Abstract—We investigated the correlation of Alzheimer´s disease (AD) severity as measured by the Mini-Mental State Examination (MMSE) to the signal complexity measures auto- mutual information, Shannon entropy and Tsallis entropy in 79 patients with probable AD from the multi-centric Prospective Dementia Database Austria (PRODEM). Using quadratic (line- ar) regressions, auto-mutual information explained up to 48% (43%), Shannon entropy up to 48% (37%) and Tsallis entropy up to 49% (35%) of the variations in MMSE scores, all at left temporal (T7) electrode site. The steepest slope of the linear regression was found for auto-mutual information (∆y/∆x = 36). For Shannon and Tsallis entropy, slopes were less steep. Comparing to traditional slowing measures, complexity measures yielded higher coefficients of determination. We con- clude that auto-mutual information is well suited to character- ize disease severity in mild to moderate AD. I. INTRODUCTION Various quantitative electroencephalogram (QEEG) measures have been shown to differentiate between normal controls, patients with mild cognitive impairment and pa- tients with Alzheimer's disease (AD) on group level. Most Research supported by Austrian Research Promotion Agency FFG, pro- ject no. 827462, including financial contributions from Dr. Grossegger and Drbal GmbH, Vienna, Austria. H. Garn is with AIT Austrian Institute of Technology GmbH, A-1220 Vienna, Austria (phone: +43505504103; fax: +43505504125; e-mail: heinrich.garn@ait.ac.at) M. Waser is with AIT Austrian Institute of Technology GmbH, A-1220 Vienna, Austria (e-mail: markus.waser@ait.ac.at) M. Deistler is with Vienna University of Technology, A-1040 Vienna, Austria (e-mail: manfred.deistler@tuwien.ac.at). T. Benke is with Innsbruck Medical University, A-6020 Innsbruck, Austria (e-mail: thomas.benke@i-med.ac.at) P. Dal-Bianco is with Vienna Medical University, A-1090 Vienna, Aus- tria (e-mail: peter.dal-bianco@meduniwien.ac.at) G. Ransmayr is with Linz General Hospital, A-4020 Linz, Austria (e-mail: gerhard.ransmayr@akh.linz.at) H. Schmidt is with Graz Medical University, A-8036 Graz, Austria (e-mail: helena.schmidt@medunigraz.at) G. Sanin with Innsbruck Medical University, A-6020 Innsbruck, Austria (e-mail: guenther.sanin@i-med.ac.at) P. Santer is with Vienna Medical University, A-1090 Vienna, Austria (e- mail: peter.santer@meduniwien.ac.at) G. Caravias is with Linz General Hospital, A-4020 Linz, Austria (e-mail: georg.caravias@akh.linz.at) S. Seiler is with Graz Medical University, A-8036 Graz, Austria (e-mail: stephan.seiler@medunigraz.at) D. Grossegger is with Dr. Grossegger & Drbal GmbH, A-1190 Vienna, Austria (e-mail: office@alphatrace.at) W. Fruehwirt is with Dr. Grossegger & Drbal GmbH, A-1190 Vienna, Austria (e-mail: office@alphatrace.at) R. Schmidt is with Graz Medical University, A-8036 Graz, Austria (e-mail: reinhold.schmidt@medunigraz.at) studies focused on band power and synchrony. Characteristic findings were the frequency slowing (increased relative delta and relative theta power, decreased relative alpha and relative beta power) and altered coherence. Electrophysiological interactions between neuronal popu- lations are highly nonlinear. Therefore, nonlinear methods for EEG analysis seem particularly suitable. As we are interested in the analysis of information processing in the brain, it seems sensible to use measures from information theory. EEG signal complexity has been characterized by various measures such as correlation dimension, omega-complexity, first positive Lyapunov exponent L1, entropy and auto- mutual information. The application of nonlinear dynamical EEG measures to EEGs of AD patients has been discussed in the reviews by Jeong [1] and Takahashi [2]. The study by Wan [3] revealed significant differences in information en- tropy, mutual entropy and approximate entropy between 103 AD patients and 124 healthy controls. Staudinger & Polikar [4] investigated spectral entropy of 79 AD patients and found significant differences to healthy individuals. Further studies used approximate entropy, mutual entropy, multi-scale entro- py, mutual information analysis, auto-mutual information and Tsallis entropy [5-10]. Reduced signal complexities were consistently reported for AD patients as compared to normal controls. However, with 17 AD patients at the most, sample sizes were rather small. The purpose of our study was to investigate whether complexity markers – in addition to the mere differentiation between AD patients and controls - can also be used to ex- plain disease severity as measured by neuropsychological tests such as, the Mini-Mental State Examination (MMSE). To our knowledge, only one study has investigated such a correlation: Yang et al. [11] found a correlation between mul- ti-scale entropy and MMSE scores of up to r = 0.44 in 108 AD patients. In our examination, we focus on entropy as a measure of uncertainty in random variables and auto-mutual information, which is a measure of the mutual dependence of two epochs. II. METHODS A. Subjects of the Prospective Dementia Database Austria Subjects were 79 probable AD cases from the multi- centric Prospective Dementia Database Austria (PRODEM) [12] at the Medical Universities of Graz, Innsbruck and Vi- enna and Linz General Hospital, Austria. PRODEM is an ongoing longitudinal multi-center cohort study of patients Electroencephalographic Complexity Markers Explain Neuropsychological Test Scores in Alzheimer´s Disease Heinrich Garn – IEEE Senior Member, Markus Waser, Manfred Deistler - IEEE Fellow, Thomas Benke, Peter Dal-Bianco, Gerhard Ransmayr, Helena Schmidt, Guenter Sanin, Peter Santer, Georg Caravias, Stephan Seiler, Dieter Grossegger, Wolfgang Fruehwirt, and Reinhold Schmidt 978-1-4799-2131-7/14/$31.00 ©2014 IEEE 496