Bootstrapped permutation test for multiresponse inference on brain behavior associations

Abstract

Despite that diagnosis of neurological disorders commonly involvesa collection of behavioral assessments, most neuroimaging studies investigatingthe associations between brain and behavior largely analyze each behavioralmeasure in isolation. To jointly model multiple behavioral scores, sparse multiresponse regression (SMR) is often used. However, directly applying SMRwithout statistically controlling for false positives could result in many spuriousfindings. For models, such as SMR, where the distribution of the modelparameters is unknown, permutation test and stability selection are typicallyused to control for false positives. In this paper, we present another technique forinferring statistically significant features from models with unknown parameterdistribution. We refer to this technique as bootstrapped permutation test (BPT),which uses Studentized statistics to exploit the intuition that the variability inparameter estimates associated with relevant features would likely be higherwith responses permuted. On synthetic data, we show that BPT provides highersensitivity in identifying relevant features from the SMR model than permutation test and stability selection, while retaining strong control on the falsepositive rate. We further apply BPT to study the associations between brainconnectivity estimated from pseudo-rest fMRI data of 1139 fourteen year oldsand behavioral measures related to ADHD. Significant connections are foundbetween brain networks known to be implicated in the behavioral tasksinvolved. Moreover, we validate the identified connections byfitting a regression model on pseudo-rest data with only those connections and applying thismodel on resting state fMRI data of 337 left out subjects to predict theirbehavioral scores. The predicted scores significantly correlate with the actualscores, hence verifying the behavioral relevance of the found connections.