February 10, 2010 (San Antonio, Texas) — Functional magnetic resonance imaging (fMRI), coupled with a support vector machine (SVM) trained to discern patterns of activation, can distinguish whether a stimulus is painful or not for a given subject, according to a study from Stanford University, Palo Alto, California.
In this trial, the SVM accurately distinguished painful heat stimuli from nonpainful heat stimuli more than 80% of the time. Further, the stimuli identified as either painful or nonpainful were correctly labeled more than 85% of the time, according to results presented here at the American Academy of Pain Medicine 26th Annual Meeting.
With the reporting of pain being so subjective and patients' accounts sometimes being called into question, the researchers set out to test a means of objectively determining whether someone is experiencing pain or not by analyzing images from the brain.
Studies have identified regions of the brain that are more active during painful stimulation but might not be useful from patient to patient.
"A tool that measures pain in my brain might not objectively measure pain in someone else's brain," Neil Chatterjee, BS, who is about to enter an MD/PhD program, told Medscape Neurology. He worked on the study with Stanford University associate professor Sean Mackey, PhD, and another undergraduate, Justin Brown.
In this study, the researchers aimed to determine whether patterns of neural activity might provide a biomarker for the presence or absence of pain.
The study involved 16 healthy subjects. Each was given a thermal stimulus hot enough to be painful for 30 seconds. That was followed by a 40-second rest period of just background stimuli. Then they were each given a thermal stimulus that was not painful, again for 30 seconds.
Each stimulus was mapped using fMRI.
The subjects were divided into a training group and a testing group.
Then researchers turned to the SVM, a tool that can sort through large amounts of data and create a model.
The maps for the 8 subjects in the training group were entered into the SVM, which plotted the maps, creating a dividing line between the 2 conditions, in this case painful and not painful.
The maps of the other 8 subjects were then plotted, and depending on how the images were categorized, the SVM predicted whether the stimulus was painful or not.
The SVM accurately identified 83% of the painful stimuli and 90% of the nonpainful stimuli. Of the stimuli identified as painful, 90% actually were painful, and the negative predictive value was 85%.
Researchers also used SVM to test whether certain regions of the brain could independently identify painful or nonpainful stimuli. Only the secondary sensory cortex was able to do so with more than chance accuracy, whereas the glandular anterior cingulate cortex, the insular cortex, and the primary sensory cortex did not.
"Because we separated the group that we tested on from the group that we created the model with, we know that to at least some extent this detection can generalize across people," Mr. Chatterjee said. But he acknowledged the test did not control for cognitive factors, such as fear and anticipation, and only tested for 1 kind of painful stimulus.
An Echocardiogram for Pain?
Jeffrey Tiede, MD, of the Columbia Interventional Pain Center in Columbia, Missouri, and the moderator of the session, said the study was exciting.
"This hopefully someday will become our echocardiogram, and we can actually screen patients on how bad they hurt and if they actually hurt, where there's addictive behavior and things of that nature," he told the audience after the presentation.
The study was funded by the National Institutes of Health and the Chris Redlich Pain Research Fund. Mr. Chatterjee and Dr. Tiede have disclosed no relevant financial relationships.
American Academy of Pain Medicine (AAPN) 26th Annual Meeting: Abstract 179. Presented February 4, 2010.
Medscape Medical News © 2010 Medscape, LLC
Send press releases and comments to firstname.lastname@example.org.
Cite this: Functional MRI Combined With a Support Vector Machine May Discern Pain - Medscape - Feb 10, 2010.