Can brain scans predict if a person is feeling pain?
By analyzing patterns across multiple brain regions simultaneously, researchers could predict whether a person perceived a subtle stimulus as painful, even when the stimulus intensity remained constant.
Source
Decoding the perception of pain from fMRI using multivariate pattern analysis
What they did
The researchers scanned 16 healthy participants using functional magnetic resonance imaging while delivering weak, near-threshold laser pulses to their feet. Across 120 trials, participants reported whether each sensation was painful or not, while the physical intensity of the laser remained unchanged. The researchers then used multivariate pattern analysis to test how well activity across 26 pre-defined brain regions could decode these subjective pain decisions.
What they found
The machine-learning model successfully predicted subjective pain reports from whole-brain activity, achieving an accuracy of 57.6% during the period preceding the stimulus and 61.4% during the stimulus itself. Predictions were most accurate when combining information from a few key regions, such as the somatosensory cortex and insula, rather than relying on a single region or single voxels. This suggests that pain perception is encoded in a widely distributed network across the brain.
The limits
What it doesn't show
The study's predictive model is highly personalized, meaning it cannot currently be used to diagnose or measure pain in a new, untested individual. The small sample of 16 healthy, young participants restricts how well these findings generalize to broader populations or people suffering from chronic pain conditions. Additionally, because the overall classification accuracies were relatively low, this method is not yet accurate enough for practical real-world applications, such as legal or clinical diagnostics.
Key terms
- Multivariate Pattern Analysis
- A machine-learning method that analyzes patterns of activity across multiple voxels simultaneously to decode a participant's mental state.
- Voxel
- A small, three-dimensional unit of volume representing a specific location in a brain scan, analogous to a pixel in a flat image.
- Support Vector Machine
- A classification algorithm used in machine learning to find a boundary that separates data into distinct categories.
- Near-threshold stimulus
- A sensory input calibrated to be right on the edge of detection, meaning it is perceived only about half of the time.
- Region of interest
- A specific, predefined anatomical area of the brain selected for targeted analysis in a neuroimaging study.
- Pain Matrix
- A network of interconnected brain regions, including sensory and emotional areas, that is consistently active during the experience of pain.
Flashcards
Want these cards to stick?
Save the deck to NoteFren and study it with spaced repetition.
Quiz yourself
What is a major difference between univariate encoding models and multivariate decoding models regarding spatial information?
Common questions
Why did the researchers use constant-intensity stimuli?
By keeping the laser's physical intensity constant, any differences in brain activity and reported pain must reflect the participant's purely subjective experience rather than physical differences in the stimulus.
Why wasn't the prediction accuracy perfect?
Subjective pain is highly complex, and fMRI scans contain noise and limitations in spatial and temporal resolution that prevent perfect decoding of such subtle mental states.
Can this technology be used in court to prove someone is in pain?
No, the authors warn that the accuracies are too low for legal diagnostics, and the model was trained on each individual's own unique brain patterns rather than a general population standard.
What is the benefit of combining brain regions for decoding?
Pain involves sensory, cognitive, and emotional processing, meaning its representation is distributed; looking at combinations of regions captures this network-level information better than looking at any single region alone.
More on Perceptual decision