Skip to main content

Table 6 Evaluation performance of generative methods presented for various anatomical regions in the examined studies

From: Enhancing cancer differentiation with synthetic MRI examinations via generative models: a systematic review

Author

Quantitative metrics

Qualitative metrics

Downstream task

 

Direct for SD

Indirect prior to SDA

Indirect after SDA

Experts/statistics(%)

Model

kitchen [78]

N/A

N/A

N/A

Plotted samples

N/A

Hu [79]

N/A

N/A

N/A

Plotted samples

N/A

Wang [81]

N/A

Acc: 0.92

(& classical DA)

Acc: 0.95

Plotted samples

FC-ANN

Yang [82]

FID: 179.54 ± 5.38

IS: 2.61 ± 0.24

MID: 0.011 ± 0.006

N/A

Acc: 0.93 ± 0.45

Three Radiologists/

R1: 4.4 FPR

R2: 91.3 FPR

R3: 7.8 FPR

FC-ANN

Wang [83]

IS: 2.24 ± 0.03

FID: 178.2 ± 3.7

SCA: 0.944 ± 0.5

N/A

Acc: 0.90

Sen(0.1): 0.26

Sen(1.0): 0.80

Three radiologists/

R1: Sen 26.0 SCA 63.0

R2: Sen 24.0 SCA 61.0

R3: Sen 82.0 SCA 89.0

FC-ANN

CNN detector

F-Quilez [84]

HD: 8.10

Dice: 0.678

MSD: 3.16

VDSC: 0.543

Dice: 0.737

MSD: 1.16

VDSC: 0.693

Visual evaluation

U-Net

Yu [85]

KLD: as low as 0.73

N/A

Acc: up-to 0.892

AUC: up-to 0.885

Two radiologists/(N/A)

LENet-NIN

Yan [88]

L2 Distance

(no values)

N/A

Acc: up-to 0.983

AUC: up-to 0.997

t-SNE & PCA

visualization

CNN

Gao [89]

N/A

N/A

AAcc: 0.8105

ma-AUC: 0.8847

Two radiologists/(N/A)

CNN

Gao [90]

N/A

N/A

IVPL AAcc: 0.715, ma-AUC: 0.9204

EVPL AAcc: 0.794, ma-AUC: 0.9451

IVPAL AAcc: 0.700, ma-AUC: 0.8250

EVPAL AAcc: 0.767, ma-AUC: 0.8646

Radiologist

IV AAcc: 0.820, ma-AUC: 0.8950

EV AAcc: 0.839, ma-AUC: 0.9063

InceptionV4

Haarburger [91]

FID: as low as 20.23

1-NN: up-to 0.268

N/A

N/A

Radiologist /70.0

Layperson /76.7

N/A

Sun [92]

N/A

Dice: 0.6856 ± 0.18

Dice: 0.6903 ± 0.20

Plotted samples

3D U-Net

  1. SD, Synthetic Data; SDA, Synthetic Data Augmentation; Acc, Accuracy; FID, Frechet Inception Distance; IS, Inception Score; MID, Mutual Information Distance; FPR, False Positive Ratio; SCA, Slice-level Classification Accuracy; Sen, Sensitivity; Spe, Specificity; HD, Hausdorff Distance; MSD, Mean Surface Distance; VDSC