Abstract
Neuroimaging studies have identified a large number of biomarkers associated with schizophrenia (SZ), but there is still a lack of biomarkers that can predict the efficacy of antipsychotic medication in SZ patients. The aim of this study was to identify neuroimaging biomarkers of antipsychotic drug response among features of the resting-state connectome. Resting-state functional magnetic resonance scans were acquired from a discovery cohort of 105 patients with SZ at baseline and after 8 weeks of antipsychotic medication treatment. Baseline clinical status and post-treatment outcome were assessed using the Positive and Negative Symptom Scale (PANSS), and clinical improvement was rated by the total score reduction. Based on acquired imaging data, a resting-state functional connectivity matrix was constructed for each patient, and a connectome-based predictive model was subsequently established and trained to predict individual PANSS total score reduction. Model performance was assessed by calculating Pearson correlation coefficients between predicted and true score reduction with leave-one-out cross-validation. Finally, the generalizability of the model was tested using an independent validation cohort of 52 SZ patients. The model incorporating resting-state connectome characteristics predicted individual treatment outcomes in both the discovery cohort (prediction vs. truth r = 0.59, mean squared error (MSE) = 0.021) and validation cohort (r = 0.41, MSE = 0.036). The model identified four positive features and eight negative features, which were respectively correlated positively and negatively with PANSS total score reduction. Among these positive features, the specific connections within the parietal lobe played a crucial role in the model's predictive performance. As for the negative features, they included the frontoparietal control network and the cerebello-thalamo-cortical connections. This study discovered and validated a set of functional features based on resting-state connectome, where higher connectivity of positive features and lower connectivity of negative features at baseline were associated with a higher reduction rate of PANSS total score in patients and a better therapeutic effect. These functional features can be used to predict the PANSS total score reduction rate of SZ patients through a model. Clinical doctors can potentially infer the individual treatment response of antipsychotic medication treatment for patients based on the predicted results.