Memory detection using fMRI - does the encoding context matter?

YNIMG-12092; No. of pages: 11; 4C: 7, 8, 9 NeuroImage xxx (2015) xxx–xxx

Contents lists available at ScienceDirect

NeuroImage journal homepage: www.elsevier.com/locate/ynimg

Memory detection using fMRI — Does the encoding context matter?

2Q2

Judith Peth a,⁎, Tobias Sommer a, Martin N. Hebart a, Gerhard Vossel b, Christian Büchel a, Matthias Gamer a

3 4

a

5

a r t i c l e

6 7 8

Article history: Accepted 18 March 2015 Available online xxxx

9 10 11 12 13 14 15 16

Keywords: Concealed Information Test Functional MRI Electrodermal responses Encoding context Memory Mock crime Multivariate pattern analysis

a b s t r a c t

R O

i n f o

O

Department of Systems Neuroscience, University Medical Center Hamburg-Eppendorf, Hamburg, Germany Department of Psychology, Interdisciplinary Research Group Forensic Psychology, Johannes Gutenberg-University Mainz, Mainz, Germany

E

D

P

Recent research revealed that the presentation of crime related details during the Concealed Information Test (CIT) reliably activates a network of bilateral inferior frontal, right medial frontal and right temporal–parietal brain regions. However, the ecological validity of these findings as well as the influence of the encoding context are still unclear. To tackle these questions, three different groups of subjects participated in the current study. Two groups of guilty subjects encoded critical details either only by planning (guilty intention group) or by really enacting (guilty action group) a complex, realistic mock crime. In addition, a group of informed innocent subjects encoded half of the relevant details in a neutral context. Univariate analyses showed robust activation differences between known relevant compared to neutral details in the previously identified ventral frontal–parietal network with no differences between experimental groups. Moreover, validity estimates for average changes in neural activity were similar between groups when focusing on the known details and did not differ substantially from the validity of electrodermal recordings. Additional multivariate analyses provided evidence for differential patterns of activity in the ventral fronto-parietal network between the guilty action and the informed innocent group and yielded higher validity coefficients for the detection of crime related knowledge when relying on whole brain data. Together, these findings demonstrate that an fMRI-based CIT enables the accurate detection of concealed crime related memories, largely independent of encoding context. On the one hand, this indicates that even persons who planned a (mock) crime could be validly identified as having specific crime related knowledge. On the other hand, innocents with such knowledge have a high risk of failing the test, at least when considering univariate changes of neural activation. © 2015 Elsevier Inc. All rights reserved.

E

C

T

b

F

1Q1

R

39 37 36

R

38

Introduction

41 42

The usage of brain imaging techniques to identify lies has been frequently investigated over the last decade (e.g., Kozel et al., 2005, 2009; Langleben et al., 2002) and critically discussed in the scientific community (e.g., Greely and Illes, 2007). Although initial studies aimed at detecting a specific signature of deceit in the brain activity, it has been recently emphasized that deception rather seems to unspecifically recruit a number of brain regions related to more general processes, especially memory (Farah et al., 2014). A technique that specifically focuses on the detection of crime related memories is the Concealed Information Test (CIT; Lykken, 1959, 1974). This test consists of multiple-choice questions that ask for details of a crime under investigation. For each question, the correct answer (e.g., the jewelry that was stolen) is presented together with different neutral answer alternatives (e.g., other potentially stolen goods). The

47 48 49 50 51 52 53 54

U

45 46

N C O

40

43 44

17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35

⁎ Corresponding author at: Department of Systems Neuroscience, University Medical Center Hamburg-Eppendorf, Martinistraße 52, 20246 Hamburg. Fax: +49 40 7410 59955. E-mail address: [email protected] (J. Peth).

general idea of the CIT is that only a guilty person will recognize the correct detail among these alternatives and show stronger physiological responses to this item. An innocent without such knowledge should respond unsystematically across answer options. Previous studies confirmed that guilty subjects show larger skin conductance responses as well as heart rate deceleration and respiratory suppression to crime related details (e.g., Gamer et al., 2006). Laboratory studies reported high validity coefficients for the differentiation between guilty and innocent persons on the basis of such autonomic measures (for reviews see Ben-Shakhar and Elaad, 2003; Meijer et al., 2014). Only few neuroimaging studies investigated neural responses in a CIT design so far. They consistently reported higher activity for critical compared to neutral answer alternatives in a ventral fronto-parietal network consisting of the bilateral inferior frontal gyrus (IFG), the right middle frontal gyrus (rMFG) and the right temporoparietal junction (rTPJ) (Gamer, 2011). These regions are not exclusively involved in the concealment of knowledge but rather reflect attentional orienting (Downar et al., 2000, 2002; Kiehl et al., 2001), response monitoring and inhibition (Aron et al., 2004; Wager et al., 2005) as well as memory processes (D'Esposito et al., 2000; Iidaka et al., 2006).

http://dx.doi.org/10.1016/j.neuroimage.2015.03.051 1053-8119/© 2015 Elsevier Inc. All rights reserved.

Please cite this article as: Peth, J., et al., Memory detection using fMRI — Does the encoding context matter?, NeuroImage (2015), http:// dx.doi.org/10.1016/j.neuroimage.2015.03.051

55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74

98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140

Participants

146

This study was approved by the local ethics committee and conducted according to the principles expressed in the Declaration of Helsinki. All participants gave written informed consent and were paid 50 Euro for participation. The final sample consisted of 60 right-handed participants with 20 participants in the guilty action group (6 women, mean age of 26.2 years), 20 participants in the guilty intention group (6 women, mean age of 23.9 years) and 20 participants in the innocent group (6 women, mean age of 25.7 years). No difference in age was found between the three groups, F(1,58) b 1. Most participants were students from various faculties (70%). During data collection, eight persons had to be excluded from the study because of different reasons and were replaced by new subjects. Reasons for exclusion were technical difficulties (n = 3), an incomplete fulfillment of the mock crime (n = 3) or alcohol intoxication (n = 2).

147 148

O

F

145

R O

96 97

Materials and methods

Design

P

94 95

141 142 143 144

149 150 151 152 153 154 155 156 157 158 159 160 161

A realistic mock crime procedure was either only planned (guilty intention group) or really enacted (guilty action group) by subjects in the guilty groups. Persons who belonged to the guilty intention group knew only relevant details that belonged to the planning phase, whereas the guilty action group knew all relevant details. In addition, a third group of persons fulfilled a non-criminal task before they were examined with the same CIT (innocent group). These subjects knew half of the relevant details from a neutral context, and found out during the CIT that these details were also part of a mock crime. Each group consisted of 20 subjects and participants were randomly assigned to their respective experimental condition by means of a predefined list.

162 163

Procedure The experiment was conducted in three different stages. Stage 1 was a planning phase that lasted for one week (study days 1–7). During this time period, participants received several emails from a fictitious confederate that asked them to either prepare a mock crime (guilty subjects) or an errand (innocents). The mock crime involved the theft of money and a CD with important study information and participants were informed about the precise place and timing of the theft during the planning phase. The innocent participants prepared to meet a person to get a CD for somebody else and received similar information on the details of this errand. To ensure active preparation of “criminal” and “neutral” activities, respectively, participants had to answer emails from the confederate and find out certain details about the upcoming event by themselves (e.g. by looking up specific details on the website of the University Medical Center Hamburg-Eppendorf where the event was about to take place). Stage 2 was the enactment phase and took place after the preparation week on study day seven. Here, the guilty action group committed the previously prepared mock crime, and the innocent participants fulfilled the errand. Participants of the guilty intention group, however, were stopped before they could enact the mock crime and were immediately investigated with the CIT (for a detailed description of stage 1 and stage 2, see Supplementary Methods). Stage 3 consisted of the CIT examination. Participants of the guilty intention group accomplished the CIT on day seven. Participants of the guilty action group and all innocent participants arrived on day eight for the CIT investigation, one day after completion of their respective task. The CIT was conducted by an examiner unknown to

173

D

92 93

that were recorded simultaneously. Therefore, this study adds significantly to the current discussion about whether neuroimaging methods are superior to traditional polygraphy in the detection of concealed information (Gamer, 2014).

