 2. Accuracy, specificity, and sensitivity of each scenario
|
Scenario |
Linear Discriminant |
||
|
Acc (%) |
Spec (%) |
Sen (%) |
|
|
A |
70.6 |
68.75 |
72.22 |
|
B |
82.4 |
77.78 |
87.5 |
|
C |
55.9 |
53.33 |
57.89 |
|
D |
73.5 |
76.92 |
71.43 |
|
E |
70.6 |
71.43 |
70 |
|
F |
76.5 |
72.22 |
81.25 |
Table 3. The confusion matrix of the highest accuracy
|
|
MCI |
Normal |
|
MCI |
14 |
4 |
|
Normal |
2 |
14 |
From all the simulated test scenarios, it is known that the highest accuracy was achieved by scenario B, which was the combination of all feature extraction methods. This study is superior to previous studies because in this study uses a larger amount of data. The proposed method in this study yielded a classification accuracy of > 80% and can be considered for early detection of AD in this case as a supplementary diagnostic criterion besides memory testing. From the low specificity values, it could be seen that it was very challenging to detect AD in the early stages since EEG MCI has similar characteristics to normal EEG signals. This happened because in the early stages of AD, neuronal damage only occurs in a partially human brain. But, an EEG analysis for early detection of AD can be considered a tool with several advantages that it is low-cost, non-invasive, and can be used continuously compared to medical imaging techniques.
People with AD tend to have particular changes of the brain waves that can be examined through EEG signals. To decide the most suitable treatment, the examination needs to be as accurate as possible. Meanwhile, the transformation from normal elderly to AD, and vice versa, is slightly different, so an automated diagnosis is indispensable. One of the characteristics of EEG signals is complex and dynamic. Therefore, this study proposed an early detection of AD based on EEG signals focusing on the signal complexity to distinguish the conditions defined in the EEG dataset used which are normal and MCI. SpecEn and FD were used as the methods for signal complexity-based approach. All features were classified using LDA to classify the normal and MCI conditions using six scenarios. The highest accuracy was performed by scenario B that employed features from all feature extraction methods in all areas. Obtaining accuracy more than 82.4% verifies that EEG is suitable for further analysis of AD detection. Furthermore, the specificity and sensitivity reached 77.8% and 87.5%, respectively, show that the proposed method is ready to support the early detection of AD in large population. However, detecting AD still faces many challenges seeing that only partial locations of brain that are affected by neuronal damage.
All authors would like to thank Telkom University for funding this research. The authors are also grateful to the Laboratory of School of Applied Science and School of Electrical Engineering for providing research facilities.
[1] Arvanitakis, Z., Shah, R.C., Bennett, D.A. (2019). Diagnosis and management of dementia: A review. JAMA, 322(16): 1589-1599. https://doi.org/10.1001/jama.2019.4782
[2] Angelucci, F., Spalletta, G., Iulio, F.D., Ciaramella, A., Salani, F., Varsi, A.E., Gianni, W., Sancesario, G., Caltagirone, C., Bossu, P. (2010). Alzheimer's disease (AD) and Mild Cognitive Impairment (MCI) patients are characterized by increased BDNF serum levels. Current Alzheimer Research, 7(1): 15-20.
[3] Al-Nuaimi, A.H.H., Al-Juboori, S., Jammeh, E., Sun, L., Ifeachor, E. (2019). Electroencephalogram based biomarkers for detection of Alzheimer’s disease. In Neuroimaging-Neurobiology, Multimodal and Network Applications. IntechOpen, pp. 1-16. https://doi.org/10.5772/intechopen.90015
[4] Cunningham, E.L., McGuinness, B., Herron, B., Passmore, A.P. (2015). Dementia. The Ulster Medical Journal, 84(2): 79-87.
