fMRI pattern classification in antisocial
adolescents with psychopathic traits
Laura Rachman
Master Brain and Cognitive Sciences, Cognitive Neuroscience
Research Project 1: VUmc, Child and Adolescent Psychiatry 9 February 2012 – 31 July 2012
Supervisor: Drs. M.D. Cohn
Co-‐assessor: Prof. dr. K.R. Ridderinkhof
Abstract
Within juvenile antisocial populations, it has been argued that those youths with psychopathic traits are at higher risk to develop persistent antisocial behavior during adulthood than antisocial adolescents without psychopathic traits. By means of fMRI pattern classification, using a Support Vector Machine (SVM) algorithm, we tried to distinguish this clinically relevant subgroup based on their neural activation patterns during the Monetary Incentive Delay (MID) task. Based on YPI total scores and scores for the callous-‐unemotional, impulsive-‐ irresponsible and grandiose-‐manipulative sub-‐dimensions, we trained classifiers to distinguish high and low scoring groups during four conditions of interest: (1) reward anticipation, (2) loss anticipation, (3) reward outcome, and (4) loss outcome. Statistically significant classification accuracies were obtained for the CU dimension during loss anticipation (70.8%) and loss outcome (70.8%) and for the II dimension during reward outcome (66.7%). These findings indicate that youths with high levels of CU and II traits show differentiated neural activation patterns during the anticipation and processing of positive and negative monetary incentives. Further development of MRI pattern classification may aid psychopathy diagnosis in the future by providing an objective classification method.
Introduction
Psychopathy is a personality disorder that is characterized by a set of interpersonal, affective and behavioral features (Dolan, 2004; Hare & Neumann, 2005). Previous research has revealed three factors underlying the psychopathy construct: an arrogant and deceitful interpersonal style, deficient affective experience, and an impulsive and irresponsible behavioral style (Cooke & Michie, 2001). While psychopathic personality traits in adults are partly captured by criteria for antisocial personality disorder, antisocial minors can be diagnosed with either oppositional defiant disorder (ODD), which is characterized by a defiant, hostile and incompliant attitude towards authority figures, or conduct disorder (CD), which is characterized by norm-‐violating, aggressive and deceitful behavior (DSM-‐IV; American Psychiatric Association, 1994). These two disorders together are commonly referred to as disruptive behavior disorders (DBD). The presence of psychopathic traits in minors has been much debated, but recent literature suggests that psychopathic traits may be present in childhood (Frick, 2009). Although their stability across a 9 year period from early adolescence to early adulthood has been shown to be only moderately stable (r=0.31; Lynam et al., 2009), psychopathic traits do seem to mark those youths at higher risk for developing persistent antisocial behavior during adulthood (Forsman et al., 2010; Loeber et al., 2009; Salekin & Frick, 2005).
of antisocial youths since this dimension shows a relatively low amount of overlap with disruptive behavior disorders as defined by the DSM-‐IV (Frick et al., 2000). Furthermore, antisocial adolescents with callous-‐unemotional traits show more aggressive behavior than antisocial adolescents without these traits. In addition, while youths without callous-‐unemotional traits mainly show reactive aggression, adolescents with callous-‐unemotional traits show both reactive and instrumental aggression (Frick & Viding, 2009).
One influential etiological theory posits that psychopathic individuals are less responsive to punishment, but are hypersensitive to reward (Lykken, 1995). Indeed, various behavioral studies have demonstrated reduced punishment sensitivity (e.g. Blair et al., 2004; Blair, Morton, Leonard, & Blair, 2006). On the other hand, while reward dominance has been found in children with psychopathic traits (e.g. Barry et al., 2000; O’Brien & Frick, 1996), it is not clear whether this is due to impairments in the processing of punishment or to impaired reward processing. Furthermore, it is possible that these impairments differ for specific sub-‐dimensions of the psychopathic syndrome. For example, the only study specifically showing impairment in neural responses during reward anticipation (Buckholtz et al., 2010) assessed the impulsive-‐antisocial dimension of psychopathy, rather than the more specific callous-‐unemotional dimension (e.g. Frick et al., 2005; Marini & Stickle, 2010).
