Abstract

Rumination, an internal cognitive state characterized by recursive thinking of current self-distress and past negative events, has been found to correlate with the development of depressive disorders. Here, we investigated the feasibility of using connectivity for distinguishing different emotional states induced by a novel free-streaming, subject-driven experimental paradigm. Connectivity between 78 functional regions of interest (ROIs) within 14 large-scale networks and 6 structural ROIs particularly relevant to emotional processing were used for classifying 4 mental states in 19 healthy controls. The 4 mental states comprised: An unconstrained period of mind wandering; a ruminative mental state self-induced by recalling a time of personal disappointment; a euphoric mental state self-induced by recalling what brings the subject joy; and a sequential episodic recollection of the events of the day. A support vector machine achieved accuracies ranging from 89% to 94% in classifying pairs of different mental states. We reported the most significant brain connections that best discriminated these mental states. In particular, connectivity changes involving the amygdala were found to be important for distinguishing the rumination condition from the other mental states. Our results demonstrated that connectivity-based classification of subject-driven emotional states constitutes a novel and effective approach for studying ruminative behavior.

Introduction

A growing body of studies using functional magnetic resonance imaging (fMRI) has shown significant differences in resting-state connectivity of depressed subjects compared with healthy controls (Mayberg et al. 2000; Greicius et al. 2007; Sheline et al. 2009, 2010; Cooney et al. 2010; Berman et al. 2011). The propensity to ruminate has been found to be a trait linked to depression (Nolen-Hoeksema 2000; Robinson and Alloy 2003). Identification of brain connections that give rise to negative emotional states is thus an important step toward understanding depressive disorders. Previous work has shown the feasibility of classifying subject-driven cognitive states with whole-brain connectivity (Shirer et al. 2012). Building upon this work, we explored the use of connectivity estimated from fMRI data acquired under a novel experimental paradigm for discriminating self-induced mental states pertaining to emotional processing. A secondary aim of this work was to identify the underlying brain connections that support these mental states with a focus on dysphoric rumination.

Past studies examining the classification of mental states largely fall into 2 categories. The first category uses functional connectivity to classify mental states. For instance, Shirer et al. (2012) employed this approach to classify mental states pertaining to subject-driven, free-streaming tasks, in which subjects: allowed their minds to wander; silently sang lyrics in their head; performed serial subtractions; and recalled the events of day of the scan in great detail. Richiardi et al. (2011) used a similar approach to classify whether subjects were resting quietly or watching a movie. The second category uses blood oxygen level-dependent (BOLD) contrast between experimental conditions for mental state discrimination. For examples, a study by Sitaram et al. (2011) used the BOLD contrast between emotional tasks and rest as features. Briefly, 12 subjects identified emotional memories that invoked happy and disgusted feelings before the experiment. An additional 4 subjects identified emotional memories that invoked happy, disgusted, and sad feelings. A prearranged image linked with each memory was shown to the subject in a block-design fashion. A support vector machine (SVM) with BOLD contrast above a certain threshold taken as features achieved an average accuracy of approximately 90% in distinguishing disgust from happiness. A multiclass classification of happy, sad, and disgust on 4 subjects had an average accuracy of approximately 62 ± 14%, where chance accuracy is 33%. Pairwise classification of happy and sad states was not reported for this study.

Numerous neuroimaging studies of negative emotional and ruminative mental states have been conducted using a wide variety of stimuli (Pardo et al. 1993; Mayberg et al. 1999; Damasio et al. 2000; Cooney et al. 2010; Smith et al. 2011; Freton et al. 2013; Wilson-Mendenhall et al. 2013; Eryilmaz et al. 2014; Mandell et al. 2014; Michl et al. 2014; Zamoscik et al. 2014). Two particular studies, of note, focused on experimental paradigms that were similar to the one used in the present study. In a study by Cooney et al. (2010), subjects were instructed to “Think about what people notice about your personality” in 30-s blocks alternating with a 10-s block of staring at a fixation cross. A within-group analysis showed greatest increases in activation in the right precuneus during a rumination task compared with an abstract and concrete distraction thought condition. In a study by Damasio et al. (2000) using [15O] PET neuroimaging, a protocol of self-recall of emotionally charged events was contrasted with self-recall of an emotionally neutral event to identify areas of altered brain activation associated with happiness, fear, sadness, and anger. Subjects were free to choose the memories for generating the emotion, although it was noted that the topic of the death of a loved-one was commonly used for invoking sadness. Once a subject believed they adequately self-invoked the emotion while in the PET scanner, they signaled to the investigators and a 30-s scan was acquired. Significant activation was observed in the somatosensory cortices and upper brain stem nuclei. No significant amygdala activation was found for any of the emotional states. Two other PET studies of emotional recall also failed to detect activation in the amygdala (Pardo et al. 1993; Mayberg et al. 1999). However, a task fMRI study (Michl et al. 2014) reported activation in the amygdala in a guilt compared with a shame invoking task.

In this study, we present a novel experimental paradigm for inducing emotional states.

In contrast to previous experimental protocols, the present study employed a longer duration, free-streaming, subject-driven experimental paradigm where no external cues or stimuli (e.g., images, music, video, or other sensory cues) were used to invoke a specific mood. This paradigm aimed to capture a more naturalistic experience of emotionally charged mental states, focusing in particular on a ruminative mental state. Our classification was based on functional connectivity derived from the temporal correlations of BOLD observations between regions of interest (ROIs). To our knowledge, no studies thus far have examined the feasibility of classifying subject-driven emotional states using connectivity. Nineteen subjects were scanned during 4 mental states, where they were asked to: (1) rest quietly; (2) recall their day in great detail; (3) think about things that bring them joy; and (4) think about a time they felt deep personal disappointment. Using functional connectivity as features, we were able to classify these 4 states in a pairwise fashion with accuracy in the range of 89–94% and determined the brain regions whose correlated activity best distinguish the different mental states.

