Cognitive Substraction: Why Brain Imaging Techniques aren’t always accurate

In imaging, there are certain methodologies, like the cognitive subtraction methodology. In this method, activity in a control task in subtracted from activity in an experimental task. So for example, take a word task. A simple model of written word recognition is used. In a famous experiment, the Peterson et al. (1988) experiment, they wanted to identify brain regions involved with 1) recognizing written words, 2) saying the words, and 3_ retrieving meaning of the words. The researchers used cognitive subtraction to tease out all the things they were testing.

So, to work out which regions are involved with recognizing written words, the researchers compared brain activity while subjects passively viewed words versus passively viewing a cross (+). The idea behind this was that the same brain regions, the same visual processing is involved in passively viewing things. But, the experimental task involved word recognition visually, and therefore, subtraction could be used to tease out the brain regions involved.

To work out which regions are involved in spoken words, the researchers compared the viewing of written words with reading a word aloud. In this task, both experimental and baseline involved visual processing of the word, and word recognition, and therefore subtracting should cancel out things, but the experimental task would be able to be analyzed afterwards.

To work out which regions are involved in retrieving the meaning of written words, the researchers compared a verb-generation task with reading aloud.

What the researchers found was that the left lateral hemisphere is involved in these processes. Recognizing written words activated bilateral sites in the visual cortex. Producing speech activates the sensorimotor cortex bilaterally. Verb generations activates the left inferior frontal gyrus.

Of course, there are some issues with cognitive subtraction. Can you think of any?

For example, let’s consider the subtraction behind determining which brain regions are associate with written word recognition. The assumptions was that both tasks involve visual processing but the experimental task involves the component of word recognition. Therefore, there is the assumptions that adding an extra component does not affect the operation of earlier ones in the sequence. This is referred to the as the assumption of pure insertion, or pure deletion. It could be that the amount, type, etc of visual processing that deals with written words is NOT the same as for the visual processing that deals with non-linguistic visuals. The added extra component in the tasks has the potential to change the operation of other components in the task. That is, there could be interactions (the effect of one variable upon another) that make the imaging data ambiguous.

PET scans versus fMRI

We have discussed PET scans and fMRI. Now the question is, which one of the two is better?


Over the last decade or so, fMRI has been more widely used than PET scans. This is because fMRI has better temporal and spatial resolution compared to PET scans. Also, fMRI does not use radioactivity, while PET scans do. However, fMRI still retains some disadvantages in comparison to PET scans.

The MRI scanner, which fMRI uses, is noisy, which of course means it can serve as a distraction to a subject, and also cannot be used to study audition. Another difficulty is that with fMRI, small movements can distort the signal being measured. You cannot speak while in the scanner, for example. Also, some brain regions are susceptible to signal distortion because nearby tissue has different magnetic properties (for example, sinuses, oral cavity, ear canes, air voids, essentially). Therefore, this makes certain regions, like the temporal region, and the orbitofrontal cortex, hard to image in fMRI. Also, PET, but not fMRI, can be used to trace pathways of chemicals.

Brain Imaging Alphabetical Soup: Making Sense of CAT, PET, MRI, fMRI, SPECT: fMRI

This article will focus on fMRI imaging.

fMRI, or functional magnetic resonance imaging, uses MRI equipment. fMRI is sensitive to the amount of deoxyhemoglobin in the blood. We know that when neurons consume oxygen, the neurons convert oxyhemoglobin to deoxyhemoglobin. Now, deoxyhemoglobin has magnetic properties, and these magnetic properties therefore introduce distortions in the local magnetic field. This distortion can be measured, and the measurement therefore gives an indication of how much deoxyhemoglobin in is in the blood. This technique is termed BOLD, for blood oxygen-level-dependent contrast.

The BOLD signal evolves over time in response to an increase in neural activity. This is called the hemodynamic response time, or HRF. HRF has three phases:


  1. The initial dip: as neurons consume oxygen, there is a small rise in deoxyhemoglobin, which results in a reduction of the BOLD signal.
  2. Overcompensation: in response to the increased consumption of oxygen, blood to a brain region increases. The increase in blood flow is greater than the increase in blood consumption, and therefore, the BOLD signal increases. This is the component that the fMRI normally measures.
  3. Undershoot: Blood flow and oxygen consumption dip before returning back to original levels.


The temporal resolution of fMRI is several seconds. This is better than PET scans, but it is still slow compared to the speeds at which cognitive processes take place.

Brain Imaging Alphabetical Soup: Making Sense of CAT, PET, MRI, fMRI, SPECT: PET scans

Structural imaging measures permanency. In other words, it measures the permanent, structural characteristics of the brain. Functional imaging measures more of the variability, the moment-to-moment changes of the brain that are associated with various cognitive processes, such as attention.