While most clinical neuroimaging studies until now have relied on conventional single-‐ voxel fMRI analysis methods, it is only recent that multi-‐voxel pattern analysis (MVPA) or pattern classification methods have been used to classify patient groups more reliably. This approach differs from conventional analysis methods because it takes patterns of neural activation into account rather than averaged activation in a region of interest, leading to increased sensitivity to detect structural or functional differences. Moreover, pattern classifiers can make inferences on an individual level, whereas activation-‐based analyses can only be used to compare group averages (Modinos et al., 2012).
The potential value of MRI pattern classification methods has not only become evident in classification of patients with depression (Fu et al., 2008; Mourão-‐Miranda et al., 2011) and autism (Coutanche et al., 2011; Ecker et al., 2010), but also, recently, in classification of patients with psychopathy (Sato et al., 2011). Current classification methods of psychopathy rely on structured interviews or self-‐report measures. While these methods may be problematic given the deceitful and manipulative characteristics that are part of the construct of psychopathy, a drawback of the widely used Psychopathy Checklist is the fact that it is time-‐extensive and that diagnosis requires additional assessment of an individual’s file information. Moreover, Skeem and Cauffman (2003) have shown that two assessment tools for psychopathic traits in youths, the Psychopathy Checklist (PCL): Youth Version and the self-‐reported Youth Psychopathic traits Inventory (YPI), only partially overlap in their conceptualization of
the construct of psychopathy in minors and question the validity of these instruments in the diagnosis of juvenile psychopaths. A more objective method to classify psychopathic traits would therefore be a welcome addition to the current classification methods. This study aims to investigate the potential usefulness of neurobiological markers for the classification of psychopathic traits. Nevertheless, it should be noted that the attempt to use neurobiological markers for classification, in order to enhance its objectivity, necessarily relies on other assessment tools that form a frame of reference. While the PCL is regarded as the ‘gold standard’ for psychopathy, Skeem and Cauffman (2003) have provided evidence that the YPI shows better divergent validity, warranting its use in the current study.
Given the heterogeneous nature of the juvenile antisocial population and the clinical relevance of psychopathic traits within this group (Frick & White, 2008), the current study aimed to identify a subgroup with high levels of psychopathic traits on the basis of reward and punishment sensitivity using MRI pattern classification methods. Whereas Sato et al. (2011) focused on structural differences, this study investigated whether functional differences during a reward-‐punishment task can be used to classify individuals with high and low levels of psychopathic traits in a juvenile population.
Following indications that callous-‐unemotional traits characterize a clinically relevant group of youths with disruptive behavior disorders (Frick, 2009), we recruited a sample of adolescents previously diagnosed with childhood-‐onset DBD to participate in this study. This sample was derived from a larger cohort of adolescents that were first arrested by the police before the age of 12, yielding a group of youths at high risk to persist in their antisocial behavior. Within the group of adolescents that were diagnosed with early-‐onset DBD we hypothesized that those scoring high on psychopathic traits would show different neural activation patterns than those scoring low on psychopathic traits during a monetary reward-‐punishment task. These differences were expected to allow a pattern classifier to discriminate between the two groups. Furthermore, we hypothesized that these differences occur during reward anticipation, as well as during punishment anticipation. Additionally, we performed this classification analysis for each of the three YPI sub-‐dimensions separately. Therefore, we tested two rather exploratory hypotheses. First, we tested the hypothesis that a pattern classifier for the reward anticipation data can yield high accuracy for the impulsive-‐irresponsible sub-‐dimension, since failure to inhibit impulses could drive excessive reward-‐seeking behavior. Second, we hypothesized that classification of data from the punishment trials can yield high accuracies on the callous-‐unemotional sub-‐dimension, since a lack of emotion might drive the insensitivity for punishment (Blair et al., 2004).
