Table 2.

Optimized parameters and prediction accuracy of SVR-LSM and SVR-MLSM models.

	Model	Baseline MoCA			Year 1 MoCA
		Prediction accuracy	p	Parameters	Prediction accuracy	p	Parameters
SVR-LSM with AIL	Linear - no volume control	0.4062	<0.001	c = 2−13	0.3788	<0.001	c = 2−10
	Linear - voxelwise normalization	0.2813	0.014	c = 4	0.3418	0.003	c = 2
	Linear - total volume regressed out*	0.2753	0.016	c = 2−11	0.1642	0.156	c = 2−3
	Nonlinear - voxelwise normalization	0.3715	<0.001	γ = 8, c = 2−20	0.3648	0.001	γ = 8, c = 2−20
MLSM with AIL and WMH	Linear - no volume control	0.3097	0.007	c = 2−17	0.4033	<0.001	c = 2−12
	Linear - voxelwise normalization	0.2509	0.029	c = 0.25	0.2899	0.011	c = 0.5
	Linear - total volumes regressed out∧	0.2538	0.027	c = 2−16	0.1818	0.116	c = 2−16

^*The total lesion burden of AIL was regressed out from the baseline and year 1 MoCA in this SVR-LSM model.

^∧The total volume of AIL and that of WMH were regressed out from the baseline and year 1 MoCA in this SVR-MLSM model.