Materials and Methods

Participants

Twenty-two healthy, right-handed subjects (mean age 25.5 ± 5.3 year, 8 females) volunteered for this study. All participants provided informed consent according to the Declaration of Helsinki and the experimental protocol was approved by the Institutional Review Board of Stanford University. Exclusion criteria included any MRI contraindications, current psychiatric or neurological disorders, previous traumatic brain injury, and use of psychotropic medication. Two subjects did not complete the scans due to discomfort in the scanner or equipment problems. The scans of one subject were omitted from analysis due to an incidental structural finding. Therefore, only the scans from 19 subjects (mean age 29 ± 5 year, 6 females) were used for our analyses. Prior to scanning, subjects were briefed on the 4 mental states they would be asked to self-induce in the scanner: “to let your mind wander and think of nothing in particular” (resting state), “to recall the present day in great detail beginning from the moment you awoke to the moment you laid down in the scanner” (memory state), to think about “what brings a smile to your face? What brings you joy?” (happy state), and to think about “a time you let yourself or someone else down” (rumination state). We also fully described the aims of the experiment, including the goal of classification of ruminative and happy mental states. While alone in a testing room, the subjects were given approximately 15 min before the scan to write about things that bring them joy and an episode of personal disappointment. The subjects were instructed to refrain from discussing the contents of their brainstorming session with the researchers. At the end of this brainstorming session, the participants completed a well-validated rumination response scale (RRS) questionnaire (Treynor et al. 2003). The purpose of administering the RRS questionnaire was 2-fold. We administered the RRS questionnaire to help subjects identify a set of statements associated with rumination. We also intended to divide the subjects based on the RRS scores and to examine whether the subjects can be correctly classified based on functional connectivity. However, the small sample size and limited range of scores in these healthy controls precludes such classification and these results are not described further.

We collected a high-resolution, T1 structural scan and four 8 min T2* functional scans. Subjects were instructed to keep their eyes closed during all functional scans. The resting-state scan was always performed first and the memory scan was always third. The happy and rumination scans were either second or fourth with the order being pseudorandomized across subjects. Subjects were prompted verbally before the beginning of each functional scan about which cognitive state to self-induce during the scan. No further instruction was given once the 8-min scan began. A set of five-point rating scales were administered after the scanning session to quantify the difficulty the subjects experienced performing the task during each scan.

MRI Acquisition

Data were collected at the Stanford Center for Cognitive and Neurobiological Imaging facility using a GE Healthcare Sigma MR750 3-T scanner with a 32-channel receive-only whole-brain RF coil array (Nova Medical, Inc.). For the functional T2*-weighted echo-planar scans, 36 interleaved axial slices were acquired with the following sequence parameters: repetition time = 2000 ms; echo time = 30 ms; flip angle = 77°; voxel size=3 × 3 × 3.6 mm. The field of view was 222 × 222 mm and the matrix size was 74 × 74. An IR-FSPGR BRAVO sagittal T1 high-resolution structural sequence had the following parameters: inversion time = 450 ms; flip angle = 12°. The matrix size was 256 × 256 with a 0.9 × 0.9 × 0.9 mm spatial resolution.

MRI Data Analysis

Slice Artifact Removal

Upon visual inspection of the raw fMRI data, occasional slice artifacts were observed. To automatically identify the time points at which a slice has artifacts, we first computed the average intensity of each slice across the whole time series. A time point with mean slice intensity deviating from the time series average by more than an empirically determined threshold of 2.8 times the standard deviation was marked as problematic. Interpolation based on time points immediately before and after the slice artifact occurrence was performed to replace these problematic time points. If slice artifacts also occurred in adjacent time points, interpolation was performed using the time points before and after the artifact. In the scan with the most artifacts, <2% of the data were interpolated by the artifact removal process. For the majority of the scans, <1% of the data were modified by the artifact removal procedure. We applied single subject-independent component analysis using FSL's melodic software (Smith et al. 2004) on selected scans to verify (by visual inspection) that the de-noising procedure removed the slice artifacts.

Preprocessing

T2* functional and T1 structural data were preprocessed using FSL tools (Smith et al. 2004; Jenkinson et al. 2012). A 6-mm smoothing kernel and a 150-s high-pass filter were applied to the fMRI data. Registration to the MNI standard space was carried out in 2 steps. First, each functional scan was aligned to the high-resolution T1-weighted image. Second, these co-registered images were registered to the MNI space using affine linear registration with 12 degrees of freedom (Jenkinson et al. 2002). Six movement parameters estimated by the FSL MCFLIRT program, cerebral spinal fluid confounds extracted from bilateral ROIs in the ventricles, white matter confounds extracted from bilateral ROIs in the white matter, and global signal was regressed out from the data using a custom Matlab script.

ROI Time Series Extraction

We employed a previously defined functional ROI atlas described in Shirer et al. (2012) for extracting ROI time series. This atlas is available at http://findlab.stanford.edu/functional_ROIs.html. We selected 78 of the 90 functionally defined ROIs with 12 cerebellum ROIs omitted due to incomplete cerebellum scan coverage. Previous research has suggested a strong role for the subgenual cingulate (Drevets et al. 1997; Greicius et al. 2007; Harrison et al. 2009; Keedwell et al. 2009) and amygdala in emotional processing (Armony 2013). Therefore, 2 bilateral structural subgenual cingulate ROIs [of 28 voxels with a center of mass at (4, 31,−10) and (−4, 31,−10) MNI coordinates (Craddock et al. 2009)] and 4 structural ROIs that incorporated cytoarchitectonically defined left and right basolateral and centromedial amygdala regions (Amunts et al. 2005) were additionally included, for a final count of 84 ROIs. For each subject, we generated 84 ROI time series for each mental state by averaging the voxel time series within each ROI.

Brain State Classification

The standard approach for classifying connectivity patterns is to compute the Pearson's correlation between the time series of all ROI pairs, which serve as connectivity estimates and apply a classifier, for example, SVM (Richiardi et al. 2013). However, elements of the Pearson's correlation matrix are inherently inter-related (Varoquaux et al. 2010; Ng et al. 2014), which violates the independent feature assumption implicit in most classifier learning algorithms (Tolos¸i and Lengauer 2011). To alleviate this fundamental limitation, we employed our recently proposed method (Ng et al. 2014), which uses tools from differential geometry to decouple the elements of the Pearson's correlation matrix. We refer to the resulting matrix as projected the Pearson's correlation matrix for reasons described in the Supplementary Materials, where greater details of the method are provided. This method requires the Pearson's correlation matrix to be well conditioned, which we achieved by employing the sparse Gaussian graphical model with the sparsity parameter set through cross-validation (Ng et al. 2012).

For classifying each pair of mental states, we used an SVM with the lower triangular elements of the projected Pearson's correlation matrix (i.e., connectivity estimates between all ROI pairs) as input features, and each sample labeled as +1/−1 depending on the mental state to which it belongs. The soft margin parameter in the SVM was left at its default value of 1. To assess classification accuracy, leave-one-subject-out cross-validation was performed and a 95% Wilson binomial confidence interval was computed (Wilson 1927; Brown et al. 2001) to estimate the upper and lower bounds of the accuracy level.