Methods
Participants
Participants were recruited from a cohort of 364 individuals who had been arrested by the police before the age of 12 (van Domburg et al., 2009). For this follow-‐up study, 55 adolescents that were diagnosed with childhood onset ODD or CD with the NIMH Diagnostic Interview Schedule for Children Version IV (NIMH DISC-‐IV) (Schaffer et al., 2000) were included for further analysis (mean age ± SD = 18.3 ± 1.2; range = 15-‐20). Exclusion criteria were standard criteria for MRI research, such as the presence of metal objects in the body (e.g. a cardiac pacemaker) or pregnancy. A standard MRI checklist from the Academic Medical Center (AMC) in Amsterdam was used for this purpose. Participants gave written informed consent to participate in the study on a first occasion where researchers visited the participants’ homes to explain the study and to obtain self-‐report data from questionnaires. Data collection for this study’s purpose took place with the approval from the VU University Medical Ethics Committee.
Psychopathic traits were assessed by means of the Youth Psychopathic Traits Inventory (YPI) (Andershed et al., 2002). This inventory uses a 4-‐point response scale and addresses ten subscales, each with five items. The YPI yields a total score for each participant as well as scores for each of the three sub-‐dimensions: ‘Grandiose-‐ Manipulative’ (GM), ‘Callous-‐Unemotional’ (CU), ‘Impulsive-‐Irresponsible’ (II) (see Table 1 for this study’s internal consistencies).
By lack of a standardized cutoff score to define people with high levels of psychopathic traits based on the YPI scores, we formed two groups based on the distribution of this study’s YPI scores. People scoring above the 75th percentile were assigned to the “high
scores” group and those scoring below the 25th percentile were assigned to the “low
scores” group (Table 1). The derived cutoff scores for high levels of psychopathic traits were comparable to the mean scores of a sample of 115 conduct disordered adolescents in secure care establishments or Young Offenders Institutions (Dolan & Rennie, 2007) (Table 2).
Table 1 Cutoff scores and internal consistencies for YPI total scores and scores on each sub-‐dimension
Low scores n High scores n Cronbach’s
Alpha Total GM CU II < 83 < 24 < 27 < 29 13 12 14 13 >116 >40 >34 >41 13 12 14 13 .934 .891 .887 .812
Table 2 YPI scores of a sample of conduct disordered adolescents as measured by Dolan and Rennie (2007) Mean S.D. YPI total Grandiose-‐Manipulative Callous-‐Unemotional Impulsive-‐Irresponsible 120.4 40.8 34.1 45.4 20.8 11.1 6.5 7.7 Reward-‐punishment task
The Monetary Incentive Delay (MID) task was used to assess reward and punishment sensitivity (Knutson et al., 2001). During 72 trials subjects were presented with one out of three possible cues (circle, square or triangle), indicating whether they could win money, loose money or neither win nor loose money. The cue was presented for 2 seconds after which subjects had to fixate on a cross-‐hair (delay, 2000-‐2500 ms). After this variable interval subjects had to respond to the appearance of a white target square by pressing a button. In a reward trial, pressing in time would lead to a monetary gain of €0.50. In a punishment trial, pressing in time would lead to avoid a monetary loss. And in a neutral trial the total earnings would remain the same, whether the participant pressed the button in time or too late. The target was presented for a variable amount of time (target, 50-‐750 ms), which was continuously adapted to the participant’s performance with increments of 50 ms, such that the hit rate for each trial type was approximately 66%. Upon disappearance of the target, feedback (1650 ms) was presented to inform whether or not the participant had succeeded on that trial and to show the cumulative earnings at that point.
fMRI measurements
All images were acquired using a Philips 3T Intera magnetic resonance scanner at the Academic Medical Center in Amsterdam. 400 T2*-‐weighted echo-‐planar images (EPI) were acquired during the reward-‐punishment task using a 8-‐channel SENSE head-‐coil. Each volume consisted of 38 slices and was obtained with a TR of 2.3 s, TE of 30 ms, and a 220 x 220 x 114 mm FOV. Slices were acquired in ascending order, oriented parallel to the AC-‐PC plane, with a thickness of 3 mm and 2.29 x 2.29 in-‐plane resolution. T1-‐ weighted anatomical scans consisting of 180 slices were acquired within the same session (TR = 9.0 ms, TE = 3.5 ms, slice thickness = 1 mm; FOV = 256 x 256 x 180 mm; voxel size = 1 x 1 x 1 mm).
