Questions: fMRI Principles and Interpretation

Question 1

A neuroscience study reports that the dorsolateral prefrontal cortex (dlPFC) activates during a demanding working memory task. A journalist writes that 'the dlPFC is the brain's working memory center.' What is wrong with this conclusion?

• A. The journalist is correct — if dlPFC activates during working memory, it performs working memory
• B. dlPFC activation correlates with the task but does not establish that dlPFC is causally necessary, nor that working memory is its unique function
• C. fMRI spatial resolution is too poor to localize activity to the dlPFC specifically
• D. The conclusion would only be wrong if the study used an inappropriate baseline condition

Answer: (B)

This is the reverse inference problem: dlPFC is recruited by many demanding cognitive processes (attention, inhibitory control, task switching), so its activation during working memory tells you something demanding is happening but does not uniquely identify working memory as the function. More critically, correlation between BOLD signal and a task never establishes causal necessity — for that, you need TMS or lesion studies that disrupt the region and test whether behavior degrades. The journalist's claim commits both errors.

Question 2

An fMRI study compares brain activity 200ms vs 800ms after stimulus onset. A colleague says this design will capture the fine-grained temporal dynamics of early perceptual processing. Why is this claim problematic?

• A. It is correct — fMRI's 2-3mm spatial resolution is insufficient for temporal comparisons
• B. The hemodynamic response peaks 5-6 seconds after neural events, so BOLD signals from 200ms and 800ms post-stimulus will both reflect nearly the same sluggish response
• C. fMRI cannot be used with such short interstimulus intervals due to scanner noise
• D. The GLM cannot model conditions separated by less than 1 second

Answer: (B)

fMRI's critical limitation here is temporal, not spatial. The hemodynamic response function (HRF) peaks roughly 5-6 seconds after neural activity and takes ~20 seconds to return to baseline. Neural events at 200ms and 800ms post-stimulus will produce nearly identical BOLD responses — the HRF blurs them together. To resolve dynamics at the millisecond-to-second scale, you need EEG or MEG, which measure neural signals directly rather than the delayed hemodynamic proxy.

Question 3

fMRI does not directly measure neural firing; it measures blood oxygen changes that serve as a delayed, indirect proxy for neural activity.

• T. True
• F. False

Answer: True

This is the foundational interpretive fact about fMRI. The BOLD signal reflects the hemodynamic response — increased blood flow delivering oxygenated hemoglobin to active regions — rather than action potentials or synaptic activity directly. Because deoxyhemoglobin (paramagnetic) and oxyhemoglobin (not paramagnetic) affect the MRI signal differently, the scanner detects blood oxygen levels, not electrical neural events. This indirect measurement is why the HRF peaks 5-6 seconds after the underlying neural activity.

Question 4

If a brain region shows significantly elevated BOLD activation during a cognitive task, this establishes that the region is causally necessary for performing that task.

• T. True
• F. False

Answer: False

Activation establishes correlation, not necessity. A region can be active during a task as a bystander — monitoring, error-checking, or as part of a network incidentally engaged — without being computationally necessary for the task. The gold standard for causal necessity is TMS (transcranial magnetic stimulation), which can temporarily disrupt a specific region; if behavior degrades, the region is necessary. fMRI tells you where to look; TMS tests whether that region is doing required work.

Question 5

Why is the 'reverse inference' problem considered a fundamental interpretive limitation of fMRI, and under what condition does it become more or less valid?

Model answer: Reverse inference means inferring a cognitive process from a brain region's activation. It is problematic because most regions are not functionally selective — dlPFC, for example, is recruited by attention, inhibitory control, working memory, and task-switching, so its activation does not uniquely diagnose which process is occurring. Reverse inference is more valid when a region is known to be highly selective (e.g., a region reliably activated only by face perception); it is less valid for multi-function regions where the same activation could reflect multiple processes.

The problem is essentially a base-rate or prior-probability issue. If a region is recruited by 20 different cognitive processes, seeing it activate during a task gives weak evidence about which of those 20 is occurring. The inference becomes stronger as selectivity increases: if a region is activated 95% of the time by process X and rarely by anything else, observing activation is informative. The lesson is that fMRI activation maps are descriptive — they tell you what regions are engaged — but cognitive-process labeling requires converging evidence from behavioral studies, lesion data, or TMS.