Structural MRI shows what the brain looks like. Functional MRI (fMRI) shows what it is doing. Over the past two decades, fMRI has become one of the dominant tools in schizophrenia research — generating thousands of papers and reshaping how scientists think about the disorder. It has also generated reasonable scepticism about reproducibility, individual prediction, and clinical translation.
Functional MRI in schizophrenia has consistently shown altered activity and connectivity in large-scale brain networks — including the default mode, salience, and frontoparietal networks — but no fMRI signature is yet specific or reliable enough to diagnose the disorder in an individual.
How fMRI works
fMRI measures the blood-oxygen-level-dependent (BOLD) signal, an indirect proxy for neural activity. When a brain region works harder, blood flow to it increases more than the oxygen extracted, shifting the magnetic properties of haemoglobin. The MRI scanner detects that shift. fMRI is performed in two main ways:
- Task-based fMRI — measures BOLD signal while a person performs a cognitive or emotional task in the scanner.
- Resting-state fMRI — measures spontaneous BOLD fluctuations while the person lies awake but not doing a task. Patterns of correlation across regions reveal "intrinsic networks."
fMRI has high spatial resolution (millimetres) and modest temporal resolution (seconds). It is non-invasive, does not use radiation, and is widely available.
The big networks
Modern fMRI research has converged on a view of the brain as organised into a small number of large-scale networks. Three are particularly relevant to schizophrenia:
- Default mode network (DMN). Active during rest, mind-wandering, and self-referential thinking. In schizophrenia, multiple meta-analyses have shown altered DMN connectivity — sometimes hyperconnected, sometimes hypoconnected, and often poorly anti-correlated with task-positive networks.
- Salience network. Centred on the anterior insula and dorsal anterior cingulate, this network detects motivationally important stimuli and toggles between rest and task networks. Disrupted salience network function is a leading candidate explanation for the "aberrant salience" model of psychosis.
- Central executive (frontoparietal) network. Engaged in working memory and goal-directed cognition. Altered activity here is among the most consistent fMRI findings in schizophrenia, often correlating with cognitive symptom severity.
Major findings
Task-based studies
People with schizophrenia have shown altered prefrontal activation during working memory tasks — sometimes hypoactivation, sometimes hyperactivation depending on task difficulty and patient state. This was characterised in early N-back fMRI studies and has been replicated across many cohorts. Reward-processing tasks have shown blunted ventral striatum activation in response to reward cues, particularly in patients with prominent negative symptoms.
Resting-state studies
Large multi-site collaborations like the SchizConnect consortium and the COBRE dataset have pooled resting-state fMRI from thousands of patients. Consistent themes include altered thalamocortical connectivity (with hyperconnectivity to sensorimotor cortex and hypoconnectivity to prefrontal cortex), and altered between-network anti-correlations.
Auditory hallucinations
fMRI studies during active auditory hallucinations have shown engagement of speech-perception areas (superior temporal gyrus, Heschl's gyrus) and atypical activity in language and self-monitoring regions. A key model holds that aberrant activation of auditory cortex without an external stimulus generates the perceptual experience of voices. See our piece on auditory hallucinations.
Cognitive control and prediction error
Models of psychosis grounded in computational neuroscience — such as the predictive-coding framework — propose altered weighting of "prediction errors" in the brain. fMRI studies of reward learning and oddball paradigms have produced consistent (if modest) evidence supporting these models.
Caveats and limitations
fMRI in schizophrenia has well-known weaknesses:
- Effect sizes are small. Group differences are real but modest, and there is substantial overlap between patient and control distributions.
- Reproducibility is mixed. Some findings replicate well; others do not. The wider "replication crisis" in neuroimaging has affected schizophrenia research.
- Confounds are hard to remove. Antipsychotic medication, smoking, head motion, sleep, and caffeine can all affect BOLD signal. Many older studies did not control these adequately.
- Specificity is limited. Many findings overlap with bipolar disorder, depression, autism, and other conditions.
- Individual prediction is not yet possible. Despite many machine-learning attempts, no fMRI biomarker reliably distinguishes individuals with schizophrenia from healthy controls in clinical settings.
Toward biomarkers
Multi-site initiatives have tried to push fMRI toward usable biomarkers. ENIGMA's schizophrenia working group, the Bipolar-Schizophrenia Network on Intermediate Phenotypes (B-SNIP), and the Human Connectome Project all contribute large pooled datasets. Progress is real but incremental. The most promising near-term clinical application is probably stratification — identifying subgroups of patients who may respond to different treatments — rather than diagnostic biomarkers.
Pharmacological fMRI
fMRI is also used to study how medications affect brain activity. Acute and chronic antipsychotic effects on BOLD signal have been mapped, often showing normalisation of striatal activation patterns. Pharmacological fMRI with ketamine — which produces a transient psychosis-like state in healthy volunteers — has been particularly productive in modelling acute psychotic mechanisms.
Functional MRI is not used in routine schizophrenia diagnosis. Standard clinical brain MRI is sometimes performed to rule out structural causes of new psychosis, but it does not include fMRI sequences.
Where the field is going
Several directions are active right now:
- Higher-field MRI. 7-Tesla scanners offer better resolution and may reveal sub-millimetre architectural details.
- Naturalistic paradigms. Watching films, hearing stories, and free conversation in the scanner produce richer signals than artificial tasks.
- Dense longitudinal sampling. Following individuals across many scans over time, including across symptomatic states, to build personalised trajectories.
- Combination with other modalities. Pairing fMRI with EEG, PET, and genomics for richer multimodal pictures.
- Better methods. More rigorous handling of motion, shorter and more reliable scans, and pre-registered hypotheses to reduce selective reporting.
The takeaway
fMRI has fundamentally changed how schizophrenia is conceptualised — not as a disease of one brain region but of brain networks and their interactions. It has clarified some mechanisms, generated useful models, and helped fix dopamine in a richer context. It has not yet produced a diagnostic test or a clinically actionable biomarker. For patients, this means fMRI remains in the research realm. For the field, it means there is still serious scientific work to do.
For more brain imaging research, see our pieces on PET imaging, MR spectroscopy, and brain volume changes.
This article is for educational purposes only and is not medical advice, diagnosis, or treatment. Always consult a qualified mental health professional. If you or someone you know is in crisis, call or text 988 in the US, or your local emergency number.