fMRI

Functional magnetic resonance imaging, fMRI for short (in German fMRT), is a brain imaging method. A person lies in a tube containing a very strong magnet. The device produces cross-sectional images of the inside of the head without any need for a surgical procedure. The addition of “functional” means: it is not just about the shape of the brain, but about what it is doing at that moment. During the measurement, the test subject solves tasks, looks at images, or listens to sentences. Afterward, it can be shown which regions of the brain received increased blood flow during these activities.

Why researchers want to look into the brain

Before fMRI existed, surprisingly little was known about the division of labor within the brain. Most knowledge came from accidents and illnesses. If someone could no longer speak after an injury, conclusions were drawn about the function of the damaged area. This was imprecise and ethically fraught, since one could only wait and see what happened. fMRI made it possible to observe healthy people during completely normal activities.

In medicine, the method is used today prior to brain tumor surgery. Surgeons check beforehand exactly where the speech center is located in this particular patient. This location varies from person to person by a few millimeters. A textbook atlas is therefore not sufficient.

For AI research, fMRI is interesting for a different reason. It provides enormous datasets on how a biological thinking system reacts to stimuli. This data serves as a benchmark for artificial neural networks — that is, for computer programs roughly modeled on nerve cells. One can test whether a language model processes sentences in a manner similar to a human.

Blood oxygen as a detour

fMRI does not measure brain activity directly. Nerve cells send electrical signals, but these can hardly be measured from outside. Instead, the device takes a detour: active nerve cells consume more oxygen. The body reacts and shortly afterward sends more oxygen-rich blood to that region.

This is exactly where physics helps. Blood with oxygen behaves differently in a magnetic field than blood without oxygen. The scanner detects this tiny difference and calculates an image from it. Experts call this signal BOLD, which stands for “blood-oxygen-level dependent.”

This detour comes at a cost. Blood flow only responds several seconds after the actual nerve activity. fMRI is therefore sluggish in terms of timing, but fairly precise spatially. A typical voxel covers about one cubic millimeter of tissue — and within it are hundreds of thousands of nerve cells. So one sees neighborhoods, not individual cells. A common misconception is therefore to mistake the colorful spots on fMRI images for direct photographs of thoughts. They are in fact statistical evaluations of differences in blood flow.

From the clinic to headlines about mind reading

In everyday life, one encounters fMRI mainly in hospitals and university clinics. A regular MRI examination of the knee or back uses the same machine, just without the functional component. A scan usually takes twenty to sixty minutes and is loud but painless.

In tech news, the method has been appearing regularly for a few years now. Research groups combine fMRI data with AI models and use it to reconstruct rough images or sentence content. The results sound spectacular but only work under narrow conditions. Each person must first be trained for hours, and anyone who doesn’t want to cooperate can easily disrupt the measurement.

For investors and observers of the tech industry, fMRI is mainly a niche topic with symbolic significance. It represents the connection between medical technology and machine learning. It is far from being portable: a scanner weighs several tons and costs millions. Anyone reading about brain-computer interfaces should therefore clearly distinguish fMRI from implants that sit directly inside the head.