Preprocessing
Functional MRI data were preprocessed using SPM8 (Wellcome Department of Cognitive Neurology, London, UK). Images were realigned and slice-‐time corrected before being co-‐registered with the T1 image. Furthermore, images were normalized to Montreal Neurological Institute (MNI) space, resampled onto a 3x3x3 mm3 grid and smoothed
with an 8 mm isotropic full-‐width half-‐maximum (FWHM) Gaussian kernel. A high-‐pass filter was then applied to each voxel of the fMRI time-‐series to remove low-‐frequency noise (cutoff of 128 s). For each of the CU and II groups, one subject had to be discarded because of MRI artifacts, leaving each CU group with 13 subjects and each II group with 12 subjects.
The preprocessed time-‐series data for each individual were further analyzed by contrasting four orthogonal regressors of interest: (1) anticipation of a reward during cue presentation vs anticipation of a neutral outcome, (2) anticipation of a monetary loss during cue presentation vs anticipation of a neutral outcome, (3) reward hit vs neutral hit outcomes, and (4) outcomes on loss trials vs outcomes on neutral trials. These individual contrast images were subsequently used as input for the pattern classification analysis.
Support Vector Machine
Pattern classification is a statistical analysis technique that takes into account the differences in activity patterns of fMRI data. In comparison with more conventional analysis methods that focus on average activations within a region of interest (ROI), pattern classification methods have the potential to detect more fine-‐grained spatial information (Mur et al., 2009) and to make inferences on an individual level (Modinos et al., 2012).
The Support Vector Machine (SVM) algorithm has been shown to be applicable to pattern recognition (Boser et al., 1992). This algorithm tries to find an optimal linear decision boundary to separate the individuals of two groups. The decision boundary, further referred to as a “hyperplane”, is composed based on the voxel values that are considered as points in a high dimensional space where the number of dimensions is equal to the number of voxels. The optimal separating hyperplane is computed such that the margin between the hyperplane and the nearest voxel values of each group is maximized. This function can then later be used to classify new subjects based on which side of the hyperplane the voxel values of this new subject are located.
The Pattern Recognition for Neuroimaging Toolbox (PRoNTo;
http://www.mlnl.cs.ucl.ac.uk/pronto/) was used for pattern classification analyses. PRoNTo was running in Matlab 7.12 (The Mathworks, Inc) in Mac OS X. The multivariate patterns of all voxel values for each of the four individual contrast images were used as
input to train four different classifiers. The classifiers were then tested by means of a leave-‐one-‐out cross-‐validation method. For this method, one subject of each group was left out, leaving the rest as input to train the classifier. When a classifier was computed, the excluded subject pair was then used to test if the classifier could correctly classify these new and unknown subjects. This process was repeated until every subject had served as test data once. The performance of the classifier was defined by averaging the classification performance of each iteration, which would yield an overall accuracy as well as the sensitivity and the specificity of the classifier.
Results
YPI total scores and Grandiose-‐Manipulative sub-‐dimension
Analyses of the YPI total and the GM groups did not reveal statistically significant classifications for the anticipation and outcome of a monetary gain or a monetary loss. Those with high levels of psychopathic traits overall or GM traits did thus not show significantly different activation patterns than those with low levels of overall psychopathic traits or GM traits during these anticipation and outcome periods.