Discriminative Connection Identification

To identify the significantly discriminative brain connections for each pair of mental states from the classifier weights learned from the projected Pearson's correlation matrices, we used our recently proposed approach that combines permutation testing with bootstrapping (Ng et al. 2014). The intuition behind this approach is as follows. The distribution of the classifier weights is unknown. Thus, classical statistical testing techniques, for example t-test, cannot be directly applied to identify the significant connections. The standard workaround is to permute the state labels and relearn the classifier weights for each permutation to generate a null distribution. That is, we search for the classifier weights that would be obtained by chance regardless of the state to which the connectivity estimates belong to. If the original classifier weight of a connection, without label permutation, is larger than, for example, the 95th percentile of the null distribution, then this weight value is unlikely to be generated by chance. Thus, the associated connection is likely relevant for state discrimination. However, with limited samples (i.e., the number of subjects multiplied by the number of mental states we are classifying) given the large number of predictors (i.e., the number of pairwise connections), the learned classifier weights would have high variance. We thus incorporated bootstrapping to identify the more stable discriminative connections. The assumption is that more stable relevant connections would presumably be assigned similar classifier weights across bootstrap samples. To threshold the original classifier weights in identifying the significantly discriminative connections, a procedure based on maximum statistics derived from the permutation-generated null distribution was employed (Nichols and Hayasaka 2003), which implicitly accounts for multiple comparisons. Statistical significance was declared at P < 0.001 (corrected). A greater detail of our approach can be found in the Supplementary Materials and Ng et al. (2014).

Post hoc Task Difficulty Regression

To test whether differences in subject self-rated difficulty of each mental state were a confounder that drives the classification performance, we learned a regression model from training data with the self-reported task difficulty as the response variable and the connectivity estimates as predictors, and applied this model to regress out the effects of task difficulty from the training and test data. If the differences in task difficulty were driving the successful classification of the mental states, we would expect to see a large drop in accuracy and a largely non-overlapping set of discriminative edges with the task difficulty effects regressed out.

Post hoc Subgenual Cingulate Seed-Based Connectivity

Post hoc seed-based analyses were performed to compare the functional connectivity in the Rumination mental state against the Memory, Rest, and Happy mental states with the right and left structural subgenual cingulate ROIs being the seeds. FSL's FLAME (Smith et al. 2004) was used in estimating seed-based functional connectivity maps. To restrict our post hoc analysis to brain areas pertinent to emotional processing, we first performed a one-sample t-test separately on the connectivity maps of the Happy and Rumination mental states. Defining significance using cluster thresholding with a Z-statistic of 2.3 and a cluster P-value threshold of 0.05 (corrected), we then generated a mask comprising voxels that were declared as significant in both the Happy and Rumination mental states. This procedure was repeated for the pairs of Memory and Rumination and Rest and Rumination mental states to generate the masks for the pairwise comparison of states. Subsequently, we applied paired t-tests on functional connectivity estimates within the appropriate mask to contrast the Rumination mental state against the Happy, Memory, and Rest mental states. Significance was declared using the same cluster thresholding criteria but without correcting for the 6 comparisons examined (2 ROIs by 3 state comparisons).

Results

Task Validation

The exit questionnaire was designed to gather information on how much effort it took the participants to achieve the mental state for the 8 min duration of the scan. Median, mean, and standard deviation for the ratings are shown in Supplementary Table 1. Our data show that subjects found the resting quietly and recent episodic recall to be easily achieved in the scanner. Completing the happy state and rumination state tasks was more difficult for the subjects, with the rumination task rated as the most difficult to achieve the 4 mental states. However, the median score for the rumination mental state was in the middle of the five-point difficulty scale and only 2 subjects reported the rumination task as most difficult on the five-point scale, indicating that, for most subjects, the rumination state was readily achieved.

Classification Accuracy

Classification accuracies achieved for each pair of mental state is shown in Figure 1. Classification accuracies between 89% and 94% were obtained with confidence intervals well above the chance level accuracy of 50%. We note that connectivity features derived with the Riemannian approach attained significantly higher classification accuracy than using conventional Pearson's correlation as features (Supplementary Fig. 1).

Figure 1.

Leave-one-out cross-validation accuracy of classifying mental states with an SVM. Dashed line shows an expected accuracy of 50% for random chance. Legend for mental states: H: Happy; R: Rest; Ru: Rumination; M: Memory.

Figure 1.

Leave-one-out cross-validation accuracy of classifying mental states with an SVM. Dashed line shows an expected accuracy of 50% for random chance. Legend for mental states: H: Happy; R: Rest; Ru: Rumination; M: Memory.

Significant Discriminative Connections

From the classifier weights, we identified significant connections (P < 0.001, corrected) that were most relevant for the successful classification of the contrasted mental state pairs. We chose a conservative P-value threshold of 0.001 to keep the number of connections reasonable for interpretation and visualization. Since our primary focus was the identification of brain interactions that support ruminative mental states, we present, in Table 1, only the significant connections associated with contrasting the Rumination condition against the other 3 mental states. In Table 1, the naming of the ROIs follows the convention used in the ROIs available online (Shirer et al. 2012). Specifically, the network an ROI belongs to as well as the number it was assigned within that network were combined and used as the ROI name. We additionally report the anatomical location of the ROIs. Supplementary Table 2 summarizes the significant connections for classifying the remaining pairs of mental states. Figures 2–4 display the significant connections for Rumination versus Rest, Rumination versus Happy, and Rumination versus Memory overlaid onto a transparent brain for visualization. The center of each ROI is represented by a sphere and the size of the sphere is proportional to the number of connections linked to the ROI. Connections with greater strength during the Rumination state are displayed with a thicker, darker line, and vice versa.

Table 1

Discriminative connections that were significant (P < 0.001, corrected) for classification of paired mental states

