Understanding our brain requires more than just analyzing its structure—we must see it in action. This led to the development of fMRI: functional magnetic resonance imaging.
As far as we know, humans are the only species on Earth that studies itself. Since ancient times, people have sought to understand complex phenomena such as consciousness and free will, art and literature, inventions, imagination and the ability to synchronize bodily movements to a degree that allows us to play the piano, dance, or perform high jumps in athletic competitions. How can all these abilities emerge from a jelly-like mass of cells in our skull, weighing an average of 1.35 kg?
The philosophers of ancient Greece were divided on the question of where emotions, reason, imagination, and consciousness reside in our bodies. The philosopher and researcher Alcmaeon of Croton argued as early as the sixth or fifth century BCE that humans differ from other animals because their perception is based on the brain. Sadly, tAlcmaeon’s writings were lost over time, but his ideas were referenced in later works by Greek scholars who continued to discuss the question. The renowned philosopher Plato also believed that the brain was the organ of sensation and rationality, but his disciple Aristotle argued that intellect resided in the heart and that that the brain's function was to cool the body.
Nearly two thousand years later, in the 16th century CE, the Belgian physician Andreas Vesalius conducted dissections of human bodies and organs, leading to the publishing of his comprehensive anatomical work De Humani Corporis Fabrica (Latin for “On the Fabric of the Human Body”). Through his investigations, Vesalius observed that the structure of the human brain resembled that of other mammals but was more complex. He attributed uniquely human traits, such as speech and rational thought, to this increased complexity. However, Vesalius did not understand the role of the brain’s internal cavities—the ventricles filled with cerebrospinal fluid—and believed that this fluid contained the "spirit of life."
The truly significant leap in our understanding of the brain occurred only in the 19th century, following the pioneering work of researchers like Santiago Ramon y Cajal, who described the brain’s structure and its different cell types, along with physicians who identified that specific brain injuries led to localized impairments in abilities such as memory, speech, and imagination. These discoveries led to the understanding that the brain is composed of nerve cells that transmit electrical signals between them, and that these signals traveling between neurons in different brain regions form the basis of sensory and cognitive functions.
Vesaluis observed that the human brain shares structural similarities with other mammals but is more complex. Vesalius’s book, “De Humani Corporis Fabrica” (“On the Fabric of the Human Body”) | Shutterstock, lev radin
A Glimspe Inside The Brain
When studying the physiology of living humans, we often cannot - and would not want to - use invasive methods that could harm the subjects. This is why advanced imaging methods play a crucial role in brain research, allowing us to peer inside the brain without the need for a surgeon’s scalpel. Several noninvasive imaging methods exist for examining different organs, each suited to visualizing specific tissues and providing unique insights. For example, X-rays produce excellent images of bone tissue, while ultrasound waves are particularly effective for imaging soft tissues.
Another widely used imaging method, applied for example in diagnosing herniated disks in the spine and many other conditions, is Magnetic Resonance Imaging (MRI). An MRI machine functions as a massive magnet, with a basic imaging unit generating a magnetic field of up to 1.5 tesla—about 30,000 times stronger than Earth's magnetic field. It enables the production of detailed and precise anatomical images of tissues without exposing patients to harmful radiation, such as X-rays.
MRI images are based on how the protons in hydrogen atoms respond to the magnetic fields generated by the machine. Each tissue in the body has its own elemental composition that is unique to it, and so each tissue will appear in the scan in a different color, according to its features. For instance, fatty tissue is denser and more viscous than fluids, allowing for clear differentiation between various tissue types within the imaged organ.
Each tissue in the body has a unique elemental composition, and so each tissue will appear in the scan in a different color, according to its features. A patient inside an MRI machine | Credit: Shutterstock, zlikovec
While anatomical scans reveal the brain’s structure, they do not show what happens when the brain functions in real time or how its activity patterns change in different states—such as when we experience fear or retrieve facts from our memory. Many of the most intriguing questions in neuroscience relate to behavior, particularly traits we consider uniquely human, such as the emotions evoked by music or the ability to feel empathy. These topics are often difficult, if not impossible, to study experimentally in living animals. If we could observe brain activity through the skull and track what happens while an individual thinks, makes decisions, or experiences emotions, we would gain profound insights into the workings of the mind.