T

90 91

C

88 89

E

86 87

R

84 85

R

82 83

O

81

C

79 80

N

77 78

Importantly, the existing neuroimaging CIT experiments were based on memory for highly salient autobiographical information (Ganis et al., 2011), artificial stimuli like playing cards (e.g., Gamer et al., 2007; Langleben et al., 2002; Nose et al., 2009) or imagined crime scenarios (Cui et al., 2014). Thus, there is a lack of research investigating the neural correlates of the CIT under realistic conditions, for example following the planning and execution of a mock crime. This approach better resembles field conditions (Osugi, 2011) and represents the optimal setting for a CIT examination in the laboratory (Ben-Shakhar and Elaad, 2003). Moreover, recent studies on autonomic responses in the CIT further improved this approach by enabling only incidental encoding of the crime related information to better resemble field conditions (e.g., Gamer et al., 2010; Peth et al., 2012). Moreover, the specific encoding context might modulate the detection of concealed information. For example, previous CIT studies reported successful detection of concealed knowledge related to criminal intentions based on skin conductance responses (Meijer et al., 2010), event-related brain potentials (Meixner and Rosenfeld, 2011) and behavioral measures (Noordraven and Verschuere, 2013), but it remains unclear whether comparable effects can also be observed in fMRI data. Furthermore, it is currently unknown whether criminal intention and action result in different physiological responses in the CIT. A second line of research contrasted participants who were guilty of committing a mock crime with a sample of innocents who were informed about crime related details. Some of these studies reported differences in autonomic responding to crime details between these groups (Ben-Shakhar et al., 1999; Bradley et al., 1996; Bradley and Rettinger, 1992; Bradley and Warfield, 1984; Giesen and Rollison, 1980; Stern et al., 1981) whereas other studies failed to find a significant differentiation (Gamer, 2010; Gamer et al., 2010). So far, no fMRI study systematically examined the influence of encoding context on recognition of relevant crime details in the CIT. To close this gap, the current study aimed at investigating three groups of subjects that differed regarding the context of information encoding (guilty action group, guilty intention group, innocent group) with an fMRI-based CIT. Electrodermal responses were additionally recorded to enable a comparison to traditional polygraph measures. While the guilty action group knew all relevant details, the guilty intention and the innocent group were unaware of half of the relevant details, respectively. The guilty intention group could only know details from the planning phase, while the innocent group could gain knowledge of half of the details from the planning and the action phase, respectively, by executing a neutral task. Such design enabled the examination of group differences in responding to known critical details as well as the calculation of validity estimates for the “traditional” comparison of informed (guilty) and uninformed (innocent) subjects (cf., Ben-Shakhar and Elaad, 2003). Based on the above-mentioned literature, we hypothesized increased activation in the previously reported ventral fronto-parietal network (Gamer, 2011) when contrasting known relevant details with neutral alternatives across all groups. In addition to this network, several regions of interest were defined to further explore whether differences in encoding context affect neural activation. As previous research on memory reported enhanced activation in the supramarginal gyrus (SMG) for information encoded during actions compared to imagined actions (Russ et al., 2003), we expected comparable differences between the guilty action and the guilty intention group. Due to their involvement in the retrieval of emotional memories (for review see Dolcos et al., 2012), we furthermore expected differences in amygdala and hippocampus activation between the guilty action and the innocent group. Moreover, we conducted multivariate analyses (Bles and Haynes, 2008) to examine whether the multivariate pattern of brain activity allows for differentiating experimental groups. Finally, the current data enable a direct comparison between the classification accuracy of neural and electrodermal response measures

U

75 76

J. Peth et al. / NeuroImage xxx (2015) xxx–xxx

E

2


164 165 166 167 168 169 170 171 172

174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200


222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266

Data preprocessing and analysis

284

For all analyses the a priori significance level was set to p = .05, but marginally significant results are reported at a threshold of p b .10. Cohens d (Cohen, 1988, p.19ff) and Cohens f (Cohen, 1988, p. 273ff) are reported as effect size estimates.

285

O

F

270

R O

220 221

Presentation of CIT questions and items as well as recording of behavioral responses was accomplished with Presentation software (Neurobehavioral Systems, Albany, CA, USA). Functional MRI was performed on a 3 T MR whole body scanner (Magnetom Trio, Siemens) using a 32-channel head coil. Forty continuous axial slices (slice thickness 2 mm, 1 mm gap) were acquired in each volume using a T2*-sensitive gradient echoplanar imaging (EPI) sequence (TR: 2390 ms; TE: 25 ms; flip angle: 80°; field of view: 216 × 216 mm; voxel size: 2 × 2 × 2 mm3, GRAPPA with PAT-factor 2). Isotropic highresolution (1 × 1 × 1 mm3) structural images were acquired using a T1-weighted coronal-oriented MPRAGE sequence with 240 slices. Skin conductance was recorded using a constant voltage system (0.5 V, Biopac MP100 System, Biopac. Inc) with electrodes placed on the thenar and hypothenar eminences of the participant's left hand.

P

218 219

269

D

216 217

Data acquisition

271 272 273 274 275 276 277 278 279 280 281 282 283

286 287 288

Recognition task From the explicit memory test, the sum of correctly recognized items was calculated for each subject. The guilty action group could recognize a maximum of 24 relevant details, while the guilty intention group could only know twelve relevant details from the planning phase. The innocent group was aware of six relevant details from the planning phase and six relevant details from their errand task. The recognition rate was calculated based on all items that could be known by the respective group to enable comparison between groups. The same measure was calculated for the 42 persons who accomplished the explicit memory test again approximately six month after the CIT.

289 290

Skin conductance responses Amplitudes of skin conductance responses that began between 1 and 3 s after stimulus onset were scored as stimulus-evoked responses and measured as changes in μS if they exceeded a threshold of 0.01 μS. The amplitudes were log-transformed according to the formula provided by Venables and Christie (1980) and we calculated difference scores between the responses to crime related details and neutral alternatives for known and unknown details, respectively. Trials with missing behavioral responses (0.15% of all trials) were excluded from further SCR analyses.

301

Preprocessing of imaging data Image processing and statistical analyses were carried out using Statistical Parametric Mapping (SPM8; Wellcome Department of Imaging Neuroscience, London, UK) running under Matlab R2009b (Mathworks, Inc., Natwick, MA, USA). The first four volumes of each time series were discarded because of T1 equilibration effects. Volumes were slice time corrected, realigned and unwarped using the mean EPI image to correct for movement artifacts. Subsequently the structural T1 image was coregistered to the mean EPI image. T1 images were segmented and resulting transformation parameters were used for spatial normalization of EPI and T1 images to MNI space. Finally, functional data were smoothed with a 6 mm full-width at half maximum (FWHM) isotropic Gaussian kernel.

311 312

E

214 215

of 188.5 days (SD = 15.8), with n = 12 from the guilty intention group, 267 n = 15 from the guilty action group and n = 15 from the innocent group. 268

T

212 213

C

210 211

E

208 209

R

207

R

205 206

N C O

203 204

the subjects, who was uninformed about the respective participant's group condition. Subjects were instructed not to mention their group condition until the whole experiment was completed. Additionally, participants were told that the experiment was designed to check whether they could cope with a deception detection test and convince the examiner that they were innocent. An additional reward of 30 Euro was offered for a successful performance in the CIT to increase their motivation. After completion of a practice session with unrelated memorized details, participants were placed in the scanner. Button presses had to be given by the index and middle finger of the participants' right hand. The CIT consisted of 24 questions about the mock crime, twelve about details from the planning phase and twelve about details from the enactment phase (see Supplementary Table S1). The order of the relevant crime detail and neutral items following a question was randomly determined but remained constant across participants. Each of the 24 questions was presented once during each of three blocks. The first block was always identical for all participants, and blocks two and three included the same questions with different orders. During the CIT, each question (e.g., How much money did you steal from the locker?) was followed by five answer options (e.g., 100 Euro?, 10 Euro?, 5 Euro?, 50 Euro?, 20 Euro?). At the beginning of each trial a fixation cross was shown (1 s), followed by the stimulus (2 s) and a blank screen (5 s–7 s). The corresponding CIT question was always shown in the upper part of the screen. The single answer options for each question were either presented as written words (e.g., for the question “What is the name of your contact person?”) or photographs (e.g., for the question “What did you steal from the locker?”) in the middle of the screen. Following standard CIT procedures (Lykken, 1959), the first item always served as a buffer and was excluded from the analyses. Among the remaining four items were two neutral items, one relevant item that mentioned or depicted the correct answer, and one target item that required a different button press to ensure that participants were paying attention to the stimulus presentation. The targets were identical for all participants and always included a small red circle next to the presented word or picture. Participants were instructed to always press the left button following an item presentation. Only when noticing the small red circle on the screen, they had to press the right button. This design resembles the three item CIT approach that was frequently used in previous CIT studies (cf., Gamer, 2011). All groups showed high detection accuracy for targets (guilty action: 92.64%; guilty intention: 92.64%; innocent: 94.31%) and no significant differences between groups were found, F(1,58) b 1. Reaction times and button presses were monitored during the CIT and subjects who did not respond, responded incorrectly or extremely slow were reminded to pay attention to the task during the short breaks between blocks. However, since we did not emphasize on speeded responses, we refrained from further analyzing the response time data. After the subjects completed the CIT in the scanner, they were asked to reveal their respective group condition and to return things they kept from their task (money and/or CD). Afterwards, an explicit memory test was conducted to check whether participants could remember the relevant details correctly. During this test, all CIT questions were presented in a multiple-choice format on a laptop outside the scanner. The question was presented auditorily and afterwards the five answer options were shown simultaneously on the screen. The subjects had to choose the correct answer option. If the correct answer was unknown to the subject, which was the case for half of the questions in the guilty intention group (i.e., details from the action phase) and the innocent group (i.e., unknown details from the planning and action phase), respectively, participants were asked to guess the correct answer. To explore long-term memory for the relevant details, subjects were invited to accomplish the same explicit memory test again approximately six months after the CIT using an online internet questionnaire. From the original sample, 42 subjects completed the questionnaire with a mean difference between study participation and response to online request