Callous-‐Unemotional sub-‐dimension
Classifiers using the loss anticipation vs neutral anticipation and feedback of loss vs neutral feedback contrast images were successful in distinguishing subjects with low levels of CU traits from subjects with high levels of CU traits. For the anticipation of loss, the classifier reached an overall accuracy of 70.8%. The rate of correct classification of low CU subjects is represented by the sensitivity, which was 75%. The specificity, reflecting the rate of high CU subjects, was 66.7%. Fig. 1 presents projections of each subject onto the weight vector and the corresponding ROC curve. Furthermore, to test the statistical significance of the classifier, we performed a random permutation test. This method randomly assigned all participants to the high or low scoring groups 1000 times to examine whether the obtained accuracies were due to chance or whether the classifier results were reliable. This resulted in a significant classification outcome in relation to chance level (p = 0.02). For the feedback of monetary loss, the SVM analysis revealed accurate classification of the two groups as well (overall accuracy 70.8%, sensitivity 58.3%, specificity 83.3%, p = 0.02).
Figure 1 A. Projection of each subject onto the calculated weight vector, with negative values (red circles) discriminating for high CU scores, and positive values (black crosses) for low CU scores during anticipation of monetary loss (p = 0.02). B. ROC curve of the same classifier (AUC = 0.75).
Impulsive-‐Irresponsible sub-‐dimension
Participants with high levels of II traits only showed significantly different activation patterns from those with low levels of II traits during the feedback of a monetary gain. The SVM analysis using the reward feedback vs neutral feedback contrast image revealed an overall classification accuracy of 66.7% (sensitivity 66.7%, specificity 66.7%, p = 0.05).
Discussion
This study investigated whether spatially distributed information in functional neuroimaging data could be used to distinguish antisocial adolescents with high levels of psychopathic traits and those with low levels of psychopathic traits. In a sample of adolescents who were arrested by the police before the age of 12 and who were previously diagnosed with childhood-‐onset DBD, we found that fMRI pattern classification could discriminate between individuals with high and low levels of callous-‐ unemotional and impulsive-‐irresponsible traits based on their neural activation patterns during the Monetary Incentive Delay task.
Previous studies have been able to detect differences in gray matter quantification (Sato et al., 2011) and functional differences in reward processing (Buckholtz et al., 2010) between psychopaths and healthy individuals. However, this study’s objective to investigate psychopathic traits within a group diagnosed with early-‐onset DBD asked for
Projection onto the weight vector False positives
Area Under Curve = 0.75
an approach that was able to detect more subtle differences in neural activations. The identification of youth with strong psychopathic tendencies within a larger population of adolescents with antisocial behavior disorders is of great relevance since it has been argued that this subgroup is at higher risk to develop persistent antisocial behavior during adulthood (Forsman et al., 2010; Loeber et al., 2009; Salekin & Frick, 2005).
By means of SVM analyses we found that people with high levels of CU traits and people with low levels of CU traits showed differentiated activation patterns during the anticipation and feedback of monetary loss. These two groups differed in such a way that a classifier could distinguish them and accurately predict to which group a new subject belonged, in line with our hypothesis. Furthermore, while we hypothesized that the II groups could be discriminated during reward anticipation, we found that the feedback of monetary rewards gave rise to differential activation patterns between subjects with low levels of II traits and those with high levels of II traits. In addition, in contrast with our hypothesis, data from the YPI total scores group did not show sufficient differentiated activation patterns to distinguish the two groups for any of the four contrast images. According to these data, high and low scoring subjects on psychopathic traits in general cannot be distinguished based on activation patterns during the MID task.
The classification results of the current study for the II sub-‐dimension suggest that people with high levels of II traits can be distinguished during reward outcome. In contrast, Buckholtz and colleagues (2010) found nucleus accumbens hyperactivation in people scoring high on the impulsive-‐antisocial factor of psychopathy during reward anticipation, but not during reward outcome. These two studies stress the importance of two different aspects of reward processing in people with impulsive personality traits. Because differential activation patterns do not necessarily arise from differences in overall activity, but can also be due to differences in functional connectivity, the univariate approach of Buckholtz et al. (2010) might not have been able to detect differences during the reward outcome phase. Furthermore, the different findings might also result from the different types of samples used in both studies. Whereas the current study was focused on adolescents, Buckholtz and colleagues investigated neural activations in adults. It has been shown that adolescents react more strongly to rewards than adults since reward systems such as the nucleus accumbens receive less inhibition compared to adults, due to the late maturation of the prefrontal cortex (Casey et al., 2008). Therefore, considering the different samples and the focus on whole-‐brain activation patterns in the current study, while Buckholtz and colleagues (2010) only focused on ventral striatal activations, future studies are required to provide more extensive insights into the neural processes underlying the anticipation and outcome of reward in impulsive and antisocial adults and adolescents.
