Table 1.

Statistic analysis of the impact of TOI selection and multi-slice training for RED-CNN and MS-RDN architectures.

	RED-CNN	P-value	MS-RDN	P-value
(a) Impact of TOI selection
NMSE	$+0.090$ dB	$<0.0001$	$+0.253$ dB	$<0.0001$
Bias	$-0.998\times {10}^{-4}{\text{cm}}^{-1}$	$<0.0001$	$-2.099\times {10}^{-4}{\text{cm}}^{-1}$	$<0.0001$
PSNR	$+0.090$ dB	$<0.0001$	$+0.253$ dB	$<0.0001$
SSIM	$+0.0009$	$<0.0001$	$+0.0011$	$<0.0001$
(b) Impact of multi-slice training
NMSE	$+0.035$ dB	$<0.0001$	Not significant	0.211
Bias	$-0.411\times {10}^{-4}{\text{cm}}^{-1}$	$<0.0001$	Not significant	0.234
PSNR	$+0.035$ dB	$<0.0001$	Not significant	0.211
SSIM	$+0.0004$	$<0.0001$	$+0.0005$	$<0.0001$

The evaluation was performed on the entire testing breast dataset using NMSE, Bias, PSNR, and SSIM metrics.

(a) shows the performance improvement by including TOI selection on single-slice ($Z=1$) RED-CNN and MS-RDN.

(b) shows the performance difference between multi-slice ($Z=5$) and single-slice training for RED-CNN and MS-RDN, respectively. Please note that the values corresponding to NMSE and PSNR are identical since these quantities are related as shown in Eqs. (6) and (7).