U

201 202

3


291 292 293 294 295 296 297 298 299 300

302 303 304 305 306 307 308 309 310

313 314 315 316 317 318 319 320 321 322 323

347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389

F

O

R O

345 346

P

343 344

D

341 342

Multivariate imaging analyses To determine whether the multivariate pattern of brain activity allows for validly differentiating experimental groups, we used a pattern classification approach. To this end, we set up similar subject-level design matrices as for the univariate analyses (first approach as described above) using normalized but unsmoothed functional images. We then calculated contrast images for known relevant minus neutral details for all groups and for unknown relevant minus neutral details for the guilty intention and innocent groups, respectively. For classification analyses, we used The Decoding Toolbox (Hebart et al., 2015). The classifier employed was a linear support vector machine (SVM) in the implementation of LIBSVM (Chang and Lin, 2011), with a fixed cost of c = 1. We ran two sets of classification analyses, one focusing on relevant ROIs and the other using all gray matter voxels in the brain. Calculations for the ROI analyses were accomplished for a combined ROI of left IFG (x = − 44, y = 19, z = 1), right IFG (x = 39, y = 23, z = −10), and right TPJ (x = 60, y = −48, z = 39) because these regions also exhibited robust effects in the univariate analyses (see results section). The combined ROI was constructed by centering 5 mm spheres on the given coordinates and encompassed 243 voxels in total. Please note that the univariate contrast that was used to define this ROI was independent of the aim of classification (i.e., group differentiation) such that classification accuracies are not biased by this selection (Kriegeskorte et al., 2009). For the exploratory whole-brain analyses, we used the probabilistic gray matter template of SPM thresholded at 50% to examine whether other brain regions might contribute to a differentiation of experimental groups. The thresholded gray matter mask comprised a total of 125,570 voxels. The classification approach was evaluated using a leave-one-pair-out resampling procedure. Here, the classifier was trained on all but two participants (one from each experimental condition), and the classification model was applied to the two left-out cases, yielding one decision value per subject which denotes the signed distance of the classified sample from the separating hyperplane. This procedure was repeated for all possible 209 pairs of participants. To estimate the discriminability of the classifier between the respective conditions, we used the distribution of decision values to calculate the area under the ROC curve (see below). To illustrate which brain regions carried information about the classes in whole-brain classification, we also reconstructed the multivariate pattern for all pairwise classification analyses. For that purpose, we trained a linear SVM on all data. This yielded a weight vector with one weight per voxel indicating the contribution of the voxel to the classification. Since the weight vector represents a combination of signal sources and class-independent covariance, we reconstructed the multivariate pattern using the following formula (Haufe et al., 2014):

T

339 340

C

337 338

E

335 336

R

333 334

R

331 332

O

330

C

328 329

N

326 327

Univariate imaging analyses Univariate data analysis was performed using a general linear model approach with two different kinds of subject-level design matrices. First, we constructed separate subject-level design matrices for each group to enable comparison between all three experimental conditions. For subjects belonging to the guilty intention group and the innocent group, one regressor was created for each item type (relevant, neutral, buffer and target) for the known and the unknown condition, respectively. Thus these matrices included eight regressors in total. For the guilty action group only four regressors, representing the different item types, were included in the design matrix, because subjects in this group were aware of all relevant details. Events were modeled as stick functions for each stimulus onset. Question onsets as well as trials with erroneous responses were modeled separately as regressors of no interest. Similar to previous CIT studies (cf., Gamer, 2011), contrast images were defined for known relevant minus neutral details for all groups. In addition, for the guilty intention and innocent groups, the contrast images for unknown relevant minus neutral details were defined. At the group-level, a one-way ANOVA design was constructed by entering one regressor for each group representing the contrast between known relevant details and neutral details. However, it is important to note that knowledge of the guilty action group was based on 24 relevant details, while the guilty intention and innocent groups only knew half of these relevant details. Second, to enable comparison between groups (guilty action versus guilty intention; guilty action versus innocent) under identical conditions, two additional subject-level models were constructed. In each model, the relevant details of the guilty action group were split according to whether they were known to the guilty intention or innocent group, respectively. Thus for the guilty action group the amount of known details was reduced based on the knowledge of one of the other groups. This way, two different subject-level models consisting of eight regressors each (relevant, neutral, buffer and target items for known and unknown details, respectively) were constructed for the guilty action group. Contrast images were defined for the known relevant minus neutral details to reveal brain regions responding to concealed knowledge. Based on this contrast, two separate grouplevel models (two-sample t-tests) were defined to compare the guilty action group with the guilty intention or innocent group, respectively. These analyses allowed for directly contrasting groups on the basis of shared knowledge that was acquired in different contexts. Regions of interest (ROI) were defined based on the coordinates reported in a meta-analysis of fMRI studies on the CIT (Gamer, 2011). The respective ROIs were the left IFG (x = − 44, y = 19, z = 1), the right IFG (x = 39, y = 23, z = − 10), the right TPJ (x = 60, y = − 48, z = 39) and the right MFG (x = 35, y = 44, z = 23). To correct for multiple comparisons, small volume corrections were applied for each ROI using 12 mm spheres. The relative difference between known relevant details and neutral details was examined by post hoc analyses of the contrast estimates, derived from 5 mm spheres around the defined ROI centers. These values were obtained using the SPMtoolbox rfxplot (Gläscher, 2009). To explore additional regions that might differ between the respective item types, a whole brain analysis with family-wise error (FWE) corrected p-values (pFWE b .05), based on Gaussian random fields (Worsley et al., 1996), was conducted within each group for the contrast known relevant minus neutral details. For comparison between the guilty action and guilty intention group and the guilty action and innocent group, respectively, small volume corrections in the hypothesized ROIs (i.e., SMG, amygdala, hippocampus) were conducted. For each ROI a bilateral anatomical mask was generated using the WFU Pickatlas (Maldjian et al., 2003; Tzourio-Mazoyer et al., 2002). Explorative whole brain analyses with pFWE b .05 (Worsley et al., 1996) were additionally conducted for these pairwise group comparisons. All activations are reported using x, y, z coordinates in MNI standard space.

U

324 325


E

4

390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434

−1

A ¼ covðXÞ w covðSÞ

which reconstructs the pattern as a scaled product of the noise covariance matrix cov(X) and the weight vector w. The pattern vector A can be mapped back to each voxel and provides an estimate of the sources of information in the brain which were discovered by the classifier. These maps were smoothed using a 6 mm FWHM isotropic Gaussian kernel and overlaid on a mean structural image of all participant for illustration purposes.

436

Validity analyses To estimate the validity of neural and electrodermal responses in differentiating the experimental groups, receiver operating characteristics (ROC) curves were calculated. As discussed by the National Research Council (2003), the calculation of accuracy rates depends on arbitrary cut-off scores and is therefore not recommended for validity estimation. Alternatively, the area A under the ROC curve was introduced to the field of CIT research by Lieblich et al. (1970). This signal detection theory based approach estimates the separation between two distributions (e.g., “guilty” and “innocent” participants) irrespective of the decision

442


437 438 439 440 441

443 444 445 446 447 448 449 450 451


To compare the percentage of correctly remembered relevant details between the three groups in the immediate memory test, a one-way ANOVA with the between-subjects factor group (guilty action, guilty intention, innocent) was calculated. As shown in Fig. 1A, no difference in memory performance was found between the three groups,

475 476 477 478 479 480 481 482

488 489 490

C

473 474

E

471 472

R

469 470

R

467 468

The comparison of all three experimental groups using an one-way ANOVA on the difference scores between known relevant and neutral details, with the between-subjects factor group (guilty action, guilty intention, innocent), revealed a significant main effect of the intercept, F(1,57) = 6.105, p = .016, f = 0.33, and a trend for a main effect of group, F(2,57) = 2.495, p = .091, f = 0.30. All other effects were not statistically significant. The direct comparison of participants with partial knowledge using a 2 × 2 ANOVA on the difference scores between relevant and neutral details, including the between-subjects factor group (guilty intention, innocent) and the within-subjects factor knowledge (known, unknown) revealed a significant main effect of the intercept, F(1,38) = 18.040, p b .001, f = 0.51, and a main effect of knowledge, F(1,38) = 19.367, p b .001, f = 0.44. All other effects were not statistically significant. As shown in Fig. 2, only for details known by the participants an increase in skin conductance responses was observed for relevant compared to neutral details. In addition, a trend for more pronounced responses in the innocent compared to the two guilty groups was observed. If the relevant details were unknown by the participants no such effect was found.