The Magnet Within
EEG uses electrodes attached to the skull to monitor the brain's electrical activity. A doctor monitoring a patient’s brain activity using EEG | Shutterstock, Roman Zaiets
A breakthrough in brain imaging occurred in 1990 when a group of researchers from AT&T Bell Laboratories discovered that the properties of red blood cells could be used in MRI scans to track changes in brain activity during cognitive processes. When a neuron is active, it consumes energy and requires oxygen. Oxygen is carried in the blood by red blood cells, which owe their red color to the protein hemoglobin, responsible for binding oxygen. As oxygen reaches the neurons, it is released from the red blood cells, which in turn bind carbon dioxide molecules. The carbon dioxide is then transported through the bloodstream to the lungs for removal.
Remember that an MRI machine is essentially a giant magnet? Hemoglobin has another crucial property: it contains an iron atom at its core. When oxygen detaches from hemoglobin, the exposed iron makes the molecule paramagnetic, meaning it responds to magnetic fields. This is similar to how an iron nail is pulled toward a magnet - the hemoglobin without oxygen causes a warp in the magnetic field. In contrast, when hemoglobin is bound to oxygen, its shape changes, making it diamagnetic, meaning it repels magnetic fields. By tracking these changes in the magnetic field, researchers can determine hemoglobin’s oxidation state—whether it is carrying oxygen to the cells or has already released it. The researchers named the measured signal Blood Oxygenation Level Dependent (BOLD).
The researchers identified the measured signal as Blood Oxygenation Level Dependent (BOLD). This MRI image of a brain highlights brown voxels in the occipital lobe, indicating significant BOLD signal changes under treatment conditions compared to control measurements. | Living Art Enterprises / Science Photo Library
Active As A Dead Salmon
This discovery paved the way for further studies that examine the changes in activity in different parts of the brain while a person is performing a task.The method was named fMRI - functional Magnetic Resonance Imaging. The scientific use of fMRI is increasingly growing: the number of scientific publications that are based on fMRI findings leaped from about 4,000 articles in 1990 to about 44 thousand articles in 2021. And so the method is currently a central scientific tool in neuroscience, moreover in studies of humans and advanced thought processes such as awareness, free will or empathy, which are difficult to find parallels for in the animal kingdom.
However, in 2009, a young researcher named Craig Bennett and his colleagues published a study that challenged the widespread trust in the method and the findings of studies based on it. Bennet recounted that during his PhD studies, he conducted fMRI experiments on adolescents and adults to examine changes in decision-making processes during adolescence. Due to the high cost of maintaining an MRI machine, most laboratories rent usage hours from hospitals or large research centers. Bennett did the same, and when he found himself with unused "dead hours" without participants, he and his colleagues decided to make use of the valuable time by scanning the most unusual objects they could find.
And they went to the supermarket, purchased and subsequently scanned a pumpkin, a frozen chicken, and one day, even an entire frozen salmon. They ran the full experiment—performing both the anatomical scan, which provides detailed cross-sections of tissue, and the functional scan, which measures changes in the BOLD signal. Amused by the beautiful images they obtained, they saved the data and forgot about the joke.
fMRI has become a central research tool in neuroscience. An fMRI image of a brain at rest | Mark And Mary Stevens Neuroimaging And Informatics Institute / Science Photo Library
In 2008, while working as a postdoctoral researcher, Bennett collaborated with his advisor, George Wolford, on the issue of multiple comparisons correction. Since the brain has no "off switch" and is constantly active, its activity patterns can only be studied comparatively—for example, by analyzing brain activity during a task versus at rest.
To measure the BOLD signal, MRI scans divide recorded brain activity into numerous small cubes called voxels—three-dimensional pixels with volume. For each voxel, a statistical test is used to determine whether there was a change in signal during task performance by the subject, compared to the activity measured under control conditions.The problem arises when conducting multiple comparisons - as you would when simultaneously analyzing thousands of voxels in the brain - the likelihood of finding a result that supports a hypothesis (such as increased activity) purely by chance increases. This type of error is known as a “Type I error” or “false positive”, and neuroscientists have sought methods to correct for it and improve the accuracy of analyses performed on thousands of voxels.