  ROI 1–ROI 2 Anatomy ROI 1  ROI 2  
Vol. Coord. Vol. Coord. 
Rumination versus Rest 
 Rumination > Rest       
  BasalGanglia03–Amygdala04 L. inf. frontal gyrus–R. baso. amygdala 18 −44, 22, 23 137 28, −3, −23 
  RECN03–Amygdala04 R. sup. parietal cortex–R. baso. amygdala 1873 48, −54, 47 137 28, −3, −23 
  dDMN02–BasalGanglia02 L. lat. occ. cortex, sup. division–L. putamen 97 −48, −68, 35 828 −14, −3, 7 
  Visuospatial01–dDMN04 L. middle frontal gyrus–Bi. precuneus 338 −27, −1, 54 1555 1, −53, 28 
  Salience04–Precuneus03 R. frontal pole–L. lat. occ. cortex, sup. division 470 28, 46, 26 388 −36, −62, 46 
  Visuospatial04–Salience04 L. lat. occ. cortex, inf. div.–R. frontal pole 93 −49, −65, −6 470 28, 46, 26 
 Rest > Rumination       
  RECN03–Language01 R. sup. parietal cortex–L. inf. frontal gyrus 1873 48, −54, 47 652 −49, 25, −4 
  Precuneus01–dDMN04 Bi. cingulate gyrus, post. div.–Bi. precuneus 579 2, −28, 27 1555 1, −53, 28 
  Visuospatial01–postSalience02 L. middle frontal gyrus–L. supramarginal gyrus, post. div. 338 −27, −1, 54 1205 −57, −38, 37 
  postSalience01–Language02 L. middle frontal gyrus–L. middle temp. gyrus, ant. div. 93 −39, 35, 30 27 −52, −1, −22 
  postSalience12–Auditory03 R. insula - R. thalamus 134 40, −6, −9 26 13, −16, −2 
Rumination versus Happy 
 Rumination > Happy       
  Precuneus02–Amygdala01 Bi. precuneus- L. CM amygdala 1572 3, −72, 40 10 −22, −8, −9 
  RECN03–LECN06 R. sup. parietal cortex–L. thalamus 1873 48, −54, 47 −15, −28, 2 
 Happy > Rumination       
  Sensorimotor03–Language03 R. supp. motor cortex–L. middle temp. gyrus, post. div. 159 3, −13, 61 317 −52, −31, −6 
  vDMN09–vDMN03 R. lat. occ. cortex, sup. div.–L. para-hippocampal gyrus 752 43, −74, 32 134 −28, −37, −15 
Rumination versus Memory 
 Rumination > Memory       
  vDMN01–dDMN01 L. RSC–MPFC 462 −12, −58, 15 5257 −3, 49, 14 
  RECN03–dDMN08 R. sup. parietal cortex–L. hippocampus 1873 48, −54, 47 393 −24, −29, −13 
  vDMN01–highVisual02 L. RSC–R. occipital pole 462 −12, −58, 15 1679 32, −85, 0 
  Visuospatial03–Language01 L. inf. frontal gyrus–L. inf. frontal gyrus 1105 −47, 13, 27 652 −49, 25, −4 
  Precuneus02–LECN02 Bi. precuneus–L. frontal pole 1572 3, −72, 40 473 −40, 48, −1 
  Visuospatial04–vDMN09 L. lat. occ. cortex, inf. div.–R. lat. occ. cortex, sup. div. 93 −49, −65, −6 752 43, −74, 32 
 Memory > Rumination       
  vDMN05–vDMN01 R. RSC–L. RSC 590 13, −53, 14 462 −12, −58, 15 
  Precuneus01–dDMN04 Bi. cingulate gyrus, post. div.–Bi. precuneus 579 2, −28, 27 1555 1, −53, 28 
  RECN03–Language01 R. sup. parietal cortex–L. inf. frontal gyrus 1873 48, −54, 47 652 −49, 25, −4 
  Amygdala02–Amygdala01 R. cent. amygdala – L. cent. amygdala 26, −7, −8 10 −22, −8, −9 
  Salience03–RECN03 Bi. paracingulate gyrus–R. sup. parietal cortex 2887 0, 17, 47 1873 48, −54, 47 
  vDMN06–vDMN01 Bi. precuneus–L. RSC 1921 1, −57, 54 462 −12, −58, 15 
  vDMN09–Salience03 R. lat. occ. cortex, sup. div.–Bi. paracingulate gyrus 752 43, −74, 32 2887 0, 17, 47 
  vDMN01–RECN01 L. RSC–R. mid. frontal gyrus 462 −12, −58, 15 2093 38, 26, 42 
  vDMN07–vDMN01 R. inf. frontal gyrus–L. RSC 399 26, 26, 45 462 −12, −58, 15 
  ROI 1–ROI 2 Anatomy ROI 1  ROI 2  
Vol. Coord. Vol. Coord. 
Rumination versus Rest 
 Rumination > Rest       
  BasalGanglia03–Amygdala04 L. inf. frontal gyrus–R. baso. amygdala 18 −44, 22, 23 137 28, −3, −23 
  RECN03–Amygdala04 R. sup. parietal cortex–R. baso. amygdala 1873 48, −54, 47 137 28, −3, −23 
  dDMN02–BasalGanglia02 L. lat. occ. cortex, sup. division–L. putamen 97 −48, −68, 35 828 −14, −3, 7 
  Visuospatial01–dDMN04 L. middle frontal gyrus–Bi. precuneus 338 −27, −1, 54 1555 1, −53, 28 
  Salience04–Precuneus03 R. frontal pole–L. lat. occ. cortex, sup. division 470 28, 46, 26 388 −36, −62, 46 
  Visuospatial04–Salience04 L. lat. occ. cortex, inf. div.–R. frontal pole 93 −49, −65, −6 470 28, 46, 26 
 Rest > Rumination       
  RECN03–Language01 R. sup. parietal cortex–L. inf. frontal gyrus 1873 48, −54, 47 652 −49, 25, −4 
  Precuneus01–dDMN04 Bi. cingulate gyrus, post. div.–Bi. precuneus 579 2, −28, 27 1555 1, −53, 28 
  Visuospatial01–postSalience02 L. middle frontal gyrus–L. supramarginal gyrus, post. div. 338 −27, −1, 54 1205 −57, −38, 37 
  postSalience01–Language02 L. middle frontal gyrus–L. middle temp. gyrus, ant. div. 93 −39, 35, 30 27 −52, −1, −22 
  postSalience12–Auditory03 R. insula - R. thalamus 134 40, −6, −9 26 13, −16, −2 
Rumination versus Happy 
 Rumination > Happy       
  Precuneus02–Amygdala01 Bi. precuneus- L. CM amygdala 1572 3, −72, 40 10 −22, −8, −9 
  RECN03–LECN06 R. sup. parietal cortex–L. thalamus 1873 48, −54, 47 −15, −28, 2 
 Happy > Rumination       
  Sensorimotor03–Language03 R. supp. motor cortex–L. middle temp. gyrus, post. div. 159 3, −13, 61 317 −52, −31, −6 
  vDMN09–vDMN03 R. lat. occ. cortex, sup. div.–L. para-hippocampal gyrus 752 43, −74, 32 134 −28, −37, −15 
Rumination versus Memory 
 Rumination > Memory       
  vDMN01–dDMN01 L. RSC–MPFC 462 −12, −58, 15 5257 −3, 49, 14 
  RECN03–dDMN08 R. sup. parietal cortex–L. hippocampus 1873 48, −54, 47 393 −24, −29, −13 
  vDMN01–highVisual02 L. RSC–R. occipital pole 462 −12, −58, 15 1679 32, −85, 0 
  Visuospatial03–Language01 L. inf. frontal gyrus–L. inf. frontal gyrus 1105 −47, 13, 27 652 −49, 25, −4 
  Precuneus02–LECN02 Bi. precuneus–L. frontal pole 1572 3, −72, 40 473 −40, 48, −1 
  Visuospatial04–vDMN09 L. lat. occ. cortex, inf. div.–R. lat. occ. cortex, sup. div. 93 −49, −65, −6 752 43, −74, 32 
 Memory > Rumination       
  vDMN05–vDMN01 R. RSC–L. RSC 590 13, −53, 14 462 −12, −58, 15 
  Precuneus01–dDMN04 Bi. cingulate gyrus, post. div.–Bi. precuneus 579 2, −28, 27 1555 1, −53, 28 
  RECN03–Language01 R. sup. parietal cortex–L. inf. frontal gyrus 1873 48, −54, 47 652 −49, 25, −4 
  Amygdala02–Amygdala01 R. cent. amygdala – L. cent. amygdala 26, −7, −8 10 −22, −8, −9 
  Salience03–RECN03 Bi. paracingulate gyrus–R. sup. parietal cortex 2887 0, 17, 47 1873 48, −54, 47 
  vDMN06–vDMN01 Bi. precuneus–L. RSC 1921 1, −57, 54 462 −12, −58, 15 
  vDMN09–Salience03 R. lat. occ. cortex, sup. div.–Bi. paracingulate gyrus 752 43, −74, 32 2887 0, 17, 47 
  vDMN01–RECN01 L. RSC–R. mid. frontal gyrus 462 −12, −58, 15 2093 38, 26, 42 
  vDMN07–vDMN01 R. inf. frontal gyrus–L. RSC 399 26, 26, 45 462 −12, −58, 15 

