Brain Imaging Studies of Anxiety: What They Show and What They Don't

A brain scan showing the amygdala “lighting up” during anxiety is one of the most recognizable images in mental health communication. It looks like hard evidence — a picture of the problem. But the science behind these images is considerably more limited than the presentation suggests.

What brain imaging studies typically do

Most fMRI studies of anxiety follow a similar pattern: a group of people with an anxiety disorder and a comparison group of people without one are scanned while performing a task — viewing threatening faces, anticipating an electric shock, or responding to emotional words. Researchers then compare brain activation between the groups, looking for regions that are more or less active in the anxious group.

A meta-analysis of these studies across PTSD, social anxiety and specific phobia found consistent hyperactivation in the amygdala and insula, with altered prefrontal cortex activity [1]. These findings form the basis of most popular accounts of the “anxious brain.”

The reverse inference problem

The most common misinterpretation of brain imaging data is reverse inference: reasoning from brain activation to a mental state. Poldrack’s influential critique explained why this fails [2]. The amygdala responds to novelty, salience, reward, surprise and social evaluation — not just fear. Seeing amygdala activation and concluding “this is anxiety” is like seeing someone’s heart rate elevated and concluding they must be scared. They might be exercising.

This matters because many popular articles about anxiety include claims like “brain scans show that anxiety activates the fear center.” The amygdala is not a fear center; it is a structure involved in many processes, and its activation alone tells you almost nothing about what someone is experiencing.

The sample size problem

Neuroimaging is expensive. An hour of scan time can cost hundreds of pounds. Recruiting people with anxiety disorders who are willing and able to lie still in a loud, confined scanner adds difficulty. As a result, most fMRI studies of anxiety have 15–30 participants per group.

This matters enormously. An analysis of statistical power in neuroscience estimated that the median study was powered to detect only large effects, with median power around 21% [3]. This means most studies are too small to reliably find the effects they’re looking for. When they do find something, the effect size is likely inflated — the “winner’s curse” that comes from publishing only the results that cross a significance threshold in underpowered studies.

A large-scale analysis demonstrated that reproducible brain-behavior associations require samples in the thousands, not dozens [4]. Most anxiety neuroimaging falls far short of this.

The flexibility problem

fMRI data analysis involves many choices: how to preprocess the images, which statistical model to use, which brain regions to examine, which threshold to apply. Carp demonstrated that different reasonable analysis pipelines applied to the same data can produce dramatically different results [5]. This means that the published result is one version of what the data could show, not the only or necessarily the most accurate version.

Combined with the pressure to publish novel, significant findings, this flexibility creates a situation where the literature may over-represent certain brain regions and under-represent null results. A region might appear consistently implicated not because it truly is, but because analyses that find it are more likely to be published.

What imaging can and cannot do

At a group level, neuroimaging has contributed genuine knowledge. The pattern of altered amygdala, prefrontal and insula activity in anxiety is probably real in broad strokes, even if individual study results are noisy.

For individual diagnosis, brain imaging is not useful. No clinical guideline recommends brain scans for anxiety diagnosis. The overlap between groups is too large and individual variability too high. Multivariate approaches — using patterns across many brain regions rather than single-region activation — show more promise [6], but none are clinically validated for anxiety.

For measuring treatment effects, some studies report brain changes after treatment, but these are typically the same small, underpowered designs. Whether brain changes add information beyond what symptom questionnaires already capture is debatable.

Questions to ask about a brain imaging study

1. How many participants were in each group?
2. Was the analysis plan pre-registered, or could the researchers have tried multiple approaches?
3. Does the study use reverse inference — concluding a mental state from brain activation?
4. Has the finding replicated in an independent sample?
5. Does the imaging add information beyond what behavioral measures already show?

What remains uncertain

Whether brain imaging will eventually become clinically useful for anxiety is an open question. Machine learning approaches may extract diagnostic information from whole-brain patterns that traditional analyses miss. But the field is currently in a position where the technology produces striking images that far outrun the conclusions the data actually support. A healthy skepticism toward “brain scan shows” claims is well justified.

Brain scans and your anxiety. If someone suggests you need a brain scan for anxiety, or you’ve seen brain imaging results that worried you, know that no brain scan is currently used or recommended for diagnosing anxiety disorders. Standard clinical assessment is the appropriate path. If anxiety is affecting your daily life, talk to a GP or mental health professional. If you are in crisis or thinking about harming yourself, contact emergency services or a crisis line now: in the US, call or text 988; in the UK and Ireland, Samaritans on 116 123.