Computational and Mathematical Methods in Medicine

Volume 2015 (2015), Article ID 953868, 8 pages

http://dx.doi.org/10.1155/2015/953868

## A Novel Application of Multiscale Entropy in Electroencephalography to Predict the Efficacy of Acetylcholinesterase Inhibitor in Alzheimer’s Disease

^{1}Neurology Department, National Yang-Ming University Hospital, Yi-Lan, Taiwan^{2}Neurology Department, National Yang-Ming University School of Medicine, Taipei, Taiwan^{3}Division of Pulmonology, Department of Internal Medicine, National Yang-Ming University Hospital, Yi-Lan, Taiwan^{4}Department of Internal Medicine, College of Medicine, National Yang-Ming University School of Medicine, Taipei, Taiwan^{5}Graduate Institute of Biomedical Electronics and Bioinformatics, National Taiwan University, Taipei, Taiwan^{6}Center for Dynamical Biomarkers and Translational Medicine, National Central University, Chungli, Taiwan^{7}Research Center for Adaptive Data Analysis, National Central University, No. 300, Jhongda Road, Taoyuan 32001, Taiwan

Received 1 December 2014; Accepted 2 January 2015

Academic Editor: Kazeem O. Okosun

Copyright © 2015 Ping-Huang Tsai et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

#### Abstract

Alzheimer’s disease (AD) is the most common form of dementia. According to one hypothesis, AD is caused by the reduced synthesis of the neurotransmitter acetylcholine. Therefore, acetylcholinesterase (AChE) inhibitors are considered to be an effective therapy. For clinicians, however, AChE inhibitors are not a predictable treatment for individual patients. We aimed to disclose the difference by biosignal processing. In this study, we used multiscale entropy (MSE) analysis, which can disclose the embedded information in different time scales, in electroencephalography (EEG), in an attempt to predict the efficacy of AChE inhibitors. Seventeen newly diagnosed AD patients were enrolled, with an initial minimental state examination (MMSE) score of . After 12 months of AChE inhibitor therapy, 7 patients were responsive and 10 patients were nonresponsive. The major difference between these two groups is Slope 2 (MSE6 to 20). The area below the receiver operating characteristic (ROC) curve of Slope 2 is 0.871 (95% CI = 0.69–1). The sensitivity is 85.7% and the specificity is 60%, whereas the cut-off value of Slope 2 is −0.024. Therefore, MSE analysis of EEG signals, especially Slope 2, provides a potential tool for predicting the efficacy of AChE inhibitors prior to therapy.

#### 1. Introduction

Alzheimer’s disease (AD) is the most common form of dementia [1, 2], with the dominant presentation of a progressive decline in cognitive functioning beyond what is expected from normal aging. The neurodegeneration in AD may be caused by deposition of amyloid beta-peptide in plaques or formation of neurofibrillary tangle in brain tissue [1, 2]. Although little is known of the actual cause of AD, many of its symptoms are generally accepted to be related to a cholinergic deficit in the cerebral cortex and other areas of the brain [3–5]. Acetylcholinesterase (AChE) inhibitors, which inhibit the acetylcholinesterase enzyme from breaking down acetylcholine and thereby increasing the level and duration of the neurotransmitter acetylcholine activity, have been proven to be an effective therapy for AD [6–11]. Pharmacoeconomic studies have demonstrated that therapies can postpone dementia from progressing to more severe stages and may also result in economic benefits for patients’ families and caregivers, as well as for society [12–16]. However, clinicians have argued that AChE inhibitors have an effect on a subgroup of only 25–50% of AD patients [17–19], which cannot be identified objectively prior to therapy. Furthermore, the time scale for measuring the effect of AChE inhibitors can last anywhere between several months to several years.

