Fig. 8
From: Mouse MRI shows brain areas relatively larger in males emerge before those larger in females
Prediction and individualization of neuroanatomical structure volumes. a Predicted and Observed p36 structure volumes for all subjects and three representative structures: primary motor cortex, striatum, and the barrel cortex. When predicting any subject at p36, the model was trained on all data excluding the predicted subject at p36 and all subjects at p65. Despite this exclusion, the model uses neuroanatomical information from earlier time points to predict structure volumes’ variation from the average (horizontal line). b Matrix plotting the root-mean-square-difference (RMSD) between the observed structure volumes (rows) and the predicted structure volumes (columns) for each subject at p36. Prediction used all data prior to and including p29. Each column (Subject X prediction) was centered so that the diagonal RMSD (observation vs prediction for Subject X) is 0. Blue or red off-diagonal cells (observation Subject Y vs prediction Subject X) indicate RMSD greater or less than diagonal entry. Density plot shows that diagonal RMSD elements (values as points and median as vertical line) tend to be less than off-diagonal RMSD (grey distribution), indicating high prediction specificity (One-sample Kolmogorov–Smirnov test: \(D^ + = 0.79894,P < 10^{ - 15},n = 28\)). c Models were trained on data closer to the time point of prediction (p36). For example, models corresponding to 'p10' were trained on p10 data and all earlier data (p3, p5, p7). As more data is included, model accuracy and specificity improved with significant specificity at p10 and above (One-sample Kolmogorov–Smirnov test: \(D^ + = 0.34392,P = 0.001,n = 28\)). d Prediction accuracy (RMSD) at p36 for each subject tends to improve over the course of neurodevelopment. When x-axis = X, subject data from ages ≤ X were used in the prediction. Trendlines denote patterns seen by sex and there is no significant difference in prediction accuracy between the sexes at any time (with shaded regions denoting standard error). Improvement in prediction accuracy (*P<0.05, **P<0.01, ***P<0.001) indicates neuroanatomy individualization, which occurs significantly earlier in males (permutation test, P = 0.025)
