A hypothesis-driven method based on machine learning for neuroimaging data analysis

There remains an open question about the usefulness and interpretation of Machine learning (MLE) approaches for discrimination spatial patterns brain images between samples or activation states. In last few decades, these have limited their operation to feature extraction linear classification tasks between-group inference. this context, statistical inference is assessed by randomly permuting image labels use random effect models that consider between-subject variability. These multivariate MLE-based pipelines, whilst potentially more effective detecting activations than hypotheses-driven methods, lost mathematical elegance, ease interpretation, localization ubiquitous General Model (GLM). Recently, estimation conventional GLM has been demonstrated be connected univariate task when design matrix expressed as a binary indicator matrix. paper we explore complete connection MLE \emph{regressions}. To purpose derive refined test with based on parameters obtained Support Vector Regression (SVR) in \emph{inverse} problem (SVR-iGLM). Subsequently, field theory (RFT) employed assessing significance following benchmark. Experimental results demonstrate how parameter estimations derived from each model (mainly SVR) result different experimental estimates are significantly related predefined functional task. Moreover, using real data multisite initiative proposed demonstrates power control false positives, outperforming regular GLM.