Recently, numerous studies have attempted to identify a prognostic predictor of AD by using artificial neural networks [20], brain magnetic resonance imaging (MRI) [18, 21], single-photon emission computed tomography (SPECT) [22], and cognitive function tests [23]. However, technical dependence, high costs, contrast-agent related allergies, and potential exposure to radionuclide irradiation have limited their clinical application. On the other hand, numerous forms of quantitative electroencephalographic (EEG) analyses have been used to elucidate the characteristics of EEGs to enhance diagnostic power with the assistance of signal processing, suggesting a potential objective tool in the evaluation of AD [24–36]. The surface EEG represents the electrical activity of innumerable cortical neurons, which determine the fundamental patterns that indicate the interaction between various mechanisms with multiple and spatial scales. As it is quite difficult to explain the underlying neurophysiology mechanism, nonlinear based methods have recently been used more frequently to explore the EEG activity [37, 38]. The majority of the studies that have used nonlinear methods showed the loss of the complexity of EEG signals to be correlated with the severity of the dementia [31, 32, 34, 35]. However, it is difficult to determine which specific physiological mechanism degrades the complexity of the EEG, and, as a result, clinicians lack information regarding the effective responder to AChE inhibitors in AD patients.

In light of fundamental nonlineal theory, biological signals represent the outcome of the nonlinear interactions between different processes at multiple temporal and spatial scales, including EEGs and electrocardiograms (ECG). With this understanding, some studies proposed a careful examination of changes in nonlinear indices with scales. The most well-known of these changes is the crossover phenomenon of the fractal correlation exponents between short and long time scales in the detrended fluctuation analysis (DFA) [39] of heart-rate dynamics. The short-term exponent is understood to be determined mostly by cardiorespiratory interaction [39, 40]. Recently, the studies of activity fluctuation with aging and in AD [41, 42] determined that fractal correlations at certain scales (i.e., 1.5–8.0 hours) declined with age. These studies also determined that an age-independent AD effect further reduced the correlations at these scales, leading to the greatest reduction of the correlations in very old people with late-stage AD. This result closely resembles the loss of correlations at long time scales in suprachiasmatic nucleus- (SCN-) lesioned animals [43]. In addition to DFA, multiscale entropy (MSE) analysis, proposed by Costa et al. [44, 45], is a possible method for measuring the complexity of nonlineal signals at different temporal scales by means of an entropy-based algorithm. In our previous study, we proposed a detrending process [31, 46, 47] to attenuate the spurious influence caused by nonstationarity in the real world. This study also demonstrated many features of the scale function of entropy, including the slope of the scale function of entropy used to check the existence of fractal correlation (a negative slope indicates a greater number of random fluctuations corresponding to loss of interaction), which can assist with clinical classification [47]. For example, sympathetic and parasympathetic activities are correlated with MSE on different scales of the heartbeat sequence (scales 3–5 and 1–4 for sympathetic activity and parasympathetic activity, resp.) [46]. Furthermore, congestive heart failure patients without *β*-blocker therapy exhibit a very negative slope for MSE1–5, indicating the lack of interaction to be a result of respiratory regulation [40].

Tracing the activation of the neurotransmitter ACh, which is released from the presynaptic vesicles that are triggered by action potential (electrical impulse), diffused in the synapse, and then finally conjugated to the postsynaptic receptor (5–50 msec) [48–50], these activations are observed at different time scales, each of which may be studied using MSE analysis.

In this study, we hypothesize that some features, such as slope, of MSE analysis of EEG data can be associated with the therapy efficacy of AChE inhibitors in AD patients, relying on the ability of MSE to demonstrate different mechanisms with multiple temporal or spatial scales.

#### 2. Materials and Methods

##### 2.1. Participants

All participants in this study were diagnosed with dementia and enrolled from the neurological clinic at the National Yang Ming University Hospital. After screening for dementia, all participants received further medical, neurological, neuropsychological, and psychiatric assessments, as well as blood examinations. Neurological assessments for all participants included cerebral computed tomography (CT) scans to rule out intracranial pathology (i.e., brain tumors or stroke) that may have contributed to cognitive decline. Trained research assistants administered the Chinese version of the minimental state examination (MMSE) [51], which features a total score of 30. The clinical dementia rating (CDR) was also used to determine the severity of dementia after a neurologist conducted separate semistructured interviews with the patient and a knowledgeable informant. The scores are as follows: 0 for normal, 0.5 for questionable, 1 for mild, 2 for moderate, and 3 for severe [52]. AD was diagnosed according to the diagnostic criteria of the Diagnostic and Statistical Manual of Psychiatric Disorders, 4th revised [53]. EEG was not included as part of the routine workup.