499

493 494 495 496 497

500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517

Imaging data

518

For each group, significant differences in neural activation between relevant and neutral items were observed in the previously identified ventral frontal-parietal network, more precisely in the bilateral IFG and the right TPJ (see lower part of Table 1 and Fig. 3). Only the guilty action group showed the predicted difference in activation between item types in the right MFG. Separate whole brain analyses for the three groups additionally revealed significant activation differences for the guilty action group and the innocent group, based on the contrast for relevant minus neutral items for known details (see upper part of Table 1). For both groups,

519 520

N C O

465 466

498

U

463 464

Skin conductance

F

486 487

461 462

O

Explicit memory

459 460

R O

485

458

491 492

P

Results

456 457

F(2,57) b 1. For the delayed memory test a 2 × 3 ANOVA was calculated on the data of 42 participants who completed the delayed memory test with the between-subjects factor group (guilty action, guilty intention, innocent) and the within-subjects factor time of measurement (immediate, delayed). Fig. 1B reveals a main effect of time of measurement, F(1,39) = 7.22, p = .011, f = 0.50. No further group differences or interaction effects were observed.

D

484

454 455

T

483

criterion (i.e., cut-off point between both distributions) and is recommended for validity estimation in the broad field of deception detection (National Research Council, 2003). In the current study, binary group comparisons were calculated on the basis of the contrast estimates (for univariate fMRI data), the difference of SCR amplitudes between relevant and neutral CIT items, or on the basis of the distribution of decision values for all possible left-out pairs (for multivariate fMRI data). For the guilty intention as well as the innocent group, known and unknown details were considered separately. For each binary group comparison, ROC curves were constructed by shifting the criterion throughout the entire distribution of values and calculating hit and false positive rate for each possible criterion. These values were used to generate the ROC curve and to calculate the area A under the curve. An area A of 1 indicates perfect differentiation of the respective groups whereas an area of 0.5 represents chance classification. To estimate the statistical significance of area statistics, 95% confidence intervals were calculated parametrically for univariate fMRI and electrodermal analyses (Bamber, 1975) and an area was considered significant if 0.5 was not included in the confidence interval. For multivariate analyses, confidence intervals were determined using a MonteCarlo permutation procedure. This non-parametric procedure was necessary, because resampling techniques such as cross-validation and bootstrapping analyses violate the assumption of independent, identically distributed samples and may for that reason yield biased variance estimates (Kohavi, 1995). For each pairwise group comparison, group assignment was randomly shuffled (20 participants in each group) and the decision values for a leave-one-pair-out procedure (see above) were calculated. From these values, ROC A values were computed, and the whole procedure was repeated 1000 times to estimate the null distribution of A. For these multivariate analyses, an area statistic was considered significant if it exceeded the 95-percentile of this empirically constructed null distribution.

E

452 453

5

Fig. 1. Explicit memory performance at the immediate (Imm) measurement occasion for the whole group of 60 participants (A), and at the immediate (Imm) and delayed measurement occasion for those 42 subjects who completed the memory test approximately six months after study participation (B). All values show the percentage of correctly remembered details for those items that could be known by the respective participants (i.e., 24 for the guilty action group, 12 for the guilty intention group, 12 for the innocent group). Error bars indicate standard errors of the mean.


521 522 523 524 525 526 527 528


541 542 543

t1:1 t1:2 t1:3 t1:4

Table 1 Results for the whole brain analysis (pFWE = .05, cluster extent threshold = 20 voxel) and the small volume corrections (12 mm sphere) in previously identified ROIs (cf., Gamer, 2011). Numbers in parentheses depict the center of the respective ROIs.

563

R O

O

F

Validity of the CIT

P

To estimate the validity of the CIT in differentiating specific experimental groups, separate ROC curves were calculated based on univariate analyses for the mean activation of the right IFG, the left IFG and the right TPJ, as well as for a combined ROI including these three areas. The right MFG was excluded from this analysis as prior results revealed that activity in this region did not differ consistently between relevant and neutral details in the current study. For the separate brain regions, a valid differentiation between subjects was possible when contrasting conditions where subjects had knowledge of crime related details with conditions where such knowledge was absent (comparison feature memory, see lower part of the legend in Fig. 5). As shown in Fig. 5, all comparisons between known and unknown conditions exceeded chance level independent of the encoding context. Only the coefficient for known details in the guilty intention group and unknown details in the innocent group in the left IFG did not reach statistical significance. Validity coefficients for similar knowledge conditions (i.e., both conditions are known or unknown) never exceeded chance level. Slightly higher validity estimates were found for the average of all single ROIs (Fig. 5, top right). Electrodermal responses were also sensitive to crime-related memory and validity coefficients were largely comparable to the estimates obtained for the univariate fMRI analyses (Fig. 5). The pattern obtained for the multivariate fMRI analyses was slightly different (see bottom row of Fig. 5). For the multivariate analysis of the combined ROI, validity coefficients were larger when comparing the guilty action group to the other two groups, and significant differentiation was possible even when the guilty action and the innocent group shared similar knowledge. By contrast, in absolute numbers, the multivariate classifier was not as sensitive to crime-related memories as electrodermal and univariate fMRI responses, and it did not allow for validly differentiating known and unknown conditions within the guilty intention group and between this group and the other two. A different pattern emerged in multivariate analyses of the whole brain data (see bottom, right of Fig. 5). In this case, the classifier was again sensitive to crime related memory (with the exception of the contrast known vs. unknown details in the guilty intention group) but it did not allow for differentiating groups with shared knowledge of crime details. Whereas validity coefficients for multivariate analyses of the combined ROI were numerically similar to those of electrodermal and univariate fMRI responses, substantially higher validity coefficients were achieved for the whole brain data1. Fig. 6 elucidates the contribution of different brain regions to the classification performance of the whole brain classifier for all group comparisons based on differences in crime related memories. Across

T

539 540

t1:5

Whole brain analysis

t1:6 t1:7 t1:8 t1:9 t1:10 t1:11 t1:12 t1:13 t1:14 t1:15 t1:16 t1:17 t1:18 t1:19 t1:20 t1:21 t1:22 t1:23 t1:24 t1:25 t1:26 t1:27 t1:28 t1:29 t1:30 t1:31 t1:32 t1:33 t1:34 t1:35 t1:36 t1:37 t1:38

Guilty action left IFG

right IFG right SMA left TPJ right TPJ

x

y

z

Z

−36 −44 −52 34 42 4 −62 52 66 48

24 18 32 26 8 24 −50 −46 −40 18

−2 24 −4 −4 34 52 32 32 34 28

6.76 5.90 5.86 6.97 5.07 6.51 5.40 5.57 5.42 5.35

b0.001 b0.001 b0.001 b0.001 0.017 b0.001 0.004 0.002 0.003 0.004

52

22

−6

5.26

0.007

−8 −48 40

16 20 22

52 10 −6

6.31 5.95 5.73

b0.001 b0.001 0.001

−36 34 52 34

24 26 −46 44

−2 −4 32 18

6.76 6.97 5.57 3.55

b0.001 b0.001 b0.001 0.036

−42 48 58 –

24 22 −44 –

0 −4 22 –

5.13 5.23 3.64 –

b0.001 b0.001 0.028 –

−48 40 66 –

20 22 −40 –

10 −6 34 –

5.95 5.73 3.93 –

b0.001 b0.001 0.011 –

U

N

C

right MFG Guilty intention right IFG Innocent left SMA left IFG right IFG ROI analysis (small volume correction) Guilty action left IFG (−44 19 1) right IFG (39 23–10) right TPJ (60–48 30) right MFG (35 44 23) Guilty intention left IFG (−44 19 1) right IFG (39 23–10) right TPJ (60–48 30) right MFG (35 44 23) Innocent left IFG (−44 19 1) right IFG (39 23–10) right TPJ (60–48 30) right MFG (35 44 23)