We know the brain consumes about 20% of the body’s total energy. A lot of this energy consumption is due to the maintenance of the membrane potential, of ion channels/gates. Now, most of the energy the brain receives is due to the vast vasculature of the brain. So, when the brain’s metabolic needs increases, you can see the increase of the blood supply to that specific region of the brain that is being most used in a given task.

PET scans are used in these situations, since these sorts of scans measure blood flow to a region directly. Other techniques use oxygenation to a brain region via fMRI.

This article will focus on PET scans.

PET, or positron emission tomography, uses radioactive tracers that have been injected into a subject’s bloodstream. The greater the blood flow to a brain region, the greater the signal emitted by the tracer in that specific region. Most commonly used tracers are oxygen-15 and fluorine-18. Oxygen-15 is usually given in the form of water, while fluorine-18 is usually given in the form of glucose.

Interestingly, though, other tracers can be used, such as radioactive neurotransmitters, which are used more commonly to study neural pathways and to study drug effects on the brain (which makes sense, considering drugs alter neurochemistry, which is based on neurotransmitters release and reuptake).

Now, with PET scans, it takes 30 seconds for the tracer to enter the brain, and then another 30 seconds for the radioactivity to reach its peak. See a problem with this?

These two 30-second windows are crucial, because in these windows, there are changes in blood flow related to cognitive tasks. Therefore, the temporal resolution (which is the accuracy that you can measure WHEN a cognitive activity is taking place) of a PET scan is about 30 seconds, which makes it unviable to use in the timing of things.

Regardless, this is how a PET scan works:

The tracer is injected into the bloodstream and it converted back from the unstable radioactive form to the stable form. As it does this, it emits a positron that ends up colliding with an electron. This releases two photons that can then be detected by machinery. A spatial image of the brain is thus constructed.

Brain Imaging Alphabetical Soup: Making Sense of CAT, PET, MRI, fMRI, SPECT: MRI

I started this mini-series with the articles on how imaging got started, and on CT scans. Now we’ll move on to MRI scans.

MRI, or magnetic resonance imaging, is used to create images of the body’s tissues, specifically of the soft tissues, like organs. X-rays pass through soft tissues undistorted and relatively easily. Now, most human tissue is water-based (which makes sense considering we’re 70% water).  The amount of water in tissues differ. So different tissues will behave differently. These differences can then be used to construct a 3D image.

MRI scans are constructed by applying a strong magnetic field around the body. In water, you can find single protons (Hydrogen atoms, like H20). In MRI scans, it is the H nuclei that create the signal.

So, initially, the fields are randomly oriented. When a strong external magnetic field is applied, a fraction will organize themselves, and align themselves with the external field. This external field is applied in a constant manner throughout the scanning session.

Now, when the protons are aligned, a brief radio frequency is applied. This radio frequency then knocks the protons out of alignment by 90 degrees to their original orientation. So now the protons are spinning in this new state, and as they do this, they produce a detectable change in the magnetic field. This then is what forms the MRI signal.

Eventually, the protons are pulled back into their original alignment (they “relax”). The MRI scanner then repeats the process by sending the radio frequency to excite the protons in different slices of the brain that is being scanned.

Now there are different types of MRI scans, including the T1 and T2 scans.

T1-weighted images are used for structural imaging of the brain. In a T1-weighted image, gray matter looks gray, and white matter looks white. Pretty simple to distinguish.

When the protons are misaligned at 90 degrees to the magnetic field, the MRI signal decays because of interactions with other nearby molecules. This is the T2 image.




Brain Imaging Alphabetical Soup: Making Sense of CAT, PET, MRI, fMRI, SPECT: CAT Scans

We’ve all heard the abbreviations: CAT, PET, MRI, fMRI, SPECT.

Now what the heck do they all mean?

Brain imaging has an interesting start. But generally, brain imaging techniques fall into two broad categories: Structural imaging, and Functional imaging. Structural imaging does just what its name implies: it provides images of the structure of the brain. There’s no indication of blood flow or brain metabolism here; just bone and tissue. Structural images are based on the idea that different bodily tissues have different densities and different physical properties. Structural images provide static, one-shot images of the brain’s structure. They say nothing of function.

Functional imaging also does what its name implies: it images the brain as it functions. Dyes, radioactive tracers, are all used in these imaging modalities to image the brain’s blood flow and metabolism. These images are dynamic, showing changes in brain function over a period of time.

The next few articles in my blog will feature one of the imaging techniques in an article. We’ll start off with CAT scans.

CAT stands for Computerized Axial Tomography. CAT scans are also abbreviated CT scans, for Computerized Tomography. CT scans are possible because different tissues have different X-ray absorptions. How much a tissue absorbs X-rays is related to that tissue’s density. So, bone absorbs the most X-rays, cerebrospinal fluid (CSF) absorbs the least. That’s why on CT scans, bone appears white and the ventricles, which contain CSF, appear black.

There’s a slight danger with CT scans, because the person being imaged is exposed to a slight amount of radiation.


CT scan image can be found here: