Contents lists available at ScienceDirect International Journal of Psychophysiology journal homepage: www.elsevier.com/locate/ijpsycho Classification of mild cognitive impairment EEG using combined recurrence and cross recurrence quantification analysis Leena T. Timothy a , Bindu M. Krishna b,⁎ , Usha Nair a a School of Engineering, Cochin University of Science and Technology, Cochin 682022, Kerala, India b Sophisticated Test and Instrumentation Centre, Cochin University of Science and Technology, Cochin 682022, Kerala, India ARTICLE INFO Keywords: Alzheimer's disease Mild cognitive impairment EEG Recurrence quantification analysis Cross recurrence quantification analysis ABSTRACT The present study is aimed at the classification of mild cognitive impairment (MCI) EEG by combining com- plexity and synchronization features based on quantifiers from the common platform of recurrence based analysis. Recurrence rate (RR) of recurrence quantification analysis (RQA) is used for complexity analysis and RR of cross recurrence quantification analysis (CRQA) is used for synchronization analysis. The investigations are carried out on EEG from two states (i) resting eyes closed (EC) and (ii) short term memory task (STM).The results of our analysis show lower levels of complexity and higher levels of inter and intra hemispheric syn- chronisation in the MCI EEG compared to that of normal controls (NC) as indicated by the statistically significant higher value of RQA RR and CRQA RR. The results also evidence the effectiveness of memory activation task by bringing out the characteristic features of MCI EEG in task specific regions of temporal, parietal and frontal lobes under the STM condition.A new approach of combining complexity and synchronization features for EEG classification of MCI subjects is proposed, based on the geometrical signal separation in a feature space formed by RQA and CRQA RR values. The results of linear classification analysis of MCI and NC EEG also reveals the effectiveness of task state analysis by the enhanced classification efficiency under the cognitive load of STM condition compared to that of EC condition. 1. Introduction Dementia caused by Alzheimer's disease (AD) is one of the most common cognitive disorders in geriatric population. Mild Cognitive Impairment (MCI), which is generally considered as an early stage of AD, is defined as a condition with memory deficits greater than normal elderly, but do not fulfil the criteria for clinically probable AD (Petersen et al., 1999). MCI is considered as a challenging condition as it is characterised by only memory impairment, leaving functions involving daily activities unaffected (Petersen et al., 2001; Petersen et al., 2009). MCI subjects are at increased risk of developing AD with a conversion rate of 12% per year (Petersen, 2004). Hence, preclinical discrimination of MCI from normal sub- jects has great significance in current research scenario and de- serves much attention. EEG signals are the representation of the complex electrical activity of the brain and hence they have the potential of providing useful in- formation about the various dynamical features of the underlying cor- tical process.Conventional linear analyses of EEG signals have identi- fied characteristic features of different brain states as well as various pathological conditions like seizures, psychiatric disorders,Alzheimer's and Parkinson's disease and toxic states (Rice et al., 1990; Bennys et al., 2001). However, considering the complex interconnections and inter- actions of the underlying neuronal networks and the identification of nonlinear nature of EEG signals, nonlinear analysis is found to provide important supplementary information in most of these cases (Park et al., 2007; Faust and Bairy, 2012; Stam, 2005; Jelles et al., 2008;Kannathal et al., 2005). In the case of AD, the established methods of spectral analysis have identified characteristic features of decreased mean frequency and co- herence (Jeong, 2004; Dauwels et al., 2010a). Taking into account the nonlinear and nonstationary nature of the EEG signals, dynamical sys- tems theory based methods are successfully applied to EEG signals for effective characterisation of AD and MCI condition. Various entropy measures like Renyi's entropy, Shannon spectral entropy, Approximate entropy, Transfer entropy, Tsalli's entropy, Lempel Ziv's complexity have indicated lower level of complexity of the EEG in MCI and AD patients compared to age matched subjects (Dauwels et al., 2011; Faust and Bairy, 2012; Abasolo et al., 2005; McBride et al., 2015; Sneddon et al., 2004; Labate et al., 2013). Entropy measures characterise the http://dx.doi.org/10.1016/j.ijpsycho.2017.07.006 Received 16 February 2017; Received in revised form 10 June 2017; Accepted 11 July 2017 ⁎ Corresponding author. E-mail address: bindum@cusat.ac.in (B.M. Krishna). International Journal of Psychophysiology 120 (2017) 86–95 Available online 13 July 2017 0167-8760/ © 2017 Elsevier B.V. All rights reserved. MARK