Bennett had an idea—he proposed retrieving the data from the frozen salmon scan, which contained approximately 65,000 voxels, and searching for "active" voxels.If such voxels were found in a dead fish, it would be a striking demonstration of Type I errors in fMRI analysis and highlight the importance of correcting for multiple comparisons. What he expected to find was a random distribution of a few isolated voxels, but to his surprise he observed three distinct clusters of active voxels - a brain activity pattern that clearly should not have existed in the motionless brain of the frozen salmon.
Bennet soon presented his findings at a major international conference held that same year, after convincing the organizers that it was not a joke. The study sparked widespread reactions—some dismissed it as a silly prank, while others argued that there must have been a mistake in the data analysis process. Bennett realized that, despite concerns about how publishing a "ridiculous" study on a dead fish might affect his reputation as an early-career scientist, his findings were too important to ignore. In 2009 Bennet and his colleagues published their fMRI scan of the frozen salmon, igniting heated discussions in the international neuroscience community. The study even earned them an Ig Nobel Prize in 2012.
Bennet was surprised to find three clear clusters of active voxels - an activity pattern that should not occur in the brain of a dead salmon. The MRI scan of the frozen fish | Bennett, C. et al. JSUR (2010), CC BY 3.0
Conflicting Conclusions
One of the core principles of science is peer review—the process in which every research article submitted to a scientific journal undergoes multiple rounds of independent evaluation by experts. These reviewers assess the study’s methodology and findings, suggesting refinements or even additional experiments to strengthen the conclusions. Bennet’s dead salmon serves a powerful demonstration of how scientists learn from errors and refine their methods over time
In another study that was published in Nature, neuroscientist Rotem Botvinik-Nezer and her colleagues from Tel Aviv University, set out to investigate how different fMRI data analysis methods might influence research findings and final conclusions. To that end, they asked seventy different laboratories from across the globe to analyse the same fMRI dataset, which contained brain activity collected from multiple subjects. Along with analyzing the data, the laboratories were asked to answer several questions, such as whether the brain activity increased or decreased under specific conditions.
At first glance, this may seem like a simple question. However, the study revealed that laboratories using different approaches and analytical tools have reached varying, and sometimes even contradictory, conclusions. Some teams found that brain activity had increased under certain conditions, while others concluded that it had actually decreased.
On other aspects, such as which brain regions were active during the test, there was greater agreement among the different research groups. These findings sparked a lively scientific debate, highlighting the fact that fMRI data analysis is a multi-step process requiring numerous decisions regarding calculations, use of software and the statistical tests used to evaluate the data. These choices can significantly influence the results and conclusions.
So how can researchers deal with this uncertainty? One key approach is replication, meaning the reproduction of studies conducted by other researchers in the hope of obtaining similar results. To facilitate effective replication, scientific journals typically require authors to provide a detailed description of all experimental and data analysis steps. Additionally, researchers are increasingly asked to openly share their raw data exactly as collected by measurement instruments, prior to processing. This openness enables other scientists to independently repeat the data analysis steps and verify if they reach the same conclusions as the original study. The trend toward providing open access to raw data is steadily growing, to the point that some scientific journals now make it a requirement for publication.
Laboratories using different approaches and analytical tools reached varying—and sometimes conflicting—conclusions. Scientists analyzing an fMRI image | Credit: Shutterstock, Gorodenkoff
The Demise of DIANA
As mentioned earlier, the signal measured fMRI reflects changes in the ratio between oxygenated and deoxygenated blood, making it an indirect measure of neural activity. If I decide to lift my hand, the neural signal from my brain to my hand travels within a fraction of a second. In contrast, the arrival of blood to a specific brain region depends on changes in blood vessel diameter—a much slower process that can take up to 20 seconds. So how can such a sluggish indicator be used for measuring neural activity, which occurs at a vastly faster timescale?
