TABLE 4.

Prediction performance of different regressors on different modal features

Features	Feature extraction	Features shape	Regressors	MAE (years)	Spearman's rank correlational analysis
MRI only	ResNet	256 × 1	Linear	4.91	.67
	ResNet	256 × 1	Support vector	5.49	.61
	ResNet	256 × 1	Random forest	4.88	.69
	ResNet	256 × 1	Lasso	5.12	.65
Blood parameters only	None	14 × 1	Linear	5.90	.53
	None	14 × 1	Support vector	5.81	.54
	None	14 × 1	Random forest	5.60	.57
	None	14 × 1	Lasso	6.10	.50
Bilinear fusion of MRI and blood parameters	ResNet + MCB	14 × 1	Linear	3.96	.76
	ResNet + MCB	14 × 1	Support vector	4.60	.74
	ResNet + MCB	14 × 1	Random forest	4.76	.72
	ResNet + MCB	14 × 1	Lasso	4.64	.71

Note: Bold values signifies that p‐value is <.001