above chance level. On the other hand the specificity was rather high, with a large number of high CU subjects correctly classified. While using different classification methods, this study’s classification accuracies are comparable to those reported by Murrie and Cornell (2002), who used self-‐report measures. They found that APSD Staff Rating and the MACI Psychopathy Content Scale measures predicted 85% of high-‐ psychopathy youths, as assessed by the PCL:YV, with an overall classification accuracy of 65% (Murrie & Cornell, 2002). This shows that the use of neurobiological markers has the potential to reach classification accuracies comparable to more widely used self-‐ report measures. Furthermore, Sato and colleagues (2011) obtained high classification results using gray matter quantification methods, with overall classification accuracies of 80% and a specificity and sensitivity of 80% and 86.7% respectively. However, while the accuracies reported in this study are similar to those reported for some psychological instruments, the use of the extreme scoring groups in this study leads to high inaccuracy rates such that the current methodology is not suitable for clinical or judicial decision making. Future studies with larger sample sizes should investigate whether the classification accuracies obtained by this study can be improved. In addition, classification of functional MRI data and structural MRI data, as performed by Sato et al. (2011), may be combined in future studies to investigate whether these methods could be improved to support clinical diagnoses.
Limitations
The lack of significant classifiers for total scores on the YPI could be due to the findings that differences in reward and punishment anticipation and processing are only represented by the CU and II sub-‐dimensions of the YPI, and not by the GM sub-‐ dimension. Furthermore, it is important to emphasize the multidimensional character of psychopathy. The diagnosis of psychopathy requires high scores on each of the dimensions. The large construct of psychopathy can thus be seen as an assembly of a wide spectrum of personality traits. The two groups that were formed based on total YPI scores might therefore be too heterogeneous for accurate classification results. In addition, a limitation of this study is the small sample size of the groups used for classification analyses. In contrast to the PCL, the YPI does not have a standardized cutoff score to identify people with high levels of psychopathic traits. We therefore formed our high and low scoring groups by taking the extremes of the population, which resulted in a large reduction in the number of subjects used for data analysis.
Unfortunately, in a very late stadium of this study we recognized that three out of fourteen subjects were incorrectly assigned to both the high and low CU groups due a flawed CU scoring algorithm. As a result, classification performance for the CU sub-‐ dimension as well as for the YPI total scores has been affected by this inaccuracy. Because of the late stage of the research in which this fault was discovered, it was not feasible to reanalyze the data, but this should be done in a future study.
Conclusion
The present fMRI study has found differences in neural activation patterns in antisocial adolescents with callous-‐unemotional and impulsive-‐irresponsible psychopathic traits during reward anticipation and during loss anticipation and processing. Even with small sample sizes, classifiers performed above chance level in discriminating and classifying people with high and low levels of psychopathic traits on the CU and II sub-‐dimensions. The accuracies reported in study shows that subtle differences within high-‐risk populations can be revealed using a multivariate approach and that these differences can be used to make inferences about the levels of CU and II traits on an individual level. However, as the percentage of inaccurate classification is still rather high, further development of this approach is warranted before this method can be used in psychopathy diagnosis.
References
American Psychiatric Association (1994). Diagnostic and statistical manual of mental disorders. DSM-‐IV, 4th Edn. Washington DC: American Psychiatric Association.