C

537 538

p corrected

E

535 536

R

533 534

R

531 532

significant activation differences were observed in the supplemental motor area (SMA). Moreover, the guilty action group showed a significant effect in the left TPJ. As shown in Fig. 4, the extracted contrast estimates for relevant minus neutral details differed significantly between known and unknown details in the guilty intention and innocent group in most ROIs (guilty intention: right IFG: t(19) = 2.32, p = .031, d = 0.76; left IFG: t(19) = 3.25, p = .004, d = 0.85, right TPJ: t(19) = 2.39, p = .027, d = 0.73; innocent: right IFG: t(19) = 3.27, p = .004, d = 1.07; left IFG: t(19) = 2.51, p = .021, d = 0.76, right TPJ: t(19) = 2.53, p = .021, d = 0.75). In both groups no significant differences between known and unknown items were found for the right MFG. A one-way ANOVA with the between subjects factor group (guilty action, guilty intention, innocent) was calculated on the contrast estimates of known relevant minus neutral details for each ROI. This analysis revealed a

O

529 530

544 545

D

Fig. 2. Skin conductance responses for difference scores between relevant and neutral details. Significance is depicted based on a one-sample t-test for significant differences from 0 (Bonferroni-corrected significance asterisks above the bars) or on paired t-tests comparing responses between known and unknown details, respectively. * p b .05, ** p b .01, *** p b .001. Error bars indicate standard errors of the mean.

main effect of the intercept within most ROIs (right IFG: F(1,57) = 42.93, p b .001, f = 0.19; left IFG: F(1,57) = 34.28, p b .001, f = 0.24, right TPJ: F(1,57) = 20.18, p b .001, f = 0.14). For the right MFG the intercept only reached trend level, F(1,57) = 3.98, p = .051, f = 0.22. Significant differences between groups were not observed in these ROIs. Additional analyses taking into account the separate presentation blocks yielded largely stable responses in the selected ROIs across the experiment (see Supplementary results). To further explore potential differences between specific groups, the known details in the guilty action group were reduced to details known by the guilty intention or innocent group, respectively, and brain activation was compared between group pairs using a two-sample t-test. For the comparison between the guilty action and guilty intention group, neither the ROI-based analyses including the bilateral SMG, nor the exploratory whole brain analysis revealed significant results. Similarly, the comparison between the guilty action and the innocent group did not reveal significant differences in a ROI-based approach including the bilateral amygdala and the hippocampus as well as in the exploratory whole brain analysis.

E

6

t1:39

Note. IFG = Inferior frontal gyrus, SMA = Supplementary motor area, TPJ =

t1:40

Temporoparietal junction, MFG = Middle frontal gyrus.


546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562

564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607

7

F


Discussion

614 615

The current study is the first one that used a complex, realistic mock crime to examine the modulation of neural activity in a CIT by encoding context. One the one hand, we could show that it is possible to detect concealed information based on criminal intentions (i.e., knowledge of relevant details about a crime that is prepared to be executed in the near future). Thus, comparable increases in neural activation were observed in the bilateral IFG and right TPJ in the guilty intention and the guilty action group. On the other hand, overall activity changes in the very same brain regions were largely indistinguishable between the guilty action and the innocent group, indicating a high risk for innocents with crime related knowledge to be misclassified as guilty. Finally, participants with crime related knowledge could be validly differentiated from subjects without such information but validity estimates were largely comparable between univariate analyses of neural and electrodermal responses. Higher validity coefficients for the detection of

626 627 628

C

E

R

624 625

R

622 623

R O

N C O

620 621

U

618 619

T

613

616 617

crime related knowledge were obtained when conducting multivariate analyses on whole brain data. The recognition of known relevant details activated regions in a ventral fronto-parietal network in all groups. This pattern of brain activity is broadly in line with a recent meta-analysis on neuroimaging data from CIT studies (Gamer, 2011), with the exception of the right MFG that was only found to be differentially activated by relevant and neutral items in the guilty action group. In addition, recognition of crime related details resulted in significant increases in neural activity in the SMA, at least in the guilty action and the innocent group. The SMA has repeatedly been found to be involved in orienting attention towards salient stimuli (Linden et al. 1999; Downar et al., 2002). Such interpretation also fits with the observation of enhanced activity in the left TPJ for the guilty action group, since this region was previously reported to be involved in the detection of task relevant changes (e.g., Uncapher et al., 2011) and therefore also reflects automatic attentional processes that are supposed to underlie the pattern of physiological responses in the CIT (Verschuere and Ben-Shakhar, 2011). Even though some previous CIT studies (e.g., Gamer et al., 2007, 2012; Langleben et al., 2002) also reported activity increases in the SMA or left IFG following the presentation of relevant details, the meta-analysis by Gamer (2011) failed to find a significant clustering of activations in these regions across available CIT studies. This might be due to differences in study designs, and future research

P

611 612

groups, activity increases in bilateral IFG, and to a lesser extent bilateral TPJ, superior and middle frontal gyrus, predicted existence of crime related memories. The absence of such memories was associated with increased activity in widespread temporal, parietal and subcortical regions.

D

609 610

E

608

O

Fig. 3. Conjunction analysis over groups (guilty action, guilty intention, innocent), for illustration purpose only. The depicted regions show stronger activation for relevant details compared to neutral items for those items that could be known by the respective participants. Warmer colors represent higher t values.

Fig. 4. Mean contrast estimates (arbitrary unit, AU) for relevant minus neutral details in the left IFG (A), right IFG (B), right TPJ (C), and right MFG (D) for known and unknown details. Significance is depicted based on a one-sample t-test for significant differences from 0 (Bonferroni-corrected significance asterisks above the bars) or on paired t-tests comparing responses between known and unknown details, respectively. ° p b .10, * p b .05, ** p b .01, *** p b .001. Error bars indicate standard error of the mean.


629 630 631 632 633 634 635 636 637 638 639 640 Q3 Q4 641 642 643 644 645 646 647 648 Q5 Q6 649 650 651 652


P

R O

O

F

8

663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686

C

E

R

661 662

R

660 Q7

O

658 659

C

656 657

has to reveal to what degree these regions are consistently involved in the recognition of salient details during a CIT and whether activity changes are modulated by CIT properties or the cognitive and emotional state of examinees. Importantly, the current findings from univariate analyses further support the assumption that memory and not deception is the key mechanism for successful detection of information with the CIT (Gamer et al., 2012). Memory for crime related details was very good in the current study and surprisingly stable across six months with no differences between groups. On this basis, it seems that activity in the fronto-parietal network that has been consistently found in previous CIT studies (Gamer, 2011) primarily reflects the recognition of personally relevant and therefore salient details. For such saliency response, the encoding context seems to be of minor importance. At first glance, this interpretation seems to contrast with results obtained in the Differentiation of Deception (DoD) paradigm. In this laboratory paradigm, deceptive and truthful responses are compared within one examinee while controlling for other potentially confounding factors (Furedy et al., 1988). Neuroimaging studies in the DoD paradigm revealed that deception similarly involves enhanced recruitment of the bilateral IFG and right TPJ along with other brain regions (Gamer, 2011). On the one hand, this indicates that these regions are additionally sensitive to deception. On the other hand, it also seems possible that deceptive responding itself increases the salience of a current stimulus and therefore also recruits a network of brain regions implicated in attentional orienting. It would be highly interesting for future research to clarify the relative contribution of attentional and response related processes to the observed activation pattern, for example by explicitly manipulating deception and memory within the CIT. Surprisingly, direct univariate comparisons of the guilty action with the guilty intention and the innocent groups, while controlling for the available crime related knowledge, did not reveal any significant differences. This was also true for predefined ROIs where we expected to find a significant differentiation. Thus, neither the SMG that was previously

N

654 655

U

653

T

E

D

Fig. 5. Areas under the ROC curve for the left IFG, right IFG, right TPJ, a combined ROI including these brain regions based on univariate fMRI analyses and skin conductance responses (SCR). Furthermore, values are given for multivariate analyses based on a combined ROI as well as on whole brain data. Warm colors represent areas close to 1, cold colors represent areas close to 0. Precise values of the areas under the ROC curve are given, with bold black letters representing significant differences from a chance classification of 0.5. To facilitate interpretation of the depicted patterns, an illustration of different comparison features is provided that highlights the hypothesized significant group differentiations when the classification is sensitive to guilty intention, guilty action, or memory, respectively.

found to reflect an enactment effect (Russ et al., 2003), nor the amygdala and hippocampus that were associated with an emotional modulation of memory (Dolcos et al., 2012) were activated stronger in the guilty action group. Thus, in line with previous research on the detection of intentions (Meijer et al., 2010), both groups of guilty participants showed similar explicit memory performance, skin conductance responding and neural activation during the CIT. This supports the idea that the CIT can be used to investigate knowledge about future events. Although such application seems promising for preventing planned crimes such as terrorist attacks, legal and ethical concerns for real-life applications of this approach have to be seriously considered. With respect to informed innocents, our results are in line with recent studies that also failed to find significant differences between guilty participants and informed innocents in autonomic measures under conditions where crime related knowledge was deeply encoded and participants were highly motivated to pass the polygraph test (Gamer, 2010; Gamer et al., 2010). Similar conditions also apply to the current study where the guilty action and the innocent group showed comparable memory performance even after a delay of six months and all subjects were motivated for successfully passing the CIT. Taken together, univariate fMRI analyses indicate that the CIT can be primarily used to detect the presence of critical information but does not directly allow for determining the source of knowledge. A somewhat different result was obtained in the multivariate analyses that provided some evidence for a valid detection of guilty action even in comparison to innocents with crime related knowledge, at least when restricting the analyses to a combined ROI of the bilateral IFG and the right TPJ. Whereas the average activity in these regions did not allow for such classification, a multivariate analysis of the activity pattern yielded a significant validity coefficient of A = 0.79. However, in contrast to the univariate analyses, the multivariate pattern in the combined ROI did not allow for reliably decoding the presence of crime related memories within and across groups. When using whole brain data for the multivariate analyses, such memory-based classification


