To explore if a DTI protocol could provide a model to predict the degree of vision recovery in NMOSDs patients. 37 patients were employed in the study, including 20 patients of well vision recovery and 17 patients of poor vision recovery. With the diffusion measure of multiple white and grey matters as features, a Lasso-Logistic regression model and a Support Vector Machine (SVM)-based classification model were constructed. The results show area under curve (AUC) of 0.7618 (P=0.008) and accuracy (ACC) of 0.7297 (0.006). The method shows promising prediction performance, and it has the potential to improve the clinical treatment design.