Note: The naming of the ROIs combines the network it was derived from and the number the ROI has within that network on the FIND lab website: findlab.stanford.edu. The volume column for each ROI has units of voxels and the location of the center of mass of each ROI is in MNI space coordinates. RSC: retrosplenial cortex; MPFC: medial prefrontal cortex.

Figure 2.

Visualization of Rumination versus Rest brain connections important for discriminating between mental states from Table 1. Connections are overlaid onto a transparent brain for visualization. The center of each ROI is represented by a sphere and the size of the sphere is proportional to the number of connections linked to the ROI. Connections with greater strength during the Rumination state are displayed with a thicker, darker line, and vice versa.

Figure 2.

Visualization of Rumination versus Rest brain connections important for discriminating between mental states from Table 1. Connections are overlaid onto a transparent brain for visualization. The center of each ROI is represented by a sphere and the size of the sphere is proportional to the number of connections linked to the ROI. Connections with greater strength during the Rumination state are displayed with a thicker, darker line, and vice versa.

Figure 3.

Visualization of Rumination versus Happy brain connections important for discriminating between mental states from Table 1. Connections are overlaid onto a transparent brain for visualization. The center of each ROI is represented by a sphere and the size of the sphere is proportional to the number of connections linked to the ROI. Connections with greater strength during the Rumination state are displayed with a thicker, darker line, and vice versa.

Figure 3.

Visualization of Rumination versus Happy brain connections important for discriminating between mental states from Table 1. Connections are overlaid onto a transparent brain for visualization. The center of each ROI is represented by a sphere and the size of the sphere is proportional to the number of connections linked to the ROI. Connections with greater strength during the Rumination state are displayed with a thicker, darker line, and vice versa.

Figure 4.

Visualization of Rumination versus Memory brain connections important for discriminating between mental states from Table 1. Connections are overlaid onto a transparent brain for visualization. The center of each ROI is represented by a sphere and the size of the sphere is proportional to the number of connections linked to the ROI. Connections with greater strength during the Rumination state are displayed with a thicker, darker line, and vice versa. MPFC: medial prefrontal cortex; RSC: retrosplenial cortex.

Figure 4.

Visualization of Rumination versus Memory brain connections important for discriminating between mental states from Table 1. Connections are overlaid onto a transparent brain for visualization. The center of each ROI is represented by a sphere and the size of the sphere is proportional to the number of connections linked to the ROI. Connections with greater strength during the Rumination state are displayed with a thicker, darker line, and vice versa. MPFC: medial prefrontal cortex; RSC: retrosplenial cortex.

Rumination Versus Rest

Eleven significant discriminative connections were found for the classification of the Rumination and Rest mental states. The connections of the right frontal pole (Salience04) to the left lateral occipital cortex, superior division (Precuneus03) and to the left lateral occipital cortex, inferior division (Visuospatial04) had higher connectivity in the Rumination state. Connectivity between the left inferior frontal gyrus (Language01) and the right superior parietal cortex (RECN03) was lower in the Rumination condition. The right superior parietal cortex (RECN03) appeared in multiple significant discriminative connections. The right basolateral amygdala (Amygdala04), which is particularly relevant to emotion processing, had higher connectivity with the left inferior frontal gyrus (BasalGanglia03) and the right superior parietal cortex (RECN03) for the Rumination state compared with the Rest mental state.

Rumination Versus Happy

Four significant discriminative connections were found for the classification of Rumination and Happy mental states. Connectivity between the right superior parietal cortex (RECN03) and the left thalamus (LECN06) was higher in the Rumination mental state. The bilateral precuneus (Precuneus02) had higher connectivity with the left centromedial amygdala (Amygdala01) in the Rumination state compared with the Happy mental state.

Rumination Versus Memory

The left retrosplenial cortex (vDMN01) was found to be present in 6 of the 15 significant discriminative connections for classifying the Rumination and Memory conditions. The left retrosplenial cortex (vDMN01) showed higher connectivity with the medial prefrontal cortex (dDMN01) and right occipital pole (highVisual02), and lower connectivity with the right retrosplenial cortex (vDMN05), bilateral precuneus (vDMN06), the right middle frontal gyrus (RECN01), and the right inferior frontal gyrus (vDMN07). The right superior parietal cortex (RECN03) had higher connectivity with the left hippocampus (dDMN08). Lower connectivity was observed between the right and left centromedial amygdala (Amygdala02 and Amydala01) in the Rumination state compared with the Memory mental state.

Regression of Subject-Reported Task Difficulty

We did not find substantial differences in the accuracy of the classifier after regressing out task difficulty ratings. Two contrasts (Rest vs. Rumination and Happy vs. Rumination) showed a 5–8% decrease in accuracy. Classification accuracy results for all contrasts are given in Supplementary Table 3. The 5–8% decrease in classification accuracy values was within the lower bound of the Wilson binomial confidence interval. The majority of the significantly discriminative edges also remained the same after regression as reported in Supplementary Table 4.

Post hoc Subgenual Cingulate Connectivity

Using the left and right subgenual cingulate ROIs as seeds in a seed-based functional connectivity analysis, no significant connectivity differences were found for Rumination versus Memory and Rumination versus Rest. For Rumination versus Happy, significantly greater connectivity was observed in the right insula for the Rumination state using the right subgenual cingulate ROI as the seed (Fig. 5), but no difference was observed with the left subgenual cingulate ROI as the seed.