[5] Grossberg, G.T., Tong, G., Burke, A.D., Tariot, P.N. (2019). Present algorithms and future treatments for Alzheimer’s disease. Journal of Alzheimer's Disease, 67(4): 1157-1171. https://doi.org/10.3233/JAD-180903
[6] Pais, M., Martinez, L., Ribeiro, O., Loureiro, J., Fernandez, R., Valiengo, L., Canineu, P., Stella, F., Talib, L., Radanovic, M., Forlenza, O.V. (2020). Early diagnosis and treatment of Alzheimer’s disease: New definitions and challenges. Brazilian Journal of Psychiatry, 42(4): 431-441. https://doi.org/10.1590/1516-4446-2019-0735
[7] Murphy, M.P., LeVine III, H. (2010). Alzheimer's disease and the amyloid-β peptide. Journal of Alzheimer's Disease, 19(1): 311-323. https://doi.org/10.3233/JAD-2010-1221
[8] Paterson, R.W., Slattery, C.F., Poole, T., et al. (2018). Cerebrospinal fluid in the differential diagnosis of Alzheimer’s disease: Clinical utility of an extended panel of biomarkers in a specialist cognitive clinic. Alzheimer's Research & Therapy, 10(1): 1-11. https://doi.org/10.1186/s13195-018-0361-3
[9] McGrowder, D.A., Miller, F., Vaz, K., Nwokocha, C., Wilson-Clarke, C., Anderson-Cross, M., Brown, J., Anderson-Jackson, L., Williams, L., Latore, L., Thompson, R., Alexander-Lindo, R. (2021). Cerebrospinal fluid biomarkers of Alzheimer’s disease: Current evidence and future perspectives. Brain Sciences, 11(2): 215. https://doi.org/10.3390/brainsci11020215
[10] van Oostveen, W.M., de Lange, E.C. (2021). Imaging techniques in Alzheimer’s disease: A review of applications in early diagnosis and longitudinal monitoring. International Journal of Molecular Sciences, 22(4): 2110. https://doi.org/10.3390/ijms22042110
[11] Park, M., Moon, W.J. (2016). Structural MR imaging in the diagnosis of Alzheimer's disease and other neurodegenerative dementia: Current imaging approach and future perspectives. Korean Journal of Radiology, 17(6): 827-845. https://doi.org/10.3348/kjr.2016.17.6.827
[12] Al-Qazzaz, N.K., Ali, S.H.B., Ahmad, S.A., Chellappan, K., Islam, M.S., Escudero, J. (2014). Role of EEG as biomarker in the early detection and classification of dementia. The Scientific World Journal, 2014: 906038. https://doi.org/10.1155/2014/906038
[13] Schumacher, J., Taylor, J.P., Hamilton, C.A., Firbank, M., Cromarty, R.A., Donaghy, P.C., Roberts, G., Allan, L., Lloyd, J., Durcan, R., Barnett, N., O’Brien, J.T., Thomas, A.J. (2020). Quantitative EEG as a biomarker in mild cognitive impairment with Lewy bodies. Alzheimer's Research & Therapy, 12: 1-12. https://doi.org/10.1186/s13195-020-00650-1
[14] Meghdadi, A.H., Stevanović Karić, M., McConnell, M., Rupp, G., Richard, C., Hamilton, J., Salat, D., Berka, C. (2021). Resting state EEG biomarkers of cognitive decline associated with Alzheimer’s disease and mild cognitive impairment. PloS One, 16(2): e0244180. https://doi.org/10.1371/journal.pone.0244180
[15] Fauzan, N., Amran, N.H. (2015). Early detection of mild cognitive impairment, dementia and Alzheimer’s using Qeeg. European Journal of Interdisciplinary Studies, 1: 149-153.
[16] Ishii, R., Canuet, L., Aoki, Y., Hata, M., Iwase, M., Ikeda, S., Nishida, K., Ikeda, M. (2018). Healthy and pathological brain aging: From the perspective of oscillations, functional connectivity, and signal complexity. Neuropsychobiology, 75(4): 151-161. https://doi.org/10.1159/000486870
[17] Handayani, N., Haryanto, F., Khotimah, S.N., Arif, I., Taruno, W.P. (2018). Coherence and phase synchrony analyses of EEG signals in Mild Cognitive Impairment (MCI): A study of functional brain connectivity. Polish Journal of Medical Physics and Engineering, 24(1): 1-9. https://doi.org/10.2478/pjmpe-2018-0001
[18] Laptinskaya, D., Fissler, P., Küster, O.C., Wischniowski, J., Thurm, F., Elbert, T., von Arnim, C.A.F., Kolassa, I.T. (2020). Global EEG coherence as a marker for cognition in older adults at risk for dementia. Psychophysiology, 57(4): e13515. https://doi.org/10.1111/psyp.13515
[19] Carone, D., Harston, G.W.J., Garrard, J., De Angeli, F., Griffanti, L., Okell, T.W., Chappell, M.A., Kennedy, J. (2019). ICA-based denoising for ASL perfusion imaging. NeuroImage, 200: 363-372. https://doi.org/10.1016/j.neuroimage.2019.07.002
[20] Kashefpoor, M., Rabbani, H., Barekatain, M. (2016). Automatic diagnosis of mild cognitive impairment using electroencephalogram spectral features. Journal of Medical Signals and Sensors, 6(1): 25.
