Table 2.

Initial learning rates (LRs) for the contrast-specific baseline and subject-specific (SS) denoising models.

	Contrast	Initial learning rate
		Baseline	Subject-specific (SS)
1	T1	2.5 × 10−4	1 × 10−5, 1 × 10−6
2	T2	1 × 10−4	1 × 10−5
3	T2 FLAIR	1 × 10−4	1 × 10−4
4	${\text{T}}_2^{*}$	2.5 × 10−4	NA
5	DWI	1 × 10−4	1 × 10−5

The baseline denoising models were finetuned on Fastest data to obtain the SS denoising models. Therefore, the initial LRs were reduced to avoid harming the learned representations. For T₁, one subject required a LR of 1 × 10⁻⁶ since the loss values did not decrease with an LR of 1 × 10⁻⁵. For T₂ FLAIR, the LR value was not changed since the model did not otherwise converge to lower loss values.