Figure 5.

Post hoc functional connectivity paired t-tests with left and right subgenual cingulate seeds showed increased connectivity in the right insula with the right subgenual cingulate seed in the contrast of rumination greater than happy.

Figure 5.

Post hoc functional connectivity paired t-tests with left and right subgenual cingulate seeds showed increased connectivity in the right insula with the right subgenual cingulate seed in the contrast of rumination greater than happy.

Discussion

Experimental Paradigm

Our presented experimental paradigm was very different from traditional block and event-related design. Using this self-directed, longer duration protocol enables us to capture more naturalistic mental states. Since no external stimuli were provided, we do not know at which time point the subjects were truly engaged in the task or mind-wandering during the non-resting fMRI scans. Without a clear mental state label of each time point, we cannot use the standard approach, in which each volume is taken as a sample with its intensity patterns as input features, to build classifiers. Considering all naturalistic cognitive function requires some coordination between brain areas, we opt to use connectivity over each scan as features with the assumption that the subjects were engaged in the task majority of the time.

Classification Accuracy

Our results demonstrated that it is possible to classify, with a high degree of accuracy, different subject-driven emotional states using functional connectivity estimates as features. The high accuracy (∼90%) between rest and the emotional states showed that it is unlikely our subjects were simply letting their minds wander during the self-induced emotional scans. Additionally, the high accuracy in distinguishing the memory and rumination mental states showed that recollecting a time of personal disappointment generates a distinct connectivity pattern compared with recalling the events of the day. Most impressively, whole-brain connectivity patterns were able to distinguish the Happy and Rumination states with an accuracy of 89%, whereas roughly chance level accuracy of 53% was obtained with Pearson correlation used as features. Our results thus indicate that the Riemannian method employed here is superior to the conventional approach of using Pearson correlations for connectivity-based classification.

Significant Discriminative Connections and Post hoc Analyses

The primary aim of this study was to determine whether functional connectivity can provide accurate classification of the emotional mental states. As a secondary aim, we wished to identify connections supporting a ruminative state. From this perspective, it is intriguing to note that the amygdala was involved only for state contrasts that included the Rumination condition and did not appear in any of the significant discriminative connections when classifying non-Rumination pairs of mental states (Supplementary Table 2). The role of the amygdala in emotional processing is most well known in the study of fear, but some evidence suggests that the amygdala may be involved in both positive and negative emotional processing as well as in detecting novel stimuli (Vuilleumier 2005; Armony 2013; Michl et al. 2014).

The experimental paradigm most similar to ours was the one proposed by Damasio et al. (2000). Our paradigm is distinct from that in Damasio et al. (2000) in a number of ways. First, subjects in our study were scanned for considerably longer stretches of time in the induced states (8 min vs. 30 s). Second, and more critically, the Damasio et al.'s study relied on activation differences between states, whereas our study used functional connectivity for classification. Furthermore, the Damasio study showed little increased activation in the amygdala during the self-generated emotional mental states, albeit the known involvement of the amygdala in emotional processing. In contrast, we detected significant changes in amygdala connectivity in all our state contrasts that involved the Rumination condition. In particular, we observed higher connectivity between the amygdala and other ROIs in the Rumination mental state compared with Resting and Happy mental states. Lower inter-amygdala connectivity was found in the Rumination condition compared with the Memory mental state. These convergent findings strongly suggest that the amygdala plays a critical role in rumination which may be a mental state that includes emotions such as guilt, disappointment, and sadness.

The right superior parietal cortex was also present in a number of discriminative edges. Some studies of emotional response have shown significant changes in activation of superior parietal cortex. A task-based fMRI study with pleasant, unpleasant, and emotionally neutral pictures showed activation in the right superior parietal cortex with the unpleasant pictures (Lang et al. 1998). In a study of negative reappraisal and suppression where women watched negative and neutral films, increased activation was observed in the superior parietal cortex in the watching negative films greater than watching neutral films contrast (Goldin et al. 2008). Superior parietal cortex is known to be important in working memory tasks, though the evidence for its role in episodic memory is less abundant (Koenigs et al. 2009). A recent review of the literature suggests that memory-related effects observed in the superior parietal cortex are related to downstream processing of task relevance or salience (Vilberg and Rugg 2008).

We note that the majority of the discriminative connections was found in ROIs that were between, instead of within, resting-state networks. Disrupted internetwork connectivity has been observed in several psychiatric and neurological diseases (Manoliu et al. 2013, 2014; von dem Hagen et al. 2013). Our results suggest that complex, self-driven mental states require coordination across multiple resting-state brain networks. A recent meta-analysis of co-activation patterns between networks in task versus rest support this interpretation (Di et al. 2013).

We also note that our results do not suggest that the particular areas highlighted in this work (e.g., the amygdala or superior parietal cortex) are the only brain regions responsible for a ruminative mental state nor that ruminative and happy states depend on connectivity between vastly different brain regions. A recent meta-analysis on emotional states indicates that distinct emotional states share common brain regions (Lindquist et al. 2012), and our results are by no means contradictory. Instead, our results suggest that modulation in connectivity patterns might be another mechanism for the generation of different self-driven mental states, in addition to changing activation level within similar brain regions.

Interestingly, the subgenual cingulate was not found in any of the significant connections for discriminating the emotional mental states. This observation led us to perform a post hoc functional connectivity analysis using subgenual cingulate seeds, from which we found significantly higher connectivity between the right subgenual anterior cingulate cortex and the right anterior insula during the Rumination state compared with the Happy state. This finding conforms to past studies that suggest the insula plays a role in emotional processing and the anterior insula, the subgenual anterior cingulate cortex, and the anterior midcingulate are strongly connected.

Conclusion

Our findings support the use of this more naturalistic experimental paradigm in generating subject-driven emotional states. With functional connectivity used as features, approximately 90% accuracy was achieved in contrasting the various pairs of mental states. An important question that deserves further investigation is which connections are generally involved with dysphoric rumination. Our current results could not fully answer this question since no significant discriminative connections were common across all 3 state contrasts examined, that is, Rumination versus Happy, Rumination versus Memory, and Rumination versus Resting. We foresee that increasing the sample size would provide the needed sensitivity to detect such connections. For future work, we hope to extend this approach to study patients with mood and anxiety disorders. In comparison with standard block-design and event-related experiments, we anticipate that the continuous, subject-driven paradigm used here will be a more naturalistic and compelling means of inducing ruminative states for studying these patient groups.

Supplementary material

Supplementary material can be found at: http://www.cercor.oxfordjournals.org/

Funding

This study was funded by the Dana Foundation, the National Institutes of Health (NS073498), the Natural Sciences and Engineering Research Council of Canada, and the Stanford Center for Cognitive and Neurobiological Imaging.

