TABLE V.

Adaptability of the LeqMod to different denoising baselines.

		GAN				SwinUNETR
		NRMSE (↓)		TLG bias (%) (o)		NRMSE (↓)		TLG bias (%) (o)
		w/o	w	w/o	w	w/o	w	w/o	w
Quadra	1%	0.273	0.266 †	−5.115	−4.444 †	0.294	0.284 †	−7.395	−5.304 †
	2%	0.252	0.234 †	−4.092	−3.376 †	0.240	0.231 †	−5.617	−3.889 †
	5%	0.196	0.194	−2.592	−2.328 †	0.207	0.216	−4.244	−2.980 †
	10%	0.178	0.169 †	−1.787	−1.583 †	0.184	0.180	−3.312	−2.133 †
	25%	0.146	0.131 †	−1.035	−0.879 †	0.173	0.137 †	−2.193	−1.163 †
	50%	0.100	0.094	−0.699	−0.610	0.140	0.114 †	−2.095	−0.964 †
uExplorer	1%	0.551	0.536 †	−13.231	−9.671 †	0.574	0.555 †	−15.354	−13.586 †
	2%	0.439	0.432 †	−8.601	−6.696 †	0.453	0.440 †	−11.286	−8.921 †
	5%	0.354	0.353 †	−4.869	−4.163 †	0.371	0.349 †	−7.069	−5.080 †
	10%	0.284	0.274 †	−3.285	−3.092 †	0.303	0.294 †	−5.371	−3.672 †
	25%	0.213	0.224	−1.800	−1.679 †	0.231	0.226 †	−3.721	−2.249 †
	50%	0.153	0.142 †	−1.016	−1.028	0.182	0.165 †	−2.805	−1.351 †
mCT	5%	0.538	0.527 †	−8.466	−7.428 †	0.531	0.524 †	−8.485	−7.507 †
	10%	0.523	0.449 †	−5.828	−5.056 †	0.459	0.430 †	−6.070	−4.990 †
	20%	0.344	0.339 †	−3.604	−3.037	0.361	0.348 †	−4.187	−3.712

^†P<0.05 based on the Wilcoxon signed-rank test between the w/o and w LeqMod groups.

(↓) and (o) represent expecting lower and zero-approaching values, respectively.