Table 1.

Overview of ML algorithms applied to lesion segmentation.

Authors	ML Algorithm	Aim	Imaging Modality	Performance
Christ et al. [20]	Cascaded CNNs based on a U-Net Architecture	Liver and tumor segmentation	CT	DICE scores: 94.3% and 91% *
Ouhmich et al. [25]	Cascaded CNNs based on a U-Net Architecture	Segmentation of healthy and cancerous liver tissues, discriminating normal parenchyma, active HCC and necrotic tumoral tissue	CT	DICE score: 90.5%, 59.6% and 75.8% **
Zhang et al. [26]	Auto-context-based CNNs based on a U-Net Architecture	Liver Tissue Classification	MRI	F-Score: 0.80, 0.83 and 0.81 ***
Han et al. [27]	CNN based on 2.5D model	Liver lesion segmentation	CT	DICE score: 0,67
Wardhana et al. [29]	CNN based on 2.5D model	Liver and lesion segmentation	CT	Lesion DICE score: 78.4 ± 16.7 and 83.6 ± 24.7% **** Liver DICE score: 95.3 ± 1.8% and 94.6 ± 2% ****
Conze PH. et al. [30]	Scale-adaptive supervoxel-based random forests	Liver tumor segmentation	CT	TN rate error of Δτ: 4.08. DSC: p95.9, a80.3, n86, t92.6 *****
Chih-Yu Hsu et al. [31]	Poisson Gradient Vector Flow-ACM based on a genetic algorithm	Liver segmentation	PET	Reduction of the iterations needed in liver’s edge processing selection

ML: machine learning; CNN: convolutional neural network; ACM: active contour model; AUC: area under the curve; TN: tumor necrosis; DSC: dice similarity coefficients. * Results respectively on 3DIRCAD (3D Image Reconstruction for Comparison of Algorithm Database) dataset and Clinical CT dataset. ** Results respectively for parenchyma, active tumor and necrosis. *** Results respectively by multi-resolution input multi-phase training, multi-resolution input single-phase training and single-resolution input single-phase training. **** Results respectively by Net01 and Net02. ***** p: parenchyma; a: active; n: necrosis; t: tumoral.