The inclusion criteria for participation in the EEG study included (1) a diagnosis of probable AD based on the Diagnostic and Statistical Manual of Psychiatric Disorders, 4th revised (DSM-IV) criteria [53] and (2) a rating of either 1 or 2 on the CDR scale [52]. Patients with dementia that may have been caused by reasons other than AD, as determined by the patient’s history, neurological examinations, imaging, and blood studies, were excluded from this study so as to ensure that alterations in EEGs were caused exclusively by AD. The research was approved by the institutional review board of National Yang-Ming University Hospital, Yi-Lan, Taiwan, a local community teaching hospital. Written informed consent was obtained from each participant prior to the study.

Seventeen newly diagnosed AD patients (9 men and 8 women, mean age ) were enrolled in this study, with an initial MMSE score of . Two of these patients were moderately demented (CDR = 2), whereas the remaining 15 were mildly demented (CDR = 1). None of the patients received any antipsychotics prior to the first EEG. Each participant received the AChE inhibitor donepezil hydrochloride (Aricept) 5 mg/d for 12 months and follow-up MMSE 12 months after therapy.

##### 2.2. EEG Recordings

EEGs were recorded for all subjects. In accordance with International Federation of Clinical Neurophysiology (IFCN) standards for the digital recording of clinical EEG [54], the surface EEG signals were recorded using a digital EEG recorder (NicoletOne) from the 19 scalp loci of the international 10–20 system (channels Fp1, Fp2, F3, F4, C3, C4, P3, P4, O1, O2, F7, F8, T3, T4, T5, T6, Fz, Cz, and Pz), with all electrodes referenced to the bilateral ear (A1, A2) for 30 min. ECGs were recorded on a separate channel as a part of EEG recordings. The sample frequency was 256 Hz and the analog-to-digital converter digitalized resolution was 12 bits. All the patients remained awake in a relaxed state in order to undergo a 30 min eyes-closed EEG.

##### 2.3. Signal Processing and Analysis: Multiscale Entropy Analysis

We adopted MSE as a complexity measure to feature digitalized EEG signals. Before conducting the analysis, one 30 sec artifact-free epoch EEG signal was selected by an experienced neurologist. Furthermore, empirical mode decomposition [55], which is considered superior to the traditional linear quantitative method to ascertain the intrinsic complexity, was utilized to remove the trend (<1 Hz) from the signals as was also the case in our previous study [31]. The MSE method is comprised of the following two steps: (1) coarse-graining the signals into different time scales and (2) quantifying the degree of irregularity in each coarse-grained time series using sample entropy (SampEn) [56, 57].

###### 2.3.1. Sampling Entropy (SampEn)

represents a time series of data length . The -length vector extracted from time series was defined by where represents the number of vectors, that are close to the given vector, . To obtain , we first defined the distance, between the given vector and as the maximum absolute difference between their corresponding scalar elements, that is, . Then, was defined by counting the number of with smaller than the threshold, , a given prior. Note that here ; that is, self-matches were not included to avoid the bias of ApEn [58]. The ratio of to total number of* m*-vectors extracted from the time series was evaluated, and the result was denoted as . The above steps were repeated to find from to ; then the natural logarithm of each was calculated and then averaged over to obtain . Theoretically, sample entropy [56, 57] is defined as . Note that the average of can be regarded as the probability that any two vectors extracted from time series are similar in some sense; therefore, and conceptually represents the average of the natural logarithm of the conditional probability that the sequences that are close to each other for consecutive data points will still be close to each other when one more point is given.