Andershed, H., Hodgins, S., & Tengström, A. (2002). Psychopathic traits in non-‐referred youths: Initial test of a new assessment tool. In E. Blaauw & L. Sheridan (Eds.),
Psychopaths: Current international perspectives: 131-‐158. The Hague, the Netherlands:
Elsevier.
Barry, C.T., Frick, P.J., DeShazo, T.M., McCoy, M.G., Ellis, M., & Loney, B.R. (2000). The importance of callous-‐unemotional traits for extending the concept of psychopathy to children. Journal of Abnormal Psychology, 109: 335-‐340.
Blair, K.S., Morton, J., Leonard, A., & Blair, R.J.R. (2006). Impaired decision-‐making on the basis of both reward and punishment information in individuals with psychopathy.
Personality and Individual Differences, 41: 155-‐165.
Blair, R.J.R., Mitchel, D.G.V., Leonard, A., Budhani, S., Peschardt, K.S., & Newman, C. (2004). Passive avoidance learning in individuals with psychopathy: modulation by reward but not by punishment. Personality and Individual Differences, 37: 1179-‐1192.
Boser, B., Guyon, I., & Vapnik, V. (1992). A training algorithm for optimal margin classifiers. Fifth Annual Workshop on Computational Learning Theory. ACM Press, Baltimore: 144-‐152.
Buckholtz, J.W., Treadway, M.T., Cowan, R.L., Woodward, N.D., Benning, S.D., Li, R. et al. (2011). Mesolimbic dopamine reward system hypersensitivity in individuals with psychopathic traits. Nature Neuroscience, 13(4): 419-‐421.
Casey, B.J., Getz, S., & Galvan, A. (2008). The adolescent brain. Developmental Review, 28: 62-‐77.
Cooke, D.J. & Michie, C. (2001). Refining the construct of psychopathy: Towards a hierarchical model. Psychological Assessment, 13: 171-‐188.
Coutanche, M.N., Thompson-‐Schill, S.L., & Schultz, R.T. (2011). Multi-‐voxel pattern analysis of fMRI data predicts clinical symptom severity. Neuroimage, 57: 113-‐123.
Dolan, M. (2004). Psychopathic personality in young people. Advances in Psychiatric
Treatment, 10: 466-‐473.
Dolan M.C. & Rennie C.E. (2007). The relationship between psychopathic traits measured by the Youth Psychopathic trait Inventory and psychopathology in a UK sample of conduct disordered boys. Journal of Adolescence, 30: 601-‐611.
Ecker, C., Rocha-‐Rego, V., Johnston, P., Mourão-‐Miranda, J., Marquand, A., Daly, E.M. et al. (2010). Investigating the predictive value of whole-‐brain structural MR scancs in autism: A pattern classification approach. Neuroimage, 49: 44-‐56.
Forsman, M., Lichtenstein, P., Andershed, H., & Larsson, H. (2010). A longitudinal twin study of the direction of effects between psychopathic personality and antisocial behavior. Journal of Child Psychology and Psychiatry, 51(1), 39-‐47.
Frick, P.J. (2009). Extending the construct of psychopathy to youth: Implications for understanding, diagnosing, and treating antisocial children and adolescents. The
Canadian Journal of Psychiatry, 54: 803-‐812.
Frick, P.J., Bodin, S.D., & Barry, C.T. (2000). Psychopathic traits and conduct problems in community and clinic-‐referred samples of children: Further development of the Psychopathy Screening Device. Psychological Assessment, 12: 382-‐393.
Frick, P.J., Stickle, T.R., Dandreaux, D.M., Farrell, J.M., & Kimonis, E.R. (2005). Callous-‐ unemotional traits in predicting the severity and stability of conduct problems and delinquency. Journal of Abnormal Child Psychology, 33: 471-‐487.
Frick, P.J. & Viding, E. (2009). Antisocial behavior from a developmental psychopathology perspective. Development and Psychopathology, 21: 1111-‐1131.
Frick, P.J. & White, S.F. (2008). Research Review: The importance of callous-‐unemotional traits for developmental models of aggressive and antisocial behavior. Journal of Child
Psychology and Psychiatry, 49: 359-‐375.
