From: Practical utility of liver segmentation methods in clinical surgeries and interventions
References | Method | Dataset | Performance |
---|---|---|---|
Lu et al. [78] | 3D-CNN employed for liver detection and probabilistic segmentation, followed by a Graphcut for segmentation refinement. | MICCAI-Sliver07, 3DIRCADB | VOE: 5.9, 9.36 RVD: 2.7%, 0.97% ASD: 0.91, 1.89 RMSD: 1.88, 4.15 MSD: 18.94 mm, 33.14 mm |
Wang et al. [79] | 2D U-Net trained in two stages to demonstrate the feasibility of transfer learning for CT segmentation | Custom Dataset (330 abdominal MRI and CT scans) | Dice: 0.94 ± 0.06 (CT) Dice: 0.95 ± 0.03 (T1-weighted MRI) Dice: 0.92 ± 0.05 (T2*-weighted MRI) |
Nakayama et al. [80] | In vivo comparison of automatic and manual volumetry for liver volume calculation | Custom Volumetric Dataset | Automatic: 982.99 cm3 ± 301.98 (volume), 4.4 minutes ± 1.9 (time) Manual: 937.10 cm3 ± 301.31 (volume), 32.8 minutes ± 6.9 (time) |
Allir et al. [81] | FCN used for coarse liver segmentation, followed by the use of region-based level set function for  tumor segmentation | LiTs, IRCAD | Liver Dice: 95.2%, 95.6% Liver Tumor Dice: 76.1%, 70% |
Yasaka et al. [82] | Custom CNN architecture for clinical retrospective study on different phases of CT scans | Custom Dataset (55536 Pictures) | Median Accuracy: 0.84 Median AUROC: 0.92 |
Vorontsov et al. [83] | FCN with two stages forliver and tumor segmentation | Custom Dataset (156 contrast material-enhanced CT scans) | Tumor Dice: 0.14 (size < 10 mm), 0.53 (size 10–20 mm), 0.68 (size > 20 mm) |