Table 4.

Segmentation methods for radiation therapy (RT)

Reference	Method	Dataset	Performance
Li et al. [108]	Voxel-based Adaboost is used for liver localization. Shape and appearance models are employed to segment the liver, followed by free form deformation for refinement	MICCAI Sliver07	Liver Dice: 0.911 ± 0.010 (CT), 0.922 ± 0.011 (CTce) Tumor burden RMSE: 0.015
Wu et al. [9]	Liver volume is extracted by histogram-based adaptive thresholding and morphological operations, followed by graph cuts	MICCAI Sliver07	VOE: 7.54% RVD: 4.16% ASD: 0.95 mm RMSD: 1.94 mm MaxD: 18.48 mm Run time: 12.21 sec Tumor burden RMSE: 0.016
Wang et al. [109]	Adaptive mesh expansion model (AMEM) is used for liver segmentation from CT scans. A virtual deformable simplex model (DSM) is introduced to represent the mesh	MICCAI Sliver07	Mean overlap error: 6.8% Mean volume difference: 2.7% ASSD: 1.3 mm RMSD: 2.7 mm Tumor burden RMSE: 0.016
Yuan et al. [110]	Hierarchical convolutional-deconvolutional neural networks (CDNN) for liver and tumor segmentation, followed tumor estimation	LiTS	Liver Dice: 0.967 Liver RMSD: 2.303 Tu mor Dice: 0.82 Tumor RMSD: 1.678 Tumor burden RMSE: 0.017