Table 1.
Related studies and methodology of CT-/MRI-based bladder image segmentation during the past 20 years.
| Study | Imaging | Approach or strategy | Region focused | Performance and Merits |
|---|---|---|---|---|
| Li et al., 2004 (86) | Multispectral MRI | Partial volume (PV) scheme | IB | More information extracted from the multispectral images, and feasible for the IB. |
| Li et al., 2008 (85) | Multispectral MRI | Markov random field (MRF) | IB | Realizing the inhomogeneity correction and overcoming the influence of partial volume and bias field. |
| Duan et al., 2010 (80) | T1WI | Coupled level-sets | *IB/OB | Realizing the simultaneous extraction of both IB and OB of the bladder. |
| Garnier et al., 2011 (87) | T2WI | 3D deformable model based on active region growing strategy | IB/OB | Achieving good performance for the IB segmentation when tumors were not existed in the bladder lumen. |
| Duan et al., 2011 (78) | T1WI | Coupled level-sets + volume-based features | Tumor | Realizing the automatic detection of BCa. |
| Duan et al., 2012 (79) | T1WI | Coupled level-sets + volume-based features + Adaptive window-setting scheme | Tumor | Realizing the automatic detection and extraction of BCa. |
| Ma et al., 2011 (88) | T2WI | Geodesic active contour (GAC) + shape-guided Chan-Vese | IB/OB | Achieving good segmentation performance for both bladder borders without tumor regions using two datasets with 2D images. |
| Han et al., 2013 (89) | T1WI | Adaptive MRF with coupled level-set constraints | IB/OB | Fast convergence, robustness to initial estimates, and robustness against noise contaminations, as well as local shape variations of the bladder wall. |
| Qin et al., 2014 (77) | T2WI | Coupled directional level-sets with adaptive shape prior constraints | IB/OB | With the average DSC of 0.96 and 0.946, respectively, for the IB and OB segmentation using 11 datasets. |
| Cha et al., 2014 (90) | #CECT | Conjoint level set analysis and segmentation system (CLASS) | IB/OB | With the average DSC of 0.842 for the IB segmentation using 182 datasets. |
| Dolz et al., 2018 (83) | T2WI | Progressive dilated convolution-based U-NET model | IB/OB/Tumor | With the average DSC of 0.9836, 0.8391 and 0.6856, respectively, for the IB, OB and tumor region segmentation using 60 datasets. |
| Gordon et al., 2018 (91) | CECT | Deep-learning convolutional neural network (DL-CNN) | IB/OB | With the average DSC of 0.9869 and 0.875, respectively, for the IB and OB segmentation using 172 datasets. |
| Ma et al., 2019 (92) | CECT | U-Net–based deep learning approach (U-DL) | IB | With the average DSC of 0.934 for the IB segmentation using 173 datasets. |
*IB and OB represent the inner and outer borders of bladder, respectively.
#CECT indicates contrast-enhanced CT.