The value of brain MRI functional connectivity data in a machine learning classifier for distinguishing migraine from persistent post-traumatic headache

Abstract

Read online

BackgroundPost-traumatic headache (PTH) and migraine often have similar phenotypes. The objective of this exploratory study was to develop classification models to differentiate persistent PTH (PPTH) from migraine using clinical data and magnetic resonance imaging (MRI) measures of brain structure and functional connectivity (fc).MethodsThirty-four individuals with migraine and 48 individuals with PPTH attributed to mild TBI were included. All individuals completed questionnaires assessing headache characteristics, mood, sensory hypersensitivities, and cognitive function and underwent brain structural and functional imaging during the same study visit. Clinical features, structural and functional resting-state measures were included as potential variables. Classifiers using ridge logistic regression of principal components were fit on the data. Average accuracy was calculated using leave-one-out cross-validation. Models were fit with and without fc data. The importance of specific variables to the classifier were examined.ResultsWith internal variable selection and principal components creation the average accuracy was 72% with fc data and 63.4% without fc data. This classifier with fc data identified individuals with PPTH and individuals with migraine with equal accuracy.ConclusionMultivariate models based on clinical characteristics, fc, and brain structural data accurately classify and differentiate PPTH vs. migraine suggesting differences in the neuromechanism and clinical features underlying both headache disorders.

Keywords

• post-traumatic headache (PTH)
• migraine
• fMRI
• classification
• machine learning