[21] Morabito, F.C., Labate, D., Foresta, F.L., Bramanti, A., Morabito, G., Palamara, I. (2012). Multivariate multi-scale permutation entropy for complexity analysis of Alzheimer’s disease EEG. Entropy, 14(7): 1186-1202. https://doi.org/10.3390/e14071186
[22] Siuly, S., Alçin, Ö.F., Kabir, E., Şengür, A., Wang, H., Zhang, Y., Whittaker, F. (2020). A new framework for automatic detection of patients with mild cognitive impairment using resting-state EEG signals. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 28(9): 1966-1976. https://doi.org/10.1109/TNSRE.2020.3013429
[23] Echegoyen, I., López-Sanz, D., Martínez, J.H., Maestú, F., Buldú, J.M. (2020). Permutation entropy and statistical complexity in mild cognitive impairment and Alzheimer’s disease: An analysis based on frequency bands. Entropy, 22(1): 116. https://doi.org/10.3390/e22010116
[24] Abásolo, D., Hornero, R., Espino, P., Alvarez, D., Poza, J. (2006). Entropy analysis of the EEG background activity in Alzheimer's disease patients. Physiological Measurement, 27(3): 241. https://doi.org/10.1088/0967-3334/27/3/003
[25] Shri, T.P., Sriraam, N. (2016). Spectral entropy feature subset selection using SEPCOR to detect alcoholic impact on gamma sub band visual event related potentials of multichannel electroencephalograms (EEG). Applied Soft Computing, 46: 441-451. https://doi.org/10.1016/j.asoc.2016.04.041
[26] Padma Shri, T.K., Sriraam, N. (2017). Pattern recognition of spectral entropy features for detection of alcoholic and control visual ERP’s in multichannel EEGs. Brain Informatics, 4(2): 147-158. https://doi.org/10.1007/s40708-017-0061-y
[27] Vidyaratne, L.S., Iftekharuddin, K.M. (2017). Real-time epileptic seizure detection using EEG. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 25(11): 2146-2156. https://doi.org/10.1109/TNSRE.2017.2697920
[28] Harne, B.P. (2014). Higuchi fractal dimension analysis of EEG signal before and after OM chanting to observe overall effect on brain. International Journal of Electrical & Computer Engineering (2088-8708), 4(4): 585-592.
[29] Al-Nuaimi, A.H.H., Jammeh, E., Sun, L., Ifeachor, E. (2018). Complexity measures for quantifying changes in electroencephalogram in Alzheimer’s disease. Complexity, 2018: 8915079. https://doi.org/10.1155/2018/8915079
[30] Jacob, J.E., Gopakumar, K. (2018). Automated diagnosis of encephalopathy using fractal dimensions of EEG sub-bands. In 2018 IEEE Recent Advances in Intelligent Computational Systems (RAICS), Thiruvananthapuram, India, pp. 94-97. https://doi.org/10.1109/RAICS.2018.8635062
[31] Varley, T.F., Craig, M., Adapa, R., Finoia, P., Williams, G., Allanson, J., Pickard, J., Menon, D.K., Stamatakis, E.A. (2020). Fractal dimension of cortical functional connectivity networks & severity of disorders of consciousness. PloS One, 15(2): e0223812. https://doi.org/10.1371/journal.pone.0223812
[32] Garner, D.M., De Souza, N.M., Vanderlei, L.C.M. (2018). Heart rate variability analysis: Higuchi and Katz’s fractal dimensions in subjects with type 1 diabetes mellitus. Romanian Journal of Diabetes Nutrition and Metabolic Diseases, 25(3): 289-295.
[33] Katz, M.J. (1988). Fractals and the analysis of waveforms. Computers in Biology and Medicine, 18(3): 145-156. https://doi.org/10.1016/0010-4825(88)90041-8
[34] Nikolopoulos, D., Moustris, K., Petraki, E., Koulougliotis, D., Cantzos, D. (2019). Fractal and long-memory traces in PM10 time series in Athens, Greece. Environments, 6(3): 29. https://doi.org/10.3390/environments6030029
[35] Wijayanto, I., Hartanto, R., Nugroho, H.A. (2020). Comparison of empirical mode decomposition and coarse-grained procedure for detecting pre-ictal and ictal condition in electroencephalography signal. Informatics in Medicine Unlocked, 19: 100325. https://doi.org/10.1016/j.imu.2020.100325
[36] Sevcik, C. (2006). On fractal dimension of waveforms. Chaos Solitons and Fractals, 28(2): 579-580. https://doi.org/10.1016/j.chaos.2005.07.003
[37] Xue, R., Liu, J., Tang, H. (2020). Two-dimensional jamming recognition algorithm based on the Sevcik fractal dimension and energy concentration property for UAV frequency hopping systems. Information, 11(11): 520. https://doi.org/10.3390/info11110520
[38] Subasi, A., Gursoy, M.I. (2010). EEG signal classification using PCA, ICA, LDA and support vector machines. Expert Systems with Applications, 37(12): 8659-8666. https://doi.org/10.1016/j.eswa.2010.06.065
[39] Nkengfack, L.C.D., Tchiotsop, D., Atangana, R., Louis-Door, V., Wolf, D. (2020). EEG signals analysis for epileptic seizures detection using polynomial transforms, linear discriminant analysis and support vector machines. Biomedical Signal Processing and Control, 62: 102141. https://doi.org/10.1016/j.bspc.2020.102141
[40] Chen, D.W., Miao, R., Yang, W.Q., Liang, Y., Chen, H.H., Huang, L., Deng, C., Han, N. (2019). A feature extraction method based on differential entropy and linear discriminant analysis for emotion recognition. Sensors, 19(7): 1631. https://doi.org/10.3390/s19071631
[41] Dodia, S., Edla, D.R., Bablani, A., Ramesh, D., Kuppili, V. (2019). An efficient EEG based deceit identification test using wavelet packet transform and linear discriminant analysis. Journal of Neuroscience Methods, 314: 31-40. https://doi.org/10.1016/j.jneumeth.2019.01.007