687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720

9

N C O

R

R

E

C

T

E

D

P

R O

O

F


U

Fig. 6. Selected axial slices of multivariate patterns (Haufe et al., 2014) overlaid on a mean structural image of all participants for comparisons of groups with and without crime related knowledge. Warm colors indicate voxels where activity increases with knowledge prediction, whereas cold colors highlight brain regions where activity increases with the prediction of the absence of knowledge. Areas under the ROC curve are shown on the right side. GA = Guilty action, GI = Guilty intention, INN = innocent. ** p b .01, *** p b .001.

721 722 723 724 725 726 727 728 729 730 731 732

was again possible but the classifier did not allow for differentiating groups with shared knowledge. Validity coefficients for whole brain data were very high indicating that brain regions outside the previously selected ROIs substantially contributed to classification accuracy. However, since the multivariate analyses were only based on data of the current study, these results have to be treated with caution until being replicated in an independent sample. To generally reduce the risk of informed innocents failing at the test, it would be advisable to either ensure that crime related knowledge was not available to innocents (Ben-Shakhar, 2012; Matsuda et al., 2012) or to preview test questions with the suspect to allow for the exclusion of questions targeting details known by innocents (Verschuere and Crombez, 2008).

Finally, it should be noted that the currently examined group of informed innocents might not be representative for most field situations since crime related knowledge was also deeply encoded during the planning and enactment of an errand in this group. The more common field situation might involve the leakage of crime related information by media. In this case, it seems possible that such knowledge is encoded more shallowly by innocents and thus, multivariate analyses could become even more valuable to differentiate contextual recollection from low confidence recognition (Rissman et al., 2010). It remains an interesting question for future research to further determine to what degree guilty suspects can be validly differentiated from informed innocents in the CIT.


733 734 735 736 737 738 739 740 741 Q8 742 743 744

768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810

The current findings demonstrate that recognition of crime related details during a CIT examination results in increased neural activation in a ventral fronto-parietal network that seems to subserve the detection of salient information among neutral details. Importantly, differences in the encoding context (criminal versus non-criminal) do not substantially influence this response. Successful differentiation of guilty examinees from uninformed participants was possible with neural and electrodermal response measures but validity coefficients were largely comparable when relying on univariate or multivariate analyses of brain activity in predefined prefrontal and temporoparietal brain regions. Higher classification accuracies for the presence of crime related memories were achieved in multivariate analyses of whole brain data.

818 819

Uncited references

830 Q9

O

F

817

R O

766 767

Conclusion

P

764 765

811 812 813 814 815 816

820 821 822 823 824 825 826 827 828 829

Hira, 2003 Seymour and Fraynt, 2009

831 832

D

762 763

be obtained for fMRI data. In this regard, it seems possible that increased emotional arousal during the mock crime leads to an enhanced involvement of the amygdala and the hippocampus during encoding (LaBar and Cabeza, 2006; Phelps, 2004) as well as during the retrieval of crime related details in the CIT (Buchanan, 2007). This hypothesis should be explicitly addressed by future research on the CIT.

833

Acknowledgments The current study was supported by grants from the European Science Foundation (09-ECRP-025) and the German Research Foundation (GA 1621/1-1). The authors are indebted to Hellena Salostowitz and Philine Böttcher for their assistance during data collection. Additionally, the authors thank Katrin Müller, Kathrin Wendt and Timo Krämer for their help with MR scanning.

834 835

Appendix A. Supplementary data

840

T

760 761

C

758 759

E

756 757

R

754 755

R

752 753

O

751

C

749 750

N

747 748

To our knowledge, this is the first study that allowed for a direct comparison of validity estimates of electrodermal and neuroimaging data in the detection of concealed information. Groups with crime related knowledge could be reliably differentiated from groups without such knowledge on the basis of both measures, and multivariate analyses provided some evidence for a significant detection of guilty action. From an applied perspective, the comparison of the guilty action group to the innocents for items where the latter group did not have crime related knowledge (uninformed innocents) seems most relevant, because this is the characteristic condition for CIT field applications (Matsuda et al., 2012). Here, validity estimates amounted to A = 0.85 (univariate analyses) and A = 0.76 (multivariate analyses) for a combined ROI of bilateral IFG and right TPJ, A = 0.98 for whole brain data (multivariate analyses) and A = 0.77 for electrodermal responses. The latter estimate is slightly lower than the average of mock crime studies reported in a recent meta-analysis on the CIT (A = 0.87, Meijer et al., 2014). This might be related to the use of a very large number of crime related details in the current study or the experimental procedure that was more optimized for fMRI purposes regarding the repetition of items and the relatively short interstimulus interval. Most interestingly, however, validity estimates were largely comparable between electrodermal responses and fMRI measures across all group comparisons at least when restricting the fMRI analyses to previously defined ROIs. Taking into account that skin conductance can be recorded much easier and cheaper than fMRI data and that electrodermal recordings can be acquired in many participants that cannot undergo fMRI testing (e.g., because of claustrophobia or metallic implants), it seems reasonable to rely on established polygraph procedures for the detection of concealed knowledge. However, under the premise that the current results can be replicated in an independent sample, the current multivariate analyses also show that these techniques seem to have potential in increasing the validity of the CIT in general as well as in improving the differentiation of guilty subjects from innocents when both groups share crime related knowledge (Gamer, 2014). Although this study provides unique insights into the processing of concealed information, some limitations should be acknowledged. First, the CIT was always administered in close temporal relationship to the planning and the execution of the mock crime or the errand, respectively. In field cases, CIT examinations can be delayed by weeks or even months and little is known about whether such delays significantly reduce CIT validity. Consistent with recent studies using shorter delays (Carmel et al., 2003; Nahari and Ben-Shakhar, 2011), memory for salient crime related details was very stable even across a period of 6 months. These results demonstrate that memory for relevant details shows some degree of persistence over time. It has to be tested, however, whether concealed information can be detected with high accuracy after such time periods as well. Moreover, a recent study indicated that the risk of failing the test for informed innocents might be reduced after some time delay (Gamer et al., 2010). It remains to be an interesting question for future research to test the stability of this effect and to examine whether this might be related to a differential change in the cortical representation of memories related to a criminal act or a neutral activity, respectively. Second, although the current study design tried to better approximate field conditions than previous fMRI studies on deception detection (Sip et al., 2008), all participants were instructed in a standardized way to prepare and enact their respective task. Such mock crime procedures are considered to be valuable for examining the detection of concealed information under relatively realistic circumstances (Ben-Shakhar and Elaad, 2003) but the transfer to real life situations is still restricted because of the artificial nature of the “criminal” activity. This is a general problem for CIT studies conducted in the laboratory and can only be solved by appropriate field studies. However, even laboratory studies might be improved with respect to the emotional context of the mock crime. We recently demonstrated that such procedure strengthens autonomic responses in the CIT (Peth et al., 2012) and it seems interesting to test whether similar results can also

U

745 746


E

10

836 837 838 839

Supplementary data to this article can be found online at http://dx. 841 doi.org/10.1016/j.neuroimage.2015.03.051. 842 References

