Transfer Learning Between fMRI-Based Predictors of Treatment Outcome with Psilocybin and Escitalopram in Patients with Major Depression

Major depression and other mood disorders often require pharmacological interventions, whose responses can vary considerably across patients, resulting in overly long and costly treatments. For this reason, there is great interest in predicting which patients will benefit the most from the treatment. In this study, we investigated baseline brain functional connectivity during rest, measured using functional magnetic resonance imaging, to predict the effectiveness of the treatment. Specifically, we propose a transfer learning scheme between two groups of independent experiments with different types of drugs: one is based on the SSRI drug escitalopram, while the other involves a novel therapy with the psychedelic compound psilocybin. We aimed to evaluate the capacity of the proposed method to predict the outcomes of each treatment. Our analysis showed that the connectivity of different large-scale functional networks predicted symptom improvement, particularly the resting-state networks related to visual and default-mode areas, which exhibited high accuracy both with individual cross-validation schemes and with transfer learning between both experimental groups. Additionally, it was observed that the connectivities with greater importance in the predictions maintained similar connectivity patterns between both independent groups. Our work highlights the value of noninvasive brain activity measurements for the prediction of treatment outcome, while also suggesting that certain functional connections support the correct prediction of both escitalopram and psilocybin treatment.