###### 2.3.2. Multiscale Entropy (MSE)

SampEn is a statistical measure based on information theory used to quantify the irregularity of a sequence of data [56–58]. It examines a time series for occurrences of similar epochs of a preassigned length; similarity is determined with respect to whether the distance between epochs is under a tolerance or not. More specifically, it calculates the negative natural logarithm of the conditional probability (i.e., those epochs) similar for points that will remain so when one additional point is added to the subseries [56–58]. The higher the SampEn is, the more unpredictable the data sequence is. However, complicated systems typically generate highly irregular output [37] and also exhibit dynamically diverse tendencies on various time scales, including, for example, the coexistence of slow and fast phenomena. Although SampEn provides an adequate way of quantifying irregularities for physiological output, it is not sufficient alone to cope with the complex content for the purpose of physiological research. In contrast, MSE analysis, based on evaluating the sample entropy at multiple time scales, has proven useful in this regard. To recast a signal in another scale, the original series were segregated into blocks, where each block contained data points. The mean value over each block then formed the coarse-grained time series at scale . It is clear that the time series at coarse-grained scale 1 is identical to the original signal. Equipped with multiple scales, the MSE method can disclose the temporal structures as well as scale-dependent characteristics of signals.

###### 2.3.3. Slope of MSE

In this study, the MSE value of each lead was calculated individually, which amounts to 380 values for each patient. Slopes of the mean value of MSE, averaged out of all leads, over scales 1–5 (Slope 1) and 6–20 (Slope 2) were estimated using the least-squares method (i.e., one Slope 1 and one Slope 2 for each patient). Throughout the analysis, we set and times the standard deviation of the processed signal, as was the case in our previous study [31].

##### 2.4. Statistical Analyses

Based on the differences between the follow-up and the initial MMSE scores, the patients were divided into two groups: a responder group and a nonresponder group. A responder was identified as a patient whose follow-up MMSE score was greater than or equal to the initial score; patients whose follow-up MMSE scores were less than the initial score were classified as nonresponders.

Descriptive statistics were presented as means ± standard deviation (SD). SPSS (ver16.0) for Windows (SPSS Inc., Chicago, IL, USA) was used for analysis. Baseline demographic characteristics, including age, MMSE, CDR, and MSE of each lead, and Slope 1 and Slope 2 were coded as continuous variables. Other demographic characteristics, such as sex, were coded as dichotomous variables. All of these characteristics were treated as predicable variables for treatment response. Because of the small sample size in this study, a Mann-Whitney* U* test (instead of Student’s* t*-test) was used to determine which factors were significant. A forward logistic regression, being a regression model in which the choice of predictive variables is carried out by an automatic procedure, which involves starting with no variables in the model, testing the addition of each variable using a chosen model comparison criterion, adding the variable (if any) that improves the model the most, and repeating this process until none improves the model was then used to calculate the odds ratio in the variants. All of the statistical tests were two-tailed and significance levels were set at a value of less than 0.05.

By conducting the receiver operating characteristic (ROC) analysis and considering the optimal combination of sensitivity and specificity, we determined the best cut-off points. Likelihood ratios and positive and negative predictive values with 95% confidence intervals (CI) were assessed at each of the cut-off point levels.

#### 3. Results

Seven patients (3M/4F, mean age ) were categorized as responders and ten patients (6M/4F, mean age ) were categorized as nonresponders. There were no significant differences in age and gender between the patients in these two groups. The initial MMSE score in the responder group was 15.6 5.1, and in the nonresponder group it was 20.9 2.8 (). The MSE values in each lead were significantly higher in F7 MSE7, Fz MSE6, Fz MSE7, Fz MSE8, C4 MSE5, T4 MSE6, T4 MSE7, T4 MSE9, Pz MSE7, Pz MSE8, and O1 MSE7 (Table 1). Slope 1 was for the responder group and for the nonresponder group () (Table 1 and Figure 1). Slope 2 was − for the responder group and − for the nonresponder group () (Table 1 and Figure 1). After applying the forward logistic regression, only Slope 2 was preserved. The other factors, including the initial MMSE, were excluded.