843

Aron, A.R., Robbins, T.W., Poldrack, R.A., 2004. Inhibition and the right inferior frontal cortex. Trends Cogn. Sci. 8 (4), 170–177. http://dx.doi.org/10.1016/j.tics.2004.02.010. Bamber, D., 1975. The area above the ordinal dominance graph and the area below the receiver operating characteristic graph. J. Math. Psychol. 12, 387–415. http://dx.doi.org/ 10.1016/0022-2496(75)90001-2. Ben-Shakhar, G., 2012. Current research and potential applications of the Concealed Information Test: an overview. Front. Psychol. 3, 1–11. http://dx.doi.org/10.3389/fpsyg. 2012.00342. Ben-Shakhar, G., Elaad, E., 2003. The validity of psychophysiological detection of information with the Guilty Knowledge Test: a meta-analytic review. J. Appl. Psychol. 88, 131–151. http://dx.doi.org/10.1037/0021-9010.88.1.131. Ben-Shakhar, G., Gronau, N., Elaad, E., 1999. Leakage of relevant information to innocent examinees in the GKT: an attempt to reduce false-positive outcomes by introducing target stimuli. J. Appl. Psychol. 84, 651–660. http://dx.doi.org/10.1037/0021-9010. 84.5.651. Bles, M., Haynes, J.-D., 2008. Detecting concealed information using brain-imaging technology. Neurocase 14, 82–92. Bradley, M.T., Rettinger, J., 1992. Awareness of crime-relevant information and the guilty knowledge test. J. Appl. Psychol. 77, 55–59. http://dx.doi.org/10.1037/0021-9010.77.1.55. Bradley, M.T., Warfield, J.F., 1984. Innocence, information, and the guilty knowledge test in the detection of deception. Psychophysiology 21, 683–689. http://dx.doi.org/10. 1111/j.1469-8986.1984.tb00257.x. Bradley, M.T., MacLaren, V.V., Carle, S.B., 1996. Deception and nondeception in guilty knowledge and guilty actions polygraph tests. J. Appl. Psychol. 81, 153–160. http:// dx.doi.org/10.1037/0021-9010.81.2.153. Buchanan, T.W., 2007. Retrieval of emotional memories. Psychol. Bull. 133, 761–779. http://dx.doi.org/10.1037/0033-2909.133.5.761. Carmel, D., Dayan, E., Naveh, A., Raveh, O., Ben-Shakhar, G., 2003. Estimating the validity of the guilty knowledge test from simulated experiments: the external validity of

844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872



N C O

R

R

E

C

D

P

R O

O

F

Lieblich, I., Kugelmass, S., Ben-Shakhar, G., 1970. Efficiency of GSR detection of information as a function of stimulus set size. Psychophysiology 6, 601–608. http://dx.doi. org/10.1111/j.1469-8986.1970.tb02249.x. Lykken, D.T., 1959. The GSR in the detection of guilt. J. Appl. Psychol. 43, 385–388. http:// dx.doi.org/10.1037/h0046060. Lykken, D.T., 1974. Psychology and the lie detector industry. Am. Psychol. 29, 725–739. Maldjian, J.A., Laurienti, P.J., Burdette, J.B., Kraft, R.A., 2003. An automated method for neuroanatomic and cytoarchitectonic atlas-based interrogation of fMRI data sets. Neuroimage 19, 1233–1239. http://dx.doi.org/10.1016/S1053-8119(03)00169-1. Matsuda, I., Nittono, H., Allen, J.J.B., 2012. The current and future status of the concealed information test for field use. Front. Psychol. 3, 532. http://dx.doi.org/10.3389/fpsyg. 2012.00532. Meijer, E.H., Verschuere, B., Merckelbach, H.L.G.J., 2010. Detecting criminal intent with the Concealed Information Test. Open Criminol. J. 3, 44–47. http://dx.doi.org/10.2174/ 1874917801003020044. Meijer, E.H., Selle, N.K., Elber, L., Ben-Shakhar, G., 2014. Memory detection with the Concealed Information Test: a meta analysis of skin conductance, respiration, heart rate, and P300 data. Psychophysiology 51, 879–904. http://dx.doi.org/10.1111/psyp.12239. Meixner, J.B., Rosenfeld, J.P., 2011. A mock terrorism application of the P300-based concealed information test. Psychophysiology 48, 149–154. http://dx.doi.org/10. 1111/j.1469-8986.2010.01050.x. Nahari, G., Ben-Shakhar, G., 2011. Psychophysiological and behavioral measures for detecting concealed information: the role of memory for crime details. Psychophysiology 48, 733–875. http://dx.doi.org/10.1111/j.1469-8986.2010.01148.x. National Research Council, 2003. The Polygraph and Lie Detection. Committee to Review the Scientific Evidence on the Polygraph. The National Academies Press, Washington, DC. Noordraven, E., Verschuere, B., 2013. Predicting the sensitivity of the reaction time-based concealed information test. Appl. Cogn. Psychol. 27, 328–335. http://dx.doi.org/10. 1002/acp.2910. Nose, I., Murai, J., Taira, M., 2009. Disclosing concealed information on the basis of cortical activations. Neuroimage 44, 1380–1386. http://dx.doi.org/10.1016/j.neuroimage. 2008.11.002. Osugi, A., 2011. Daily application of the concealed information test: Japan. Memory Detection: Theory and Application of the Concealed Information Test. Cambridge University Press, Cambridge, pp. 253–275. Peth, J., Vossel, G., Gamer, M., 2012. Emotional arousal modulates the encoding of crimerelated details and corresponding physiological responses in the Concealed Information Test. Psychophysiology 49, 381–390. http://dx.doi.org/10.1111/j.1469-8986. 2011.01313.x. Phelps, E.A., 2004. Human emotion and memory: interactions of the amygdala and hippocampal complex. Curr. Opin. Neurobiol. 14, 198–202. http://dx.doi.org/10.1016/j. conb.2004.03.015. Rissman, J., Greely, H.T., Wagner, A.D., 2010. Detecting individual memories through the neural decoding of memory states and past experience. Proc. Natl. Acad. Sci. U. S. A. 107, 9849–9854. http://dx.doi.org/10.1073/pnas.1001028107. Russ, M.O., Mack, W., Grama, C.-R., Lanfermann, H., Knopf, M., 2003. Enactment effect in memory: evidence concerning the function of the supramarginal gyrus. Exp. Brain Res. 149, 497–504. http://dx.doi.org/10.1007/s00221-003-1398-4. Seymour, T.L., Fraynt, B.R., 2009. Time and encoding effects in the concealed knowledge test. Appl. Psychophysiol. Biofeedback 34, 177–187. http://dx.doi.org/10.1007/ s10484-009-9092-3. Sip, K.E., Roepstorff, A., McGregor, W., Frith, C.D., 2008. Detecting deception: the scope and limits. Trends Cogn. Sci. 12, 48–53. http://dx.doi.org/10.1016/j.tics.2007.11.008. Stern, R.M., Breen, J.P., Watanabe, T., Perry, B.S., 1981. Effect of feedback of physiological information on responses to innocent associations and guilty knowledge. J. Appl. Psychol. 66, 677–681. http://dx.doi.org/10.1037/0021-9010.66.6.677. Tzourio-Mazoyer, N., Landeau, B., Papathanassiou, D., Crivello, F., Etard, O., Delcroix, N., Mazoyer, B., Joliot, M., 2002. Automated anatomical labeling of activations in SPM using a macroscopic anatomical parcellation of the MNI MRI single-subject brain. Neuroimage 15, 273–289. http://dx.doi.org/10.1006/nimg.2001.0978. Uncapher, M.R., Hutchinson, J.B., Wagner, A.D., 2011. Dissociable effects of top-down and bottom-up attention during episodic encoding. J. Neurosci. 31, 12613–12628. http:// dx.doi.org/10.1523/JNEUROSCI. 0152-11.2011. Venables, P.H., Christie, M.J., 1980. Electrodermal activity. In: Martin, I., Venables, P.H. (Eds.), Techniques in Psychophysiology. Wiley, New York, pp. 3–67. Verschuere, B., Ben-Shakhar, G., 2011. Theory of the concealed information test. Memory detection: Theory and application of the Concealed Information Test. Cambridge University Press, Cambridge, pp. 128–150. Verschuere, B., Crombez, G., 2008. Deja Vu! The effect of previewing test items on the validity of the Concealed Information polygraph Test. Psychol. Crime Law 14, 287–297. http://dx.doi.org/10.1080/10683160701786407. Wager, T.D., Sylvester, C.-Y.C., Lacey, S.C., Nee, D.E., Franklin, M., Jonides, J., 2005. Common and unique components of response inhibition revealed by fMRI. Neuroimage 27, 323–340. http://dx.doi.org/10.1016/j.neuroimage.2005.01.054. Worsley, K.J., Marrett, S., Neelin, P., Vandal, A.C., Friston, K.J., Evans, A.C., 1996. A unified statistical approach for determining significant signals in images of cerebral activation. Hum. Brain Mapp. 4, 58–73. http://dx.doi.org/10.1002/(SICI)10970193(1996)4:1b58::AID-HBM4N3.0.CO;2-O.

