Multivariate pattern classification of pediatric Tourette syndrome using functional connectivity MRI

Deanna J. Greene, Jessica A. Church, Nico U.F. Dosenbach, Ashley N. Nielsen, Babatunde Adeyemo, Binyam Nardos, Steven E. Petersen, Kevin J. Black, Bradley L. Schlaggar
  • Developmental Science, February 2016, Wiley
  • DOI: 10.1111/desc.12407

Machine learning "diagnosis" of Tourette syndrome by resting brain activity

What is it about?

We applied a multivariate method–support vector machine (SVM) classification–to patterns in brain network activity, measured with resting state functional connectivity (RSFC) MRI in 42 children with Tourette syndrome (age 8–15 yrs) and 42 tic-free controls matched for age, IQ, and head movement during the scan. Strict methods were used to remove data potentially compromised by head movement in the scanner. Univariate tests at single voxels (volume elements) across the brain identified no significant group differences. By contrast, SVM correctly classified group membership in ~70% of subjects, a degree of accuracy that is very unlikely to happen by chance (p < .001).

Why is it important?

Our results support the contention that multivariate data analysis methods can better capture the complexity of some brain disorders, and hold promise for predicting prognosis and treatment outcome for individuals with TS. We also report a novel adaptation of SVM binary classification that provides a confidence measure for the classification of each individual.

Perspectives

Dr Kevin J. Black
Washington University in St. Louis

Think about QR codes, those little squares with black and white patches that your smartphone can read. The pattern is too complex for you to interpret with your naked eye. You can't just look at a bunch of QR codes and sort them into two piles, one for ads and one for business cards. And if you did sort them, you probably couldn't say, "Aha! The 'business card' spot is this little square here!" But the information is there. This study was a little like that. Brain activity networks in kids with Tourette syndrome are different from brain activity networks in kids without tics, but there's no one pretty "hot spot" to look at to tell who's who. But if you feed the data to a boredom-resistant, error-resistant computer program that is designed to look at big patterns, it can find patterns that let it sort the kids into two groups. The sorting is far from perfect, but so much better than chance that the computer must be onto something.

Read Publication

http://dx.doi.org/10.1111/desc.12407

The following have contributed to this page: Dr Kevin J. Black