Anjali Krishnan

Cluster-extent based thresholding in fMRI analyses: Pitfalls and recommendations

Cluster-extent based thresholding is currently the most popular method for multiple comparisons correction of statistical maps in neuroimaging studies, due to its high sensitivity to weak and diffuse signals. However, cluster-extent based thresholding provides low spatial specificity; researchers can only infer that there is signal somewhere within a significant cluster and cannot make inferences about the statistical significance of specific locations within the cluster. This poses a particular problem when one uses a liberal cluster-defining primary threshold (i.e., higher p-values), which often produces large clusters spanning multiple anatomical regions. In such cases, it is impossible to reliably infer which anatomical regions show true effects. From a survey of 814 functional magnetic resonance imaging (fMRI) studies published in 2010 and 2011, we show that the use of liberal primary thresholds (e.g., p<.01) is endemic, and that the largest determinant of the primary threshold level is the default option in the software used. We illustrate the problems with liberal primary thresholds using an fMRI dataset from our laboratory (N=33), and present simulations demonstrating the detrimental effects of liberal primary thresholds on false positives, localization, and interpretation of fMRI findings. To avoid these pitfalls, we recommend several analysis and reporting procedures, including 1) setting primary p<.001 as a default lower limit; 2) using more stringent primary thresholds or voxel-wise correction methods for highly powered studies; and 3) adopting reporting practices that make the level of spatial precision transparent to readers. We also suggest alternative and supplementary analysis methods.

A Sensitive and Specific Neural Signature for Picture-Induced Negative Affect

Neuroimaging has identified many correlates of emotion but has not yet yielded brain representations predictive of the intensity of emotional experiences in individuals. We used machine learning to identify a sensitive and specific signature of emotional responses to aversive images. This signature predicted the intensity of negative emotion in individual participants in cross validation (n =121) and test (n = 61) samples (high–low emotion = 93.5% accuracy). It was unresponsive to physical pain (emotion–pain = 92% discriminative accuracy), demonstrating that it is not a representation of generalized arousal or salience. The signature was comprised of mesoscale patterns spanning multiple cortical and subcortical systems, with no single system necessary or sufficient for predicting experience. Furthermore, it was not reducible to activity in traditional “emotion-related” regions (e.g., amygdala, insula) or resting-state networks (e.g., “salience,” “default mode”). Overall, this work identifies differentiable neural components of negative emotion and pain, providing a basis for new, brain-based taxonomies of affective processes.

Somatic and vicarious pain are represented by dissociable multivariate brain patterns

Understanding how humans represent others’ pain is critical for understanding pro-social behavior. ‘Shared experience’ theories propose common brain representations for somatic and vicarious pain, but other evidence suggests that specialized circuits are required to experience others’ suffering. Combining functional neuroimaging with multivariate pattern analyses, we identified dissociable patterns that predicted somatic (high versus low: 100%) and vicarious (high versus low: 100%) pain intensity in out-of-sample individuals. Critically, each pattern was at chance in predicting the other experience, demonstrating separate modifiability of both patterns. Somatotopy (upper versus lower limb: 93% accuracy for both conditions) was also distinct, located in somatosensory versus mentalizing-related circuits for somatic and vicarious pain, respectively. Two additional studies demonstrated the generalizability of the somatic pain pattern (which was originally developed on thermal pain) to mechanical and electrical pain, and also demonstrated the replicability of the somatic/vicarious dissociation. These findings suggest possible mechanisms underlying limitations in feeling others’ pain, and present new, more specific, brain targets for studying pain empathy. DOI: http://dx.doi.org/10.7554/eLife.15166.001

Quantifying cerebral contributions to pain beyond nociception

Cerebral processes contribute to pain beyond the level of nociceptive input and mediate psychological and behavioural influences. However, cerebral contributions beyond nociception are not yet well characterized, leading to a predominant focus on nociception when studying pain and developing interventions. Here we use functional magnetic resonance imaging combined with machine learning to develop a multivariate pattern signature—termed the stimulus intensity independent pain signature-1 (SIIPS1)—that predicts pain above and beyond nociceptive input in four training data sets (Studies 1–4, N=137). The SIIPS1 includes patterns of activity in nucleus accumbens, lateral prefrontal and parahippocampal cortices, and other regions. In cross-validated analyses of Studies 1–4 and in two independent test data sets (Studies 5–6, N=46), SIIPS1 responses explain variation in trial-by-trial pain ratings not captured by a previous fMRI-based marker for nociceptive pain. In addition, SIIPS1 responses mediate the pain-modulating effects of three psychological manipulations of expectations and perceived control. The SIIPS1 provides an extensible characterization of cerebral contributions to pain and specific brain targets for interventions.

The impact of T1 versus EPI spatial normalization templates for fMRI data analyses

Spatial normalization of brains to a standardized space is a widely used approach for group studies in functional magnetic resonance imaging (fMRI) data. Commonly used template‐based approaches are complicated by signal dropout and distortions in echo planar imaging (EPI) data. The most widely used software packages implement two common template‐based strategies: (1) affine transformation of the EPI data to an EPI template followed by nonlinear registration to an EPI template (EPInorm) and (2) affine transformation of the EPI data to the anatomic image for a given subject, followed by nonlinear registration of the anatomic data to an anatomic template, which produces a transformation that is applied to the EPI data (T1norm). EPI distortion correction can be used to adjust for geometric distortion of EPI relative to the T1 images. However, in practice, this EPI distortion correction step is often skipped. We compare these template‐based strategies empirically in four large datasets. We find that the EPInorm approach consistently shows reduced variability across subjects, especially in the case when distortion correction is not applied. EPInorm also shows lower estimates for coregistration distances among subjects (i.e., within‐dataset similarity is higher). Finally, the EPInorm approach shows higher T values in a task‐based dataset. Thus, the EPInorm approach appears to amplify the power of the sample compared to the T1norm approach when not using distortion correction (i.e., the EPInorm boosts the effective sample size by 12–25%). In sum, these results argue for the use of EPInorm over the T1norm when no distortion correction is used. Hum Brain Mapp 38:5331–5342, 2017.

When pain really matters: A vicarious-pain brain marker tracks empathy for pain in the romantic partner

In a previous study (Krishnan, 2016) we identified a whole-brain pattern, the Vicarious Pain Signature (VPS), which predicts vicarious pain when participants observe pictures of strangers in pain. Here, we test its generalization to observation of pain in a close significant other. Participants experienced painful heat (Self-Pain) and observed their romantic partner in pain (Partner-Pain). We measured whether (i) the VPS would respond selectively to Partner-Pain and (ii) the Neurologic Pain Signature (NPS), a measure validated to track somatic pain, would selectively respond to Self-Pain, despite the high interpersonal closeness between partners. The Partner-Pain condition activated the VPS (t = 4.71, p = 0.00005), but not the NPS (t = -1.03, p = 0.308). The Self-Pain condition activated the NPS (t = 13.70, p < .00005), but not the VPS (t = -1.03 p = 0.308). Relative VPS-NPS response differences strongly discriminated Partner-Pain vs. Self-Pain (cross-validated accuracy=97%, p < .000001). Greater interpersonal closeness between partners predicted greater VPS responses during Partner-Pain (r = 0.388, p = 0.050) and greater unpleasantness when observing the romantic partner in pain (r = 0.559, p = 0.003). The VPS generalizes across empathy paradigms and to an interactive social setting, and strongly activates when observing a close significant other in pain. VPS responses may be modulated by relevant interpersonal relationship factors. Self-Pain and Partner-Pain evoke non-overlapping large-scale neural representations.

Distance-Based Partial Least Squares Analysis

Distances matrices are traditionally analyzed with statistical methods that represent distances as maps such as Metric Multidimensional Scaling (mds), Generalized Procrustes Analysis (gpa), Individual Differences Scaling (indscal), and distatis. Mds analyzes only one distance matrix at a time while gpa, indscal and distatis extract similarities between several distance matrices. However, none of these methods is predictive. Partial Least Squares Regression (plsr) predicts one matrix from another, but does not analyze distance matrices. We introduce a new statistical method called Distance-based Partial Least Squares Regression (displsr), which predicts one distance matrix from another. We illustrate displsr with data obtained from a neuroimaging experiment, which explored semantic categorization.

Multiple brain networks mediating stimulus-pain relationships in humans

The brain transforms nociceptive input into a complex pain experience comprised of sensory, affective, motivational, and cognitive components. However, it is still unclear how pain arises from nociceptive input, and which brain networks coordinate to generate pain experiences. We introduce a new high-dimensional mediation analysis technique to estimate distributed, network-level patterns mediating the relationship between stimulus intensity and pain. In a large-scale analysis of functional magnetic resonance imaging data (N=284), we identify both traditional mediators in somatosensory brain regions and additional mediators located in prefrontal, midbrain, striatal, and default-mode regions unrelated to nociception in standard analyses. The whole brain mediators are specific for pain vs. aversive sounds and are organized in five functional networks. Brain mediators explain 32% more within-subject variance of single-trial pain ratings than previous brain-based models. Our results provide a new, broader view of the networks underlying pain experience, as well as distinct targets for interventions.