E

T

mock crime studies. J. Exp. Psychol. Appl. 9, 261–269. http://dx.doi.org/10.1037/ 1076-898X.9.4.261. Chang, C.C., Lin, C.J., 2011. LIBSVM: a library for support vector machines. ACM Trans. Intell. Syst. Technol. 2, 1–27. Cohen, J., 1988. Statistical Power Analysis for the Behavioral Sciences. Erlbaum, Hillsdale, NJ. Cui, Q., Vanman, E.J., Wei, D., Yang, W., Jia, L., Zhang, Q., 2014. Detection of deception based on fMRI activation patterns underlying the production of a deceptive response and receiving feedback about the success of the deception after a mock murder crime. Soc. Cogn. Affect. Neurosci. 9, 1472–1480. http://dx.doi.org/10.1093/scan/nst134. D'Esposito, M., Postle, B.R., Rypma, B., 2000. Prefrontal cortical contributions to working memory: evidence from event-related fMRI studies. Exp. Brain Res. 133, 3–11. http://dx.doi.org/10.1007/s002210000395. Dolcos, F., Denkova, E., Dolcos, S., 2012. Neural correlates of emotional memories: a review of evidence from brain imaging studies. Psychologia 55 (2), 80–111. Downar, J., Crawley, A.P., Mikulis, D.J., Davis, K.D., 2000. A multimodal cortical network for the detection of changes in the sensory environment. Nat. Neurosci. 3, 277–283. http://dx.doi.org/10.1038/72991. Downar, J., Crawley, A.P., Mikulis, D.J., Davis, K.D., 2002. A cortical network sensitive to stimulus salience in a neutral behavioral context across multiple sensory modalities. J. Neurophysiol. 87, 615–620. http://dx.doi.org/10.1152/jn.00636.2001. Farah, M.J., Hutchinson, J.B., Phelps, E.A., Wagner, A.D., 2014. Functional MRI-based lie detection: scientific and societal challenges. Nat. Rev. Neurosci. 15, 123–131. http://dx. doi.org/10.1038/nrn3702. Furedy, J.J., Davis, C., Gurevich, M., 1988. Differentiation of deception as a psychological process: a psychophysiological approach. Psychophysiology 25, 683–688. Gamer, M., 2010. Does the Guilty Actions Test allow for differentiating guilty participants from informed innocents? A re-examination. Int. J. Psychophysiol. 76, 19–24. http:// dx.doi.org/10.1016/j.ijpsycho.2010.01.009. Gamer, M., 2011. Detecting deception using neuroimaging techniques. Memory Detection: Theory and Application of the Concealed Information Test. Cambridge University Press, Cambridge, pp. 90–113. Gamer, M., 2014. Mind reading using neuroimaging. Is this the future of deception detection? Eur. Psychol. 19, 172–183. http://dx.doi.org/10.1027/1016-9040/a000193. Gamer, M., Rill, H.-G., Vossel, G., Gödert, H.W., 2006. Psychophysiological and vocal measures in the detection of guilty knowledge. Int. J. Psychophysiol. 60, 76–87. http://dx. doi.org/10.1016/j.ijpsycho.2005.05.006. Gamer, M., Bauermann, T., Stoeter, P., Vossel, G., 2007. Covariations among fMRI, skin conductance, and behavioral data during processing of concealed information. Hum. Brain Mapp. 28, 1287–1301. http://dx.doi.org/10.1002/hbm.20343. Gamer, M., Kosiol, D., Vossel, G., 2010. Strength of memory encoding affects physiological responses in the Guilty Actions Test. Biol. Psychol. 83 (2), 101–107. http://dx.doi.org/ 10.1016/j.biopsycho.2009.11.005. Ganis, G., Rosenfeld, J.P., Meixner, J., Kievit, R.A., Schendan, H.E., 2011. Lying in the scanner: covert countermeasures disrupt deception detection by functional magnetic resonance imaging. Neuroimage 55, 312–319. http://dx.doi.org/10.1016/j.neuroimage. 2010.11.025. Giesen, M., Rollison, M.A., 1980. Guilty knowledge versus innocent associations: effects of trait anxiety and stimulus context on skin conductance. J. Res. Pers. 14, 1–11. http:// dx.doi.org/10.1016/0092-6566(80)90035-5. Gläscher, J., 2009. Visualization of group inference data in functional neuroimaging. Neuroinformatics 7, 73–82. http://dx.doi.org/10.1007/s12021-008-9042-x. Greely, H.T., Illes, J., 2007. Neuroscience-based lie detection: the urgent need for regulation. Am. J. Law Med. 33 (2–3), 377–431. Haufe, S., Meinecke, F., Görgen, K., Dähne, S., Haynes, J.-D., Blankertz, B., Bießmann, F., 2014. On the interpretation of weight vectors of linear models in multivariate neuroimaging. Neuroimage 87, 96–110. http://dx.doi.org/10.1016/j.neuroimage.2013.10.067. Hebart, M.N., Görgen, K., Haynes, J.-D., 2015. The Decoding Toolbox (TDT): a versatile software package for multivariate analyses of functional imaging data. Front. Neuroinform. 8, 88. http://dx.doi.org/10.3389/fninf.2014.00088. Hira, S., 2003. The P300-based guilty knowledge test: does it stand the test of time? Psychophysiology 40, S10–S11. Iidaka, T., Matsumoto, A., Nogawa, J., Yamamoto, Y., Sadato, N., 2006. Frontoparietal network involved in successful retrieval from episodic memory. Spatial and temporal analyses using fMRI and ERP. Cereb. Cortex 16, 1349–1360. http://dx.doi.org/10. 1093/cercor/bhl040. Kiehl, K.A., Laurens, K.R., Duty, T.L., Forster, B.B., Liddle, P.F., 2001. Neural sources involved in auditory target detection and novelty processing: an event-related fMRI study. Psychophysiology 38, 133–142. http://dx.doi.org/10.1111/1469-8986.3810133. Kohavi, R., 1995. A study of cross-validation and bootstrap for accuracy estimation and model selection. International Joint Conference on Artifical Intelligence 14, pp. 1137–1143. Kozel, F.A., Johnson, K.A., Mu, Q., Grenesko, E.L., Laken, S.J., George, M.S., 2005. Detecting deception using functional magnetic resonance imaging. Biol. Psychiatry 58, 605–613. http://dx.doi.org/10.1016/j.biopsych.2005.07.040. Kozel, F.A., Laken, S.J., Johnson, K.A., Boren, B., Mapes, K.S., Morgan, P.S., George, M.S., 2009. Replication of functional MRI detection of deception. Open Forensic Sci. J. 2, 6–11. http://dx.doi.org/10.2174/1874402800902010006. Kriegeskorte, N., Simmons, W.K., Bellgowan, P.S., Baker, C.I., 2009. Circular analysis in systems neuroscience: the dangers of double dipping. Nat. Neurosci. 12, 535–540. http:// dx.doi.org/10.1038/nn.2303. LaBar, K.S., Cabeza, R., 2006. Cognitive neuroscience of emotional memory. Nat. Rev. Neurosci. 7, 54–64. http://dx.doi.org/10.1038/nrn1825. Langleben, D.D., Schroeder, L., Maldjian, J.A., Gur, R.C., McDonald, S., Ragland, J.D., Childress, A.R., 2002. Brain activity during simulated deception: an event-related functional magnetic resonance study. Neuroimage 15, 727–732. http://dx.doi.org/ 10.1006/nimg.2001.1003.

U

873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958

11


959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036

Context Dependent Encoding Using Convolutional Dynamic Networks.

ASSIMILATION CHOICES AMONG IMMIGRANT FAMILIES: DOES SCHOOL CONTEXT MATTER?

Attitudes toward Assisted Suicide: Does Family Context Matter?

Memory consolidation in children with epilepsy: does sleep matter?

Context reinstatement and memory for intrinsic versus extrinsic context: the role of item generation at encoding or retrieval.

Goal-Directed Modulation of Neural Memory Patterns: Implications for fMRI-Based Memory Detection.

Does detection range matter for inferring social networks in a benthic shark using acoustic telemetry?

The association between education and work stress: does the policy context matter?

Does encoding matter? A novel view on the quantitative genetic trait prediction problem.

Examining the significance of urban-rural context in tobacco quitline use: does rurality matter?

The Reduction of Fear of Movement-related Pain: Does Motivational Context Matter?

Commentary on Vink et al. (2014): The polygenic basis of drug use-does context matter?

Symptotic detection of chimerism: Y does it matter?

Does the thought of death contribute to the memory benefit of encoding with a survival scenario?

Axion dark matter detection using atomic transitions.

Rats Remember Items in Context Using Episodic Memory.

Context-specific prospective-memory processing: evidence for flexible attention allocation adjustments after intention encoding.

Imagining another context during encoding offsets context-dependent forgetting.

Does Context Matter? Mastery Motivation and Therapy Engagement of Children with Cerebral Palsy.

Laparoscopic colpotomy using monopolar electrocautery: does power really matter?

The person within: memory codes for persons and traits using fMRI repetition suppression.

Event familiarity influences memory detection using the aIAT.

Identification of the neural correlates of cyclothymic temperament using a working memory task in fMRI.

Detection of inducible clindamycin resistance in group B Streptococcus: does the distance between the disks matter?