Fu, C.H.Y., Mourão-‐Miranda, J., Costafreda, S.G., Khanna, A., Marquand, A.F., Williams, S.C.R., & Brammer, M.J. (2008). Pattern classification of sad facial processing: Toward the development of neurobiological markers in depression. Biological Psychiatry, 63: 656-‐ 662.
Hare, R.D. & Neumann, C.S. (2005). Structural models of psychopathy. Current Psychiatry
Reports, 7: 57-‐64.
Knutson, B., Fong, G.W., Adams, C.M., Varner, J.L., & Hommer, D. (2001). Dissociation of reward anticipation and outcome with event-‐related fMRI. Neuroreport, 12(17): 3683-‐ 3687.
Loeber, R., Burke, J., & Pardini, D.A. (2009). Perspectives on oppositional defiant disorder, conduct disorder, and psychopathic features. Journal of Child Psychology and
Psychiatry, 50: 133-‐142.
Lykken, D.T. (1995). The antisocial personalities. Hillsdale, NJ: Lawrence Erlbaum Associates.
Marini, V.A. & Stickle, T.R. (2010). Evidence for deficits in reward responsivity in antisocial youth with callous-‐unemotional traits. Personality Disorders: Theory, Research,
and Treatment, 1: 218-‐229.
Modinos, G., Pettersson-‐Yeo, W., Allen, P., McGuire, P.K., Aleman, A., & Mechelli, A. (2012). Multivariate pattern classification reveals differential brain activation during emotional processing in individuals with psychosis proneness. Neuroimage, 59: 3033-‐ 3041.
Mourão-‐Miranda, J., Hardoon, D.R., Hahn, T., Marquand, A.F., Williams, S.C.R., Shawe-‐ Taylor, J., & Brammer, M. (2011). Patient classification as an outlier detection problem: An application of the one-‐class support vector machine. Neuroimage, 58: 793-‐804.
Mur, M., Bandettini, P.A., & Kriegeskorte, N. (2009). Revealing representational content with pattern-‐information fMRI – an introductory guide. Social Cognitive and Affective
Neuroscience, 4: 101-‐109.
Murrie, D.C. & Cornell, D.G. (2002). Psychopathy screening of incarcerated juveniles: A comparison of measures. Psychological Assessment, 14: 390-‐396.
O’Brien, B.S. & Frick, P.J. (1996). Reward dominance: Associations with anxiety, conduct problems, and psychopathy in children. Journal of Abnormal Child Psychology, 24: 223-‐ 240.
Salekin, R.T., & Frick, P.J. (2005). Psychopathy in children and adolescents: The need for a developmental perspective. Journal of Abnormal Child Psychology, 33: 403-‐409.
Sato, J.R., de Oliveira-‐Souza, R., Thomaz, C.E., Basílio, R., Bramati, I.E., Amaro Jr., E. et al. (2011). Identification of psychopathic individuals using pattern classification of MRI images. Social Neuroscience, iFirst: 1-‐13.
Shaffer, D., Fisher, P., Lucas, C.P., Dulcan, M.K., & Schwab-‐Stone, M.E. (2000). NIMH Diagnostic Interview Schedule for Children Version IV (NIMH DISC-‐IV): Description, differences from previous versions, and reliability of some common diagnoses. Journal of
the American Academy of Child and Adolescent Psychiatry, 39(1), 28-‐38.
Skeem, J.L. & Cauffman, E. (2003). Views of the downward extension: Comparing the Youth Version of the Psychopathy Checklist with the Youth Psychopathic traits Inventory. Behavioral Sciences and the Law, 21: 737-‐770.
Van Domburgh, L., Vermeiren, R., Blokland, A.A.J., & Doreleijers, T.A.H. (2009). Delinquent development in Dutch childhood arrestees: Developmental trajectories, risk factors and co-‐morbidity with adverse outcomes during adolescence. Journal of
Abnormal Child Psychology, 37: 93-‐105.