Fig. 1

Overall architecture of the proposed method

Fig. 2

The architecture of the convolutional neural network used in the proposed method. The inputs of the network are 12$T_{2}^{*}$-weighted images acquired at different TEs, and the outputs are 2 parameter maps corresponding to${{S}_{0}}$and$R_{2}^{*}$maps, respectively

Fig. 3

$R_{2}^{*}$ maps reconstructed by UNet-TVp with different λ values. The NRMSE values of $R_{2}^{*}$ maps are shown in the bottom right corner of the maps. The change of color in the color bar on the right side of the figure from blue to red represents the $R_{2}^{*}$ value from small to large (GT: ground truth)

Table 1

Quantitative results of different methods on the simulated testing datasets (mean ± standard deviation)

Fig. 4

$R^{*}_{2}$maps reconstructed by different methods for one simulated severe iron-loaded liver dataset and corresponding absolute difference maps (Difference) under each $R^{*}_{2}$ map. $T^{*}_{2}$ w: $T^{*}_{2}$-weighted images (TE1 = 0.93 ms, TE2 = 2.27 ms). GT: ground truth of $S_{0}$ and $R^{*}_{2}$ maps. The SSIM of each $R^{*}_{2}$ map is shown in its top right corner, and the NRMSE is shown in the top right corner of the corresponding absolute difference map. The change of color in the color bar on the right side of the figure from blue to red represents the $R^{*}_{2}$ value from small to large

Fig. 5

Bland-Altman analysis for the agreement between the mean $R^{*}_{2}$ values in liver parenchyma (excluding vasculatures) and the reference, and the $R^{*}_{2}$ maps reconstructed from different methods on the simulated testing datasets. The solid lines represent mean differences and the dashed lines indicate 95% confidence intervals

Fig. 6

Bland-Altman analysis for the agreement of the mean $R^{*}_{2}$ values in liver parenchyma (excluding vasculatures) with the reference, and the $R^{*}_{2}$ maps reconstructed from other methods on the clinical testing datasets. The PCANR algorithm was used as the reference method. The solid lines represent mean differences and the dashed lines indicate 95% confidence intervals

Fig. 7

$R^{*}_{2}$ maps estimated by different reconstruction methods for one representative clinical testing data, which has moderate hepatic iron overload. First row: $T^{*}_{2}$-weighted images (TE1 = 0.93 ms, TE2 = 2.27 ms, TE3 = 3.61 ms, TE4 = 4.95 ms). The mean $R^{*}_{2}$ value (s-1) in liver parenchyma (excluding vasculatures) is shown in the top right corner of each $R^{*}_{2}$ map. The change of color in the color bar on the right side of the figure from blue to red represents the $R^{*}_{2}$ value from small to large

Fig. 8

$R^{*}_{2}$ maps estimated by different reconstruction methods for one representative clinical testing data, which has severe hepatic iron overload. First row: $T^{*}_{2}$-weighted images (TE1 = 0.93 ms, TE2 = 2.27 ms, TE3 = 3.61 ms, TE4 = 4.95 ms). The mean $R^{*}_{2}$ value (s-1) in liver parenchyma (excluding vasculatures) is shown in the top right corner of each $R^{*}_{2}$ map. The change of color in the color bar on the right side of the figure from blue to red represents the $R^{*}_{2}$ value from small to large