Estimating Gender and Age from Brain Structural MRI of Children and Adolescents: A 3D Convolutional Neural Network Multitask Learning Model
Table 2
Performance metrics of the test procedure.
Regression models
n
MAE, y
r
value
R2-scr
Age: ABIDE-II 10-fold CV on test set
58
1.63 ± 0.28
0.76 ± 0.07
<0.001
0.54 ± 0.1
Age: ABIDE-II model on ADHD-200 full data
922
1.64
0.72
<0.001
0.50
Age: ADHD-200 10-fold CV on test set
92
1.43 ± 0.22
0.84 ± 0.04
<0.001
0.62 ± 0.08
Age: ADHD-200 model on ABIDE-II full data
580
1.57
0.75
<0.001
0.56
Classification models
n
Precision
Recall
F1-scr
AUC-ROC
Gender: ABIDE-II, 10-fold CV on test set
58
0.87 ± 0.06
0.80 ± 0.08
0.83 ± 0.04
0.82 ± 0.06
Gender: ABIDE-II model on ADHD-200 full data
922
0.76
0.80
0.78
0.79
Gender: ADHD-200, 10-fold CV on test set
92
0.84 ± 0.03
0.81 ± 0.06
0.83 ± 0.03
0.85 ± 0.04
Gender: ADHD-200 model on ABIDE-II full data
580
0.90
0.87
0.89
0.89
ASD: ABIDE-II, 10-fold CV on test set
58
0.46 ± 0.04
0.70 ± 0.18
0.55 ± 0.06
0.54 ± 0.06
ADHD: ADHD-200, 10-fold CV on test set
92
0.48 ± 0.07
0.55 ± 0.20
0.50 ± 0.11
0.61 ± 0.07
The performance indicators from 10-fold cross-validation are presented in their averaged values ± standard deviation. The chosen model for the cross-data set evaluation is the best-performing model of 10-fold cross-validation. For the column titles, r is the Pearson’s correlation between predicted and target ages, n is the sample size, and R2-scr is the prediction R2 (also known as cross-validation R2 or q2). Values in bold are metrics of the best-performing trained models. ASD: autism spectrum disorder; ADHD: attention deficit hyperactivity disorder.
