Brain imaging refers to a family of technologies that let researchers and doctors look at the structure and activity of a living brain without surgery. From the earliest X-rays of the skull over a century ago to today’s ultra-high-resolution scanners, the field has given us tools to diagnose strokes in real time, track proteins linked to Alzheimer’s disease, and even reconstruct rough images of what a person was looking at based on their brain activity. Each technique captures something different, and knowing what each one does well and where it falls short matters for understanding both the headlines and the clinical decisions that follow.
From Skull X-Rays to Modern Scanners
The story of brain imaging begins in the early 1900s with plain skull radiographs, which could reveal fractures and some tumors but gave almost no information about brain tissue itself. In 1918 and 1919, doctors introduced ventriculography and pneumoencephalography, methods that involved injecting air or contrast agents into the brain’s fluid spaces to make structures visible on film. These procedures worked, but they were invasive, painful, and carried real risks. Carotid angiography followed in 1927, and for roughly the next four decades, those techniques were the best available options for seeing inside a living skull.1PubMed Central. Neuroradiology back to the future: brain imaging
The arrival of CT scanning in 1971 changed everything. Suddenly, doctors could see the brain’s soft tissue in cross-section without injecting anything into the skull, and procedures like pneumoencephalography became obsolete almost overnight. MRI followed in the early 1980s, offering even sharper contrast between different types of brain tissue and no ionizing radiation. From there, the pace of innovation only accelerated: functional MRI arrived in the 1990s, PET imaging became a mainstream research tool, and a growing list of portable and wearable sensors continues to expand what is possible.
How fMRI Reads Brain Activity
Functional MRI is probably the technique most people picture when they hear “brain scan.” It works by detecting changes in blood oxygenation. When a patch of brain tissue becomes active, local blood flow increases to deliver oxygen. That oxygen is carried by hemoglobin, and hemoglobin behaves differently in a magnetic field depending on whether it is carrying oxygen or not. The signal that emerges from this difference is called the blood-oxygen-level-dependent, or BOLD, signal.2PubMed Central. Coupling mechanism and significance of the BOLD signal: a status report
Crucially, fMRI does not directly measure neurons firing. It measures the vascular response to neural activity, a process called neurovascular coupling. This distinction matters because anything that changes the blood vessels, such as aging, certain medications, or vascular disease, can alter the BOLD signal independently of what the neurons are actually doing. Aging, for instance, changes both the structure and function of the brain’s blood vessels, which can affect how the BOLD signal is recorded and interpreted.3PubMed Central. Age-related alterations in the cerebrovasculature affect neurovascular coupling and BOLD fMRI responses: Insights from animal models of aging Researchers comparing brain activity across age groups need to account for this vascular confound, not just assume that a weaker signal means less neural activity in older brains.
Other Ways to Watch the Brain
fMRI captures spatial detail well but is slow: the BOLD response peaks a few seconds after neurons fire. Other techniques fill different niches.
- EEG and MEG: Electroencephalography records electrical signals from the scalp, and magnetoencephalography picks up the tiny magnetic fields those currents produce. Both track neural activity on a millisecond timescale, far faster than fMRI, though they are less precise about where in the brain the signal originates. Combining MEG and EEG data consistently improves localization accuracy over either one alone, because the two methods are sensitive to different physical properties of brain tissue rather than simply adding more sensors.4PubMed Central. The advantage of combining MEG and EEG: comparison to fMRI in focally stimulated visual cortex
- fNIRS: Functional near-infrared spectroscopy shines light through the skull and detects changes in blood oxygenation, similar in principle to fMRI but using optics rather than magnets. It is portable, tolerant of movement, and well-suited for bedside monitoring, making it especially useful in newborns and other populations that cannot stay still inside a scanner.5PubMed Central. Monitoring neonatal brain function: recent advances in functional near-infrared spectroscopy (fNIRS)
- Diffusion MRI: Water molecules in the brain’s white matter pathways diffuse more easily along the length of nerve fibers than across them. Diffusion imaging exploits that asymmetry to map the brain’s long-range wiring, revealing bundles of connections that no other non-invasive method can see.6PubMed Central. MAPPING BRAIN ANATOMICAL CONNECTIVITY USING WHITE MATTER TRACTOGRAPHY
- PET: Positron emission tomography uses radioactive tracers that bind to specific molecules in the brain, such as the glucose neurons consume or the abnormal proteins that accumulate in Alzheimer’s disease. PET sacrifices some spatial resolution compared with MRI but can answer chemical questions that MRI cannot, like how much of a particular protein has built up in a given region.
Stroke Diagnosis at the Bedside
When someone arrives at a hospital with stroke symptoms, every minute counts. Doctors need to know whether the stroke is caused by a clot or a bleed, how much brain tissue is already dead, and how much is still salvageable. Both MRI-based and CT-based perfusion imaging can answer these questions, and the evidence consistently shows that the two approaches lead to nearly identical treatment decisions.
In one study of 42 stroke patients scanned within hours of symptom onset, CT perfusion with angiography and MRI agreed on treatment eligibility in all but one case.7PubMed. Comparison of CT perfusion and angiography and MRI in selecting stroke patients for acute treatment Earlier work found extremely high correlations between CT perfusion and MRI diffusion-weighted imaging for measuring infarct size, concluding the two techniques are essentially equivalent for identifying salvageable tissue.8PubMed. Comparison of admission perfusion computed tomography and qualitative diffusion- and perfusion-weighted magnetic resonance imaging in acute stroke patients A later study confirmed strong agreement: CT and MRI matched on whether the infarct core exceeded a critical size threshold about 91% of the time and agreed on trial enrollment criteria in roughly 84% of patients.9PubMed. Quantitative assessment of core/penumbra mismatch in acute stroke: CT and MR perfusion imaging are strongly correlated when sufficient brain volume is imaged
This matters practically because CT scanners are far more widely available than MRI machines, especially in smaller hospitals and emergency departments. Knowing that CT perfusion provides equivalent information means stroke patients in facilities without MRI can still receive the imaging needed to guide clot-dissolving therapy.
Tracking Alzheimer’s Proteins with PET
Alzheimer’s disease involves two hallmark proteins that accumulate in the brain years before memory loss becomes obvious: amyloid plaques and tau tangles. PET tracers that bind to each of these proteins have become vital tools for diagnosis and for monitoring whether experimental treatments are working.10PubMed Central. Amyloid and Tau Positron Emission Tomography Imaging in Alzheimer’s Disease and Other Tauopathies
Longitudinal PET studies have revealed that tau accumulates in a particular pattern: significant rates of buildup appear only after amyloid levels have already become abnormal. Once the disease progresses to the point of cognitive impairment, tau accumulation spreads broadly across the brain rather than marching through one region at a time.11Brain. Longitudinal tau PET in ageing and Alzheimer’s disease In people with mild cognitive impairment who showed high levels of both amyloid and tau at the start of a study, tau levels continued to rise over two years alongside a correlated rise in brain inflammation.12PubMed Central. The relationships between neuroinflammation, beta-amyloid and tau deposition in Alzheimer’s disease: a longitudinal PET study
These findings have practical implications for drug trials. Because tau accumulation can be tracked with PET over relatively short periods, tau PET scans can serve as an efficient way to measure whether a therapy is slowing the disease, without waiting years for cognitive decline to become measurable on tests.
Mapping the Brain Before Tumor Surgery
When a brain tumor sits near regions that control movement, language, or vision, surgeons face a delicate tradeoff: remove too little and the tumor returns; remove too much and the patient loses function. Preoperative fMRI can identify which nearby brain areas are active during specific tasks, and diffusion-based tractography can show the white-matter pathways threading past the tumor.13PubMed. A systematic review of functional magnetic resonance imaging and diffusion tensor imaging modalities used in presurgical planning of brain tumour resection Used together, these techniques map the functional landscape around the tumor so the surgeon can plan a safer route.
There is an important caveat: when a tumor sits very close to functional areas, it can distort the local BOLD signal, making fMRI-guided mapping less reliable. In those cases, navigated transcranial magnetic stimulation has proven more accurate for guiding tractography of the motor pathways than fMRI-based approaches.14PubMed Central. Preoperative mapping techniques for brain tumor surgery: a systematic review Still, preoperative fMRI combined with tractography and intraoperative stimulation is associated with better surgical outcomes, including greater tumor removal and fewer functional deficits.15PubMed Central. Current State of Functional MRI in the Presurgical Planning of Brain Tumors
The Default Mode Network and Resting-State Research
One of the most surprising discoveries to come out of brain imaging was that the brain does not go quiet when you stop doing tasks. In the 1990s and early 2000s, PET and fMRI studies noticed a consistent set of brain regions that were more active during rest than during focused tasks. This network, now called the default mode network, was initially puzzling because researchers were accustomed to thinking about the brain in terms of what activates during a specific challenge, not what deactivates.16PubMed. The brain’s default mode network
Early connectivity analyses of resting-state data provided some of the first compelling evidence that this network functions as a cohesive unit, with its component regions tightly synchronized even when a person is simply lying still with their eyes open.17PubMed Central. Functional connectivity in the resting brain: a network analysis of the default mode hypothesis The discovery reignited interest in the brain’s spontaneous background activity and launched an entire subfield. Resting-state studies now play a major role in studying conditions from depression to autism to Alzheimer’s disease, because disrupted connectivity patterns in this and other networks often show up before obvious symptoms do.
The False-Positive Problem
A brain scan showing colorful blobs of “activation” looks authoritative, but producing those maps requires a chain of statistical decisions, each of which can go wrong. One major concern is the multiple-comparisons problem. A typical fMRI analysis tests tens of thousands of small brain volumes simultaneously for a significant response. Without proper correction, the odds of finding something by chance alone are enormous.18PubMed Central. Zen and the art of multiple comparisons
The most common correction method, cluster-extent thresholding, groups neighboring voxels that pass a preliminary threshold and asks whether the cluster as a whole is large enough to be unlikely by chance. This approach is sensitive to weak, spread-out signals but carries a trade-off: it has low spatial specificity, meaning researchers can only say there is a signal somewhere in the cluster, not pinpoint exactly where. Using lenient preliminary thresholds makes the problem worse by producing very large clusters that span multiple brain regions.19PubMed Central. Cluster-extent based thresholding in fMRI analyses: pitfalls and recommendations
A widely cited 2016 analysis drove the point home. Using three million random group analyses on real resting-state data, the authors tested the major fMRI software packages and found that parametric cluster-based methods produced inflated false-positive rates, sometimes dramatically so. The principal culprit was a flawed assumption about the spatial smoothness of the data.20PubMed Central. Cluster failure: Why fMRI inferences for spatial extent have inflated false-positive rates The paper prompted software updates and renewed attention to statistical rigor. It does not mean that all prior fMRI work is wrong, but it underscores that the pretty brain maps consumers of science encounter are only as reliable as the statistical choices behind them.
Higher Resolution and Wearable Scanners
Conventional clinical MRI scanners typically operate at 1.5 or 3 Tesla. Moving to 7 Tesla dramatically improves both signal-to-noise ratio and contrast, making it possible to visualize tiny structures like the subthalamic nucleus, individual cortical layers, and small perforating arteries that are invisible at lower field strengths.21PubMed Central. Neuroimaging at 7 Tesla: a pictorial narrative review A next-generation 7T system has pushed functional imaging down to roughly 0.4 mm resolution, fine enough to distinguish activity in individual layers of cortex.22PubMed Central. Next-generation MRI scanner designed for ultra-high-resolution human brain imaging at 7 Tesla That level of granularity opens up questions about how information flows between cortical layers, something that was previously accessible only in animal studies using invasive electrodes.
Meanwhile, a different revolution is unfolding outside the scanner room. Traditional MEG systems require liquid helium to keep their sensors at near absolute zero, which means the sensors sit inside a rigid, one-size-fits-all helmet at a fixed distance from the scalp. Optically pumped magnetometers, or OPMs, are a new generation of magnetic-field sensors that need no cryogenic cooling and can be mounted directly on the head in a flexible, lightweight array.23PubMed Central. Optically pumped magnetometers: From quantum origins to multi-channel magnetoencephalography Early whole-head OPM systems with far fewer sensors have already shown performance comparable to conventional cryogenic MEG for detecting standard brain rhythms.24PubMed Central. Multi-channel whole-head OPM-MEG: Helmet design and a comparison with a conventional system Because participants can move naturally while wearing the system, OPM-MEG could eventually be used with children, patients who cannot hold still, or in real-world settings far removed from a traditional laboratory.
AI, Image Reconstruction, and the Privacy Question
Recent work has pushed brain imaging into territory that feels like science fiction. Researchers have trained machine-learning models to take a person’s fMRI data while they view an image and generate a rough reconstruction of what the person was seeing. One approach combined a language model for extracting semantic content from brain activity with a diffusion model for generating the reconstructed image.25Scientific Reports. Improved image reconstruction from brain activity through automatic image captioning Another used a generative adversarial network within a Bayesian framework to improve both the visual fidelity and the natural appearance of reconstructed images from fMRI data.26PubMed. BigGAN-based Bayesian Reconstruction of Natural Images from Human Brain Activity Deep learning has also been applied to clinical image analysis: models for simultaneous lesion and brain segmentation in multiple sclerosis have matched or exceeded the accuracy of individual human raters.27Scientific Reports. Simultaneous lesion and brain segmentation in multiple sclerosis using deep neural networks
These advances raise genuinely difficult ethical questions. Brain data is uniquely sensitive because it can reveal information a person never intended to share, from health conditions to cognitive tendencies. As neurotechnologies generate ever-larger databases of brain-derived data, the risks to privacy multiply. Profiling based on neural data could marginalize individuals or groups, and existing legal frameworks were not designed for data this intimate.28PubMed Central. Brain Recording, Mind-Reading, and Neurotechnology: Ethical Issues from Consumer Devices to Brain-Based Speech Decoding Several countries and the Organisation for Economic Co-operation and Development have begun exploring “neurorights” legislation, but the regulatory landscape is still far behind the technology.
Watching the Developing Brain
Imaging the brains of children and adolescents presents unique challenges. Kids move more, have smaller heads, and their brains are changing rapidly. But longitudinal structural MRI studies have started to reveal how the developing brain’s anatomy relates to behavior. One study tracking adolescents over time found that decreasing hippocampal volume was linked to parent-reported aggression, and that the coupling between prefrontal regions, the limbic system, and the striatum during maturation was associated with both low and high levels of externalizing behavior.29PubMed Central. Longitudinal structural brain development and externalizing behavior in adolescence
The broader trend in pediatric neuroimaging has shifted from simply comparing average brain sizes at different ages toward tracking individual developmental trajectories and studying how circuits mature together rather than in isolation.30Neuron. Structural MRI of Pediatric Brain Development: What Have We Learned and Where Are We Going? This shift matters because two children with the same overall brain volume may have very different patterns of cortical thinning, white-matter maturation, and connectivity, and those differences may relate to cognitive and behavioral outcomes in ways that a single snapshot never could.
Comparing Brains Across Species
Brain imaging is not limited to humans. The same MRI, PET, and diffusion techniques used in clinical and research settings have been adapted for comparative studies across primate species, revealing what makes the human brain distinctive. Structural MRI has been used to compare brain proportions and aging patterns; PET to compare baseline glucose metabolism; fMRI to compare sensory pathways and resting-state networks; and diffusion imaging to compare structural wiring. Together, these methods have highlighted human-specific features in cortical organization, connectivity, and the trajectory of brain development.31PubMed. Comparative primate neuroimaging: insights into human brain evolution
An MRI-based analysis of brain folding across 34 primate species turned up a striking regularity: despite a twenty-fold range in brain volume among species with folded cortices, the characteristic wavelength of cortical folds stayed remarkably stable at roughly 12 mm. The analysis also suggested, based on evolutionary modeling, that the common ancestor of all primates likely had a folded brain resembling that of a small lemur.32PubMed. Evolution of neocortical folding: A phylogenetic comparative analysis of MRI from 34 primate species Findings like these are only possible because imaging lets researchers study brain anatomy across many species using the same measurement framework, without requiring any invasive procedure.