Notes

Leah Friedman provided helpful comments for the manuscript. We are grateful to Jonas Richiardi, Nora Leonardi, and Dimitri Van De Ville for sharing their brain visualization software used to generate the figures. Susan Nolen-Hoeksema generously shared her ruminative responses questionnaire. Bob Dougherty provided advice and example code to perform artifact correction. We thank our participants for their enthusiasm and participation. Conflict of Interest: None declared.

References

Amunts
K
Kedo
O
Kindler
M
Pieperhoff
P
Mohlberg
H
Shah
NJ
Habel
U
Schneider
F
Zilles
K
.
2005
.
Cytoarchitectonic mapping of the human amygdala, hippocampal region and entorhinal cortex: intersubject variability and probability maps
.
Anat Embryol
 .
210
:
343
352
.
Armony
JL
.
2013
.
Current emotion research in behavioral neuroscience: the role(s) of the amygdala
.
Emotion Rev
 .
5
:
104
115
.
Berman
MG
Peltier
S
Nee
DE
Kross
E
Deldin
PJ
Jonides
J
.
2011
.
Depression, rumination and the default network
.
Soc Cogn Affect Neurosci
 .
6
:
548
555
.
Brown
LD
Cai
TT
DasGupta
A
.
2001
.
Interval estimation for a binomial proportion
.
Stat Sci
 .
16
:
101
117
.
Cooney
RE
Joormann
J
Eugène
F
Dennis
EL
Gotlib
IH
.
2010
.
Neural correlates of rumination in depression
.
Cogn Affect Behav Neurosci
 .
10
:
470
478
.
Craddock
RC
Holtzheimer
PE
Hu
XP
Mayberg
HS
.
2009
.
Disease state prediction from resting state functional connectivity
.
Magn Reson Med
 .
62
:
1619
1628
.
Damasio
AR
Grabowski
TJ
Bechara
A
Damasio
H
Ponto
LLB
Parvizi
J
Hichwa
RD
.
2000
.
Subcortical and cortical brain activity during the feeling of self-generated emotions
.
Nat Neurosci
 .
3
:
1049
1056
.
Di
X
Gohel
S
Kim
EH
Biswal
BB
.
2013
.
Task vs. rest—different network configurations between the coactivation and the resting-state brain networks
.
Front Hum Neurosci
 .
7
:
1
9
.
Drevets
WC
Price
JL
Simpson
JR
Jr
Todd
RD
Reich
T
Vannier
M
Raichle
ME
.
1997
.
Subgenual prefrontal cortex abnormalities in mood disorders
.
Nature
 .
386
:
824
827
.
Eryilmaz
H
Van De Ville
D
Schwartz
S
Vuilleumier
P
.
2014
.
Lasting impact of regret and gratification on resting brain activity and its relation to depressive traits
.
J Neurosci
 .
34
:
7825
7835
.
Freton
M
Lemogne
C
Delaveau
P
Guionnet
S
Wright
E
Wiernik
E
Bertasi
E
Fossati
P
.
2013
.
The dark side of self-focus: brain activity during self-focus in low and high brooders
.
Soc Cogn Affect Neurosci
 . .
Goldin
PR
McRae
K
Ramel
W
Gross
JJ
.
2008
.
The neural bases of emotion regulation: reappraisal and suppression of negative emotion
.
BiolPsychiatry
 .
63
:
577
586
.
Greicius
MD
Flores
BH
Menon
V
Glover
GH
Solvason
HB
Kenna
H
Reiss
AL
Schatzberg
AF
.
2007
.
Resting-state functional connectivity in major depression: abnormally increased contributions from subgenual cingulate cortex and thalamus
.
Biol Psychiatry
 .
62
:
429
437
.
Harrison
NA
Brydon
L
Walker
C
Gray
MA
Steptoe
A
Critchley
HD
.
2009
.
Inflammation causes mood changes through alterations in subgenual cingulate activity and mesolimbic connectivity
.
Biol Psychiatry
 .
66
:
407
414
.
Jenkinson
M
Bannister
P
Brady
M
Smith
S
.
2002
.
Improved optimization for the robust and accurate linear registration and motion correction of brain images
.
NeuroImage
 .
17
:
825
841
.
Jenkinson
M
Beckmann
CF
Behrens
TEJ
Woolrich
MW
Smith
SM
.
2012
.
FSL
.
NeuroImage
 .
62
:
782
790
.
Keedwell
P
Drapier
D
Surguladze
S
Giampietro
V
Brammer
M
Phillips
M
.
2009
.
Neural markers of symptomatic improvement during antidepressant therapy in severe depression: subgenual cingulate and visual cortical responses to sad, but not happy, facial stimuli are correlated with changes in symptom score
.
J Psychopharmacol
 .
23
:
775
788
.
Koenigs
M
Barbey
AK
Postle
BR
Grafman
J
.
2009
.
Superior parietal cortex is critical for the manipulation of information in working memory
.
J Neurosci
 .
29
:
14980
14986
.
Lang
PJ
Bradley
MM
Fitzsimmons
JR
Cuthbert
BN
Scott
JD
Moulder
B
Nangia
V
.
1998
.
Emotional arousal and activation of the visual cortex: An fMRI analysis
.
Psychophysiology
 .
35
:
199
210
.
Lindquist
KA
Wager
TD
Kober
H
Bliss-Moreau
E
Barrett
LF
.
2012
.
The brain basis of emotion: a meta-analytic review
.
Behav Brain Sci
 .
35
:
121
143
.
Mandell
D
Siegle
GJ
Shutt
L
Feldmiller
J
Thase
ME
.
2014
.
Neural substrates of trait ruminations in depression
.
J Abnorm Psychol
 .
123
:
35
48
.
Manoliu
A
Meng
C
Brandl
F
Doll
A
Tahmasian
M
Scherr
M
Schwerthöffer
D
Zimmer
C
Förstl
H
Bäuml
J
et al
2014
.
Insular dysfunction within the salience network is associated with severity of symptoms and aberrant inter-network connectivity in major depressive disorder
.
Front Hum Neurosci
 .
7
:
930
.
Manoliu
A
Riedl
V
Doll
A
Bauml
JG
Muhlau
M
Schwerthoffer
D
Scherr
M
Zimmer
C
Forstl
H
Bauml
J
et al
2013
.
Insular dysfunction reflects altered between-network connectivity and severity of negative symptoms in schizophrenia during psychotic remission
.
Front Hum Neurosci
 .
7
:
1
16
.
Mayberg
HS
Brannan
SK
Tekell
JL
Silva
JA
Mahurin
RK
McGinnis
S
Jerabek
PA
.
2000
.
Regional metabolic effects of fluoxetine in major depression: serial changes and relationship to clinical response
