The core of brain imaging is settled: it maps structure superbly and reveals broad, replicated patterns of function, with real clinical and scientific value. What is mixed sits one level finer: how well activity maps onto specific thoughts, and which statistical methods are trustworthy. What is genuinely contested is reproducibility in small studies, over-interpretation in the media, and commercial or legal claims from neuro-marketing and neuro-law. The honest reader keeps the solid centre and the shaky edges clearly apart.
How brain imaging is studied and checked
To judge which claims are strong, it helps to know how imaging evidence is built and stress-tested. Confidence comes not from any single striking scan but from methods that hold up when others repeat, combine, and criticise them.
Task studies and meta-analysis
Many groups run similar tasks and pool the results across dozens or hundreds of studies. Patterns that recur across independent samples, such as which networks engage during language, are the ones that earn confidence.
Large shared datasets
Big open projects scan thousands of people with common protocols. Large samples tame the noise that plagued small early studies and let researchers check whether an effect survives outside one lab.
Converging methods
Imaging findings are tested against other tools: what happens after injury, what temporary disruption does, what electrical recordings show. When these agree, a claim moves from suggestive to solid.
Methods criticism
A vigorous internal literature probes the statistics themselves, exposing pitfalls like uncorrected multiple comparisons. This self-policing, including the dead-salmon demonstration, is how the field keeps itself honest.
No single study settles anything alone. A finding is trustworthy when it replicates in large samples, survives proper statistics, and lines up with evidence from other methods. That is the yardstick applied below.
A few figures worth knowing
These numbers give a feel for the scale and the pitfalls of the field. Treat them as rough, illustrative characterisations rather than precise constants.
Settled: what the evidence firmly supports
These claims rest on decades of work across independent methods and are not seriously disputed among researchers who study imaging.
Imaging maps brain structure with high reliability. Structural MRI and CT show the anatomy of a living brain clearly enough to find and follow tumours, strokes, and tissue changes. This is among the most valuable diagnostic advances of the last century, is used routinely worldwide, and is not in question. When it comes to what the brain physically looks like, imaging is on entirely firm ground.
Functional imaging reveals broad, replicated patterns of activity. Across thousands of studies, particular classes of task reliably engage particular networks: vision at the back of the brain, movement along a strip near the top, language typically toward the left. That the brain is organised into coordinating networks, active even at rest, with a fairly consistent layout across people, is well established. Imaging has genuinely mapped the coarse functional geography of the brain.
Imaging has real clinical and scientific value. Beyond diagnosis, imaging guides surgery, tracks recovery, and has advanced understanding of perception, memory, and attention. Its usefulness does not depend on the exaggerated claims sometimes made for it. The tool earns its place through the solid, unglamorous work of mapping structure and broad function accurately, and that value is not seriously contested.
Mixed: where the science is still moving
These questions are actively researched and partly answered, but the evidence is incomplete or the estimates shift with method. Here the honest posture is confidence about the broad shape and caution about the details.
How finely activity maps onto specific thoughts. Pattern-based analysis can, in constrained laboratory settings, distinguish which of a small, known set of images or categories a person is viewing from their brain activity. That is a real and impressive advance. But it is a long way from general mind-reading: it works within narrow, pre-trained choices, often needs the same person's own data, and degrades badly outside the lab. How far this decoding can be pushed toward reading open-ended, spontaneous thought is unresolved, and easy to overstate.
Which statistical methods are trustworthy. The field has learned, sometimes painfully, that certain older analysis choices produced fragile results. Standards have tightened: stricter correction for testing many voxels, larger samples, pre-registration, and sharing of data and code. These are real improvements, but they are unevenly adopted, and how best to analyse complex imaging data is still an active methodological debate rather than a solved problem.
Individual-level prediction and biomarkers. There is real hope that imaging might one day help predict outcomes or guide treatment for individuals. Progress is genuine but slow, and most candidate markers that look promising in one sample shrink or vanish in another. Group differences are robust; turning them into a reliable test for one person is much harder, and confident claims of a scan-based diagnosis for most conditions run ahead of the evidence.
Contested: where the real disputes are
These are genuine debates, shaped by data, incentives, and interpretation, where thoughtful people reach different conclusions. This page describes the disagreement rather than taking a side.
Reproducibility and the small-sample problem. Many influential early studies used very small samples, which makes findings unstable and easy to over-fit to noise. The dead-salmon demonstration by Craig Bennett and colleagues dramatised the danger: run enough voxel comparisons without correction, and even a dead fish appears to show brain activity. How much of the older imaging literature holds up, and how thoroughly newer standards have fixed it, is actively argued. Reformers say the field has turned a corner; sceptics say a large back-catalogue of underpowered results still circulates.
Over-interpretation in the media, and neuro-hype. There is broad agreement that imaging is often over-sold in the press, with colourful pictures and reverse-inference claims, but the harder question is how much this distorts public understanding and policy. Some argue it mostly makes for silly headlines; others see real damage when shaky brain claims sway courts, classrooms, or clinics. Where legitimate communication ends and hype begins is a genuine, ongoing dispute.
Neuro-marketing and neuro-law. Commercial claims that a scan reveals a hidden buying trigger, and legal claims that imaging can establish a defendant's state of mind or detect a lie, are the most disputed applications of all. Many rest on reverse inference or on individual predictions the science cannot yet support. Courts have generally been sceptical of brain-based lie detection. Supporters see early promise; critics see over-reach dressed in the authority of a brain scan, and the debate is unresolved.
The mistake to avoid
Because the contested edges are real, it is tempting to conclude the whole enterprise is unreliable. That inversion is as wrong as the hype it reacts against. The dead-salmon result is not proof that fMRI is worthless; it is proof that the field polices its own statistics, and it led directly to better standards. Healthy self-criticism at the edges of a science is a sign of maturity, not fraud. The solid centre, structure and broad function, stands regardless of how the arguments about fine-grained decoding and commercial claims are eventually settled. Keeping the strong core and the weak edges distinct is the whole art of reading imaging honestly, a theme the what it reveals page develops in detail.
Where to go next
To see the interpretive errors behind the contested claims, read what brain imaging reveals. To understand why the signal invites over-reading, revisit how brain imaging works. Or return to the overview.
Sources
- Bennett CM, Baird AA, Miller MB, Wolford GL. Neural correlates of interspecies perspective taking in the post-mortem Atlantic salmon: an argument for multiple comparisons correction. Journal of Serendipitous and Unexpected Results. 2010;1(1):1-5.
- Poldrack RA. Can cognitive processes be inferred from neuroimaging data? Trends in Cognitive Sciences. 2006;10(2):59-63.
- Logothetis NK. What we can do and what we cannot do with fMRI. Nature. 2008;453:869-878.
- Farah MJ, Hutchinson JB, Phelps EA, Wagner AD. Functional MRI-based lie detection: scientific and societal challenges. Nature Reviews Neuroscience. 2014;15:123-131.
This page is educational and weighs the state of the evidence on imaging as a scientific tool. It does not diagnose any individual, and the findings it describes are group-level, statistical results.