Fig. 1: Schematic representation of GMV-based predictive modeling for arithmetic abilities.

Step 1: we extracted GMV of each region of interest (ROI) in Destrieux atlas. The Destrieux atlas includes 148 ROI. Based on GMV data from 130 children, a GMV matrix of size 130 by 148 was constructed. Step 2: one participant was left out on each LOO iteration as the testing set, and the remaining participants were regarded as the training set. Partial correlations were calculated between the GMV of each ROI and the arithmetic skills, while controlling for age and estimated total intracranial vol (eTIV). The significantly arithmetic-correlated ROIs (P < 0.01) were included in a spatial profile to build the predictive model. Step 3: a multivariate linear regression model was trained to relate the GMV of each ROI to the arithmetic skills based on selected ROIs. Step 4: The model was applied to novel subjects in left-out participants to assess the generalizability.