.
Biol Psychiatry
 .
48
:
830
843
.
Mayberg
HS
Liotti
M
Brannan
SK
McGinnis
S
Mahurin
RK
Jerabek
PA
Silva
JA
Tekell
JL
Martin
CC
Lancaster
JL
et al
1999
.
Reciprocal limbic-cortical function and negative mood: converging PET findings in depression and normal sadness
.
Am J Psychiatry
 .
156
:
675
682
.
Michl
P
Meindl
T
Meister
F
Born
C
Engel
RR
Reiser
M
Hennig-Fast
K
.
2014
.
Neurobiological underpinnings of shame and guilt: a pilot fMRI study
.
Soc Cogn Affect Neurosci
 .
9
:
150
157
.
Ng
B
Dresler
M
Varoquaux
G
Poline
J-B
Greicius
M
Thirion
B
.
2014
.
Transport on riemannian manifold for functional connectivity-based classification
. In:
Golland
P
Hata
N
, editors.
MICCAI 2014
 .
Switzerland
:
Springer International Publishing
. p.
405
413
.
Ng
B
Varoquaux
G
Poline
J-B
Thirion
B
.
2012
.
A novel sparse graphical approach for multimodal brain connectivity inference
. In:
Ayache
N
Delingette
H
Golland
P
Mori
K
, editors.
Medical image computing and computer-assisted intervention—MICCAI 2012
 .
Berlin Heidelberg
:
Springer
. p.
707
714
.
Nichols
T
Hayasaka
S
.
2003
.
Controlling the familywise error rate in functional neuroimaging: a comparative review
.
Stat Methods Med Res
 .
12
:
419
446
.
Nolen-Hoeksema
S
.
2000
.
The role of rumination in depressive disorders and mixed anxiety/depressive symptoms
.
J Abnorm Psychol
 .
109
:
504
511
.
Pardo
JV
Pardo
PJ
Raichle
ME
.
1993
.
Neural correlates of self-induced dysphoria
.
Am J Psychiatry
 .
150
:
713
719
.
Richiardi
J
Achard
S
Bunke
H
Van De Ville
D
.
2013
.
Machine learning with brain graphs: predictive modeling approaches for functional imaging in systems neuroscience
.
IEEE Signal Process Magazine
 .
30
:
58
70
.
Richiardi
J
Eryilmaz
H
Schwartz
S
Vuilleumier
P
Van De Ville
D
.
2011
.
Decoding brain states from fMRI connectivity graphs
.
Neuroimage
 .
56
:
616
626
.
Robinson
MS
Alloy
LB
.
2003
.
Negative cognitive styles and stress-reactive rumination interact to predict depression: a prospective study
.
Cogn Ther Res
 .
27
:
275
291
.
Sheline
YI
Barch
DM
Price
JL
Rundle
MM
Vaishnavi
SN
Snyder
AZ
Mintun
MA
Wang
S
Coalson
RS
Raichle
ME
.
2009
.
The default mode network and self-referential processes in depression
.
Proc Natl Acad Sci USA
 .
106
:
1942
1947
.
Sheline
YI
Price
JL
Yan
Z
Mintun
MA
.
2010
.
Resting-state functional MRI in depression unmasks increased connectivity between networks via the dorsal nexus
.
Proc Natl Acad Sci USA
 .
107
:
11020
11025
.
Shirer
WR
Ryali
S
Rykhlevskaia
E
Menon
V
Greicius
MD
.
2012
.
Decoding subject-driven cognitive states with whole-brain connectivity patterns
.
Cereb Cortex
 .
22
:
158
165
.
Sitaram
R
Lee
S
Ruiz
S
Rana
M
Veit
R
Birbaumer
N
.
2011
.
Real-time support vector classification and feedback of multiple emotional brain states
.
NeuroImage
 .
56
:
753
765
.
Smith
R
Fadok
RA
Purcell
M
Liu
S
Stonnington
C
Spetzler
RF
Baxter
LC
.
2011
.
Localizing sadness activation within the subgenual cingulate in individuals: a novel functional MRI paradigm for detecting individual differences in the neural circuitry underlying depression
.
Brain Imaging Behav
 .
5
:
229
239
.
Smith
SM
Jenkinson
M
Woolrich
MW
Beckmann
CF
Behrens
TEJ
Johansen-Berg
H
Bannister
PR
De Luca
M
Drobnjak
I
Flitney
DE
et al
2004
.
Advances in functional and structural MR image analysis and implementation as FSL
.
NeuroImage
 .
23
(Suppl. 1)
:
S208
S219
.
Tolos¸i
L
Lengauer
T
.
2011
.
Classification with correlated features: unreliability of feature ranking and solutions
.
Bioinformatics
 .
27
:
1986
1994
.
Treynor
W
Gonzalez
R
Nolen-Hoeksema
S
.
2003
.
Rumination reconsidered: a psychometric analysis
.
Cogn Ther Res
 .
27
:
247
259
.
Varoquaux
G
Baronnet
F
Kleinschmidt
A
Fillard
P
Thirion
B
.
2010
.
Detection of brain functional-connectivity difference in post-stroke patients using group-level covariance modeling
. In:
Jiang
T
Navab
N
Pluim
JPW
Viergever
MA
, editors.
Medical image computing and computer-assisted intervention—MICCAI 2010
 .
Berlin, Heidelberg
:
Springer
. p.
200
208
.
Vilberg
KL
Rugg
MD
.
2008
.
Memory retrieval and the parietal cortex: a review of evidence from a dual-process perspective
.
Neuropsychologia
 .
46
:
1787
1799
.
Von dem Hagen
EAH
Stoyanova
RS
Baron-Cohen
S
Calder
AJ
.
2013
.
Reduced functional connectivity within and between “social” resting state networks in autism spectrum conditions
.
Soc Cogn Affect Neurosci
 .
8
:
694
701
.
Vuilleumier
P
.
2005
.
How brains beware: neural mechanisms of emotional attention
.
Trends Cogn Sci
 .
9
:
585
594
.
Wilson
EB
.
1927
.
Probable inference, the law of succession, and statistical inference
.
J Am Statist Assoc
 .
22
:
209
212
.
Wilson-Mendenhall
CD
Barrett
LF
Barsalou
LW
.
2013
.
Neural evidence that human emotions share core affective properties
.
Psychol Sci
 .
24
:
947
956
.
Zamoscik
V
Huffziger
S
Ebner-Priemer
U
Kuehner
C
Kirsch
P
.
2014
.
Increased involvement of the parahippocampal gyri in a sad mood predicts future depressive symptoms
.
Soc Cogn Affect Neurosci
 .
1
7
. .

Author notes

Anna-Clare Milazzo and Bernard Ng contributed equally.