This is one of the limitations that fMRI researchers face. To study brain activity—such as what happens when a participant sees an image of a cat—the image must be displayed for several seconds to allow enough time for blood to flow to the active brain region and alter the BOLD signal. However, if the participant stares at the same cat for too long, they might get bored, and instead of identifying brain areas activated by viewing the cat, the scan might capture regions associated with mind-wandering. For this reason, experiments typically involve multiple sets of images, each shown for only a few seconds, repeated several times. While this approach allows researchers to track changes in the BOLD signal, it does not perfectly reflect how the brain processes information in real-life situations.
For each voxel, researchers examine whether the signal changed during a task compared to its baseline level under control conditions.Brain scan highlighting voxels that increase in activity (warm colors) and those that decrease (cool colors) when participants empathize with others’ emotions. | Image by Dr. Adi Yaniv
In 2022 a research group from South Korea published a study in Science, claiming that they had successfully developed an fMRI scanning technique capable of detecting brain activity at a millisecond - a thousandth of a second - resolution. The method, called DIANA (Direct Imaging of Neuronal Activity), generated enormous excitement. It promised to combine the wealth of spatial data obtained by MRI with ultrafast temporal precision matching the speed of real-time neural processes.
DIANA involved delivering electrical stimuli to an anesthetized animal at a rate of five pulses per second, while the MRI scanner sampled activity at specific points in the brain every five milliseconds. The collected data was then compiled to create a comprehensive view of the brain’s response over time. The researchers claimed that this technique suppressed the slow BOLD response, enabling the MRI to detect electrical potential differences associated with neural activity rather than blood flow.
Unfortunately, the excitement soon turned to disappointment. A year later, the editors of Science issued a warning to readers, stating that no follow-up studies had successfully replicated the results. One attempt to use the method failed outright, while another research group managed to measure a similar electrical signal to the one reported in the original DIANA study—but concluded that it was not directly linked to neural activity. They suggested it was a byproduct of the electrical stimulation pattern applied to the animal, rather than the brain’s intrinsic activity. An attempt was also made to apply the method to human subjects, but the researchers noted that part of the recorded signal might have resulted from the experimental conditions, as well as the equipment and techniques used for stimulation and measurement. They argued that more controlled studies in animal models are necessary to determine what the measured signal actually represents and whether it remains consistent under different conditions.
Capable of scanning at a resolution of 0.2 mm and completing a full brain scan in under four minutes. MRI brain scan at an intensity of 11.7 tesla (right) compared to other imaging devices | © CEA
Facing The Future
The tools we use in studying the living brain continue to improve. Among these innovations is the increasing use of more powerful MRI machines that provide exceptionally detailed anatomical scans. The most powerful MRI device is currently located in France and reaches a magnetic field intensity of 11.7 tesla. It can scan at a resolution of 0.2 mm and complete a full scan of all brain slices in just four minutes. For comparison, most hospitals in Israel use 1.5-tesla MRI machines, which are sufficient for routine medical needs. For brain research, 3-tesla MRI scanners are commonly used, and the Weizmann Institute of Science even has one that reaches 7 tesla.
However, higher magnetic field strength comes at a cost—subjects may experience nausea and dizziness, limiting the possible duration of experiments. Additionally, maintaining such a machine is complex and expensive, as the coils generating the magnetic field require cooling to near absolute zero. The cooling is required to minimize resistance, enabling the passage of the electric current that generates the magnetic field. Thus, despite its advantages, an especially powerful MRI device is not necessarily the optimal choice for an experiment and additional factors should be taken into account.
Efficient research and data analysis tools, combined with transparent data-sharing practices—allowing open access to raw data and peer review—can significantly deepen our understanding of ourselves and the most fundamental question of all: what shapes us and enables us to plan, imagine, create, remember, and collaborate? Perhaps in the not-so-distant future we will be able to grasp the extent of complexity of brain activity that allows me to write this article - gathering information from multiple sources and assembling letters into words—and enables you to read, understand, and (hopefully) remember it. And who knows? One day down the road, you might retrieve this knowledge from memory and apply it in new and unexpected ways.