Characterization of Fear Conditioning and Fear Extinction by Analysis of Electrodermal Activity Rose T. Faghih, ∗ Member, IEEE, Patrick A. Stokes, Marie-France Marin, Rachel G. Zsido, Sam Zorowitz, Blake L. Rosenbaum, Huijin Song, Mohammed R. Milad, Darin D. Dougherty, Emad N. Eskandar, Alik S. Widge, Emery N. Brown, Fellow, IEEE, Riccardo Barbieri, Senior Member, IEEE Abstract— Electrodermal activity (EDA) is a measure of physical arousal, which is frequently measured during psy- chophysical tasks relevant for anxiety disorders. Recently, specific protocols and procedures have been devised in order to examine the neural mechanisms of fear conditioning and extinc- tion. EDA reflects important responses associated with stimuli specifically administrated during these procedures. Although several previous studies have demonstrated the reproducibility of measures estimated from EDA, a mathematical framework associated with the stimulus-response experiments in question and, at the same time, including the underlying emotional state of the subject during fear conditioning and/or extinction experiments is not well studied. We here propose an ordinary differential equation model based on sudomotor nerve activity, and estimate the fear eliciting stimulus using a compressed sensing algorithm. Our results show that we are able to recover the underlying stimulus (visual cue or mild electrical shock). Moreover, relating the time-delay in the estimated stimulation to the visual cue during extinction period shows that fear level decreases as visual cues are presented without shock, suggesting that this feature might be used to estimate the fear state. These findings indicate that a mathematical model based on electrodermal responses might be critical in defining a low- dimensional representation of essential cognitive features in order to describe dynamic behavioral states. I. INTRODUCTION Human fear conditioning models have brought signifi- cant new insights into the pathophysiology of psychiatric disorders [10], [9]. In particular, studies using a two-day fear conditioning and extinction procedure demonstrated that patients diagnosed with schizophrenia and posttraumatic stress disorder (PTSD) exhibit deficits in fear extinction memory recall [8], [14] and impaired extinction retention compared to trauma-exposed normal controls [13]. More recently, a study considering electrodermal activity (EDA) to measure conditioned response during fear conditioning ∗ Corresponding Author’s Email: rfaghih@mit.edu Rose T. Faghih, Patrick A. Stokes, Alik S. Widge, Emery N. Brown, and Riccardo Barbieri are with Massachusetts Institute of Technology and Massachusetts General Hospital. Marie-France Marin, Rachel G. Zsido, Sam Zorowitz, Blake L. Rosen- baum, Huijin Song, Mohammed R. Milad, Darin D. Dougherty, and Emad N. Eskandar are with Massachusetts General Hospital. PAS, SZ, DDD, ENE, ASW, ENB, RB were supported by the Defense Advanced Research Projects Agency (DARPA) under contract number W911NF-14-2-0045. RTF and ENB were supported by the 2013 Har- vard/MIT Joint Research Grants Program in Basic Neuroscience. The opinions presented are those of the authors and not of DARPA or the institutions. and fear extinction responses on day 1, and during fear extinction recall and fear renewal on day 2, found that a skin conductance response (SCR) during conditioning and extinction recall are not significantly different across time and are correlated within subjects [18]. Changes in EDA have been proposed as neurophysiologic arousal measures and as estimates of sympathetic nervous system activity [16]. When an outgoing sympathetic nervous burst occurs, resulting from temporal and spatial summation of spikes triggered by sudormotor nerve, SCR is generated. Generally, an increase of the frequency or amplitudes in SCR signal is interpreted as an increase of sympathetic nervous system activity level [11]. The sudomotor nerve stimulating the sweat glands’ activity triggering EDA is composed of separate, discrete and temporally short bursts. Importantly, SCRs recorded on skin surface are not always distinguishable due to overlapping responses. Several investigators have tried to decode SRC by overcoming the overlapping problem [12], [1], [2]. The main concept behind these approaches implies a deconvolution technique relying on the existence of a stereotyped and stable impulse response function (IRF) that can be estimated in order to reveal the underlying bursts generating the sudomotor signal. In the present work we describe a novel approach based on a model where the input is a shock event generated within a behavioral paradigm. We use the sparsity of shocks (i.e., there are a small number of stimulations compared to the experiment time) and recover the timing and amplitude of individual stimulations using compressed sensing techniques. For compressible signals, where only a small number of co- efficients are large (i.e., most coefficients are small or zero), the signal can be approximated by a sparse representation and recovered using optimization or greedy algorithms [3]. We then use a coordinate descent approach based on [5] to recover stimulus (visual cue or mild electric shock) and parameters. II. METHODS A. Experiment To test our model and estimation algorithm, we used SCR data of 8 healthy subjects. A detailed description of the experiment and characteristics of the participants are in [18]. None of the participants had medical conditions, or neurological disorders, none were using psychoactive drugs 978-1-4244-9270-1/15/$31.00 ©2015 IEEE 7814