Unpacking the functional heterogeneity of the Emotional Face Matching Task: a normative modelling approach

Abstract

In this study we dissect the heterogeneity that underlies traditional group-level analyses, and determine how individualised patterns of predicted activation relate to age, sex, and variations in acquisition parameters and task design choices. To this end we take advantage of six large open-access/shared datasets and collate a large representative sample of over 7500 participants from which we build a normative of task-evoked activation during a widely used emotional reactivity task, the Emotional Face Matching Task. This enables us to bind heterogeneous datasets to a common reference model and enables meaningful comparisons between them. We then apply this model to the naturalistic and clinically realistic MIND-Set cohort, which is a heterogeneous and highly comorbid sample containing individuals with one or more current diagnosis (affective and anxiety disorders, autism spectrum disorders and/or attention deficit hyperactivity disorder). This enables us to determine whether, and if so how, participants with mental illness and/or neurodivergence differ from the reference cohort, both at the group level and at the level of the individual and in terms of cross-diagnostic symptom domains in addition to diagnosis. We show that patients have, on average, a higher frequency of extreme deviations, and have unique spatial distributions depending on the DSM-IV diagnosis and the number of co-occurring diagnoses when models are constructed using the face>shapes task contrast. Models built using the face>baseline task contrast, have, by comparison, greater predictive value for individuals' functioning across four transdiagnostic domains. We demonstrate the application of the normative modelling framework to task-based functional neuroimaging data, discuss its potential to further our understanding of individual differences in brain function within reference populations, and further validate the clinical relevance of these models.

Predictive Clinical Neuroscience Portal (PCNportal): instant online access to research-grade normative models for clinical neuroscientists.

“…We also welcome model contributions from other groups that can be added to the website after validation tests. For example, we plan to host the fine-grained normative models for the cerebellum reported in ( Gaiser et al ., 2023 ) and voxel-wise models of functional MRI data ( Savage et al ., 2023 ) sin the coming months.…”

Do you match what I match? A systematic review of 20 years of the emotional face matching task reveals vast inconsistencies in design and implementation.

“…With our study and others revealing the potentially huge influence of small design choices [47,66], it was near impossible and would have been uninformative to try to infer the relation of these design changes to task activation, as was our original aim. Furthermore summarising task activation was a major challenge due to the inconsistent naming conventions applied.…”

“…Considering the large heterogeneity between studies and subjects, additional variance within-subject only adds to variable responses that can be observed. Using the normative modeling framework, which will be explained below, may prove a suitable alternative to assessing Test-ReTest using individual activation maps, as the derived deviation scores reportedly capture stable individual differences in task evoked activation [66].…”

“…Another alternative to address the challenge of heterogeneity embraces recent methodological advances such as the normative modeling framework that facilitate the aggregation of datasets that use distinct versions of a task to a common reference model. Provided that datasets are transparently and completely reported, this enables meaningful comparisons to be drawn across studies, and affords the opportunity to quantitatively determine the influence of different task parameters and sample characteristics on functional activity across cohorts; this approach was recently applied to combine data from 6 independent sites, each with unique combinations of EMFT design and acquisition parameters [66].…”

Embracing variability in the search for biological mechanisms of psychiatric illness

“…The application of normative models to diverse psychiatric groups has consistently shown group means derived from case-control group mean comparisons are not representative of most individual patients. Specifically, while individuals with a psychiatric diagnosis typically show a higher frequency of deviations in measures of brain structure [33] and function [34] compared to controls, the location of these deviations varies considerably, regardless of whether robust group-average case-control differences are present. In most cases less than 10-20% of individuals with a given diagnosis exhibit extreme deviations in the same brain locus [33,35,36].…”