Epileptic patients may suffer from severe brain damages during seizures. There is thus a significant need for automated seizure prediction. Independently, brain macroscopic activity has been shown to display scalefree temporal dynamics, which, in turn, were involved into seizure prediction. Selfsimilarity, the paradigm model for scalefree dynamics, has however mostly been defined in univariate settings, thus yielding a collection of independent analyses of recorded signals. Yet, nonnegligible correlations exist in multi-channel recordings of brain activity and may prove useful in seizure prediction. This work aims to assess the benefits of using a recently developed multivariate eigen-wavelet framework for multivariate selfsimilarity analysis in seizure prediction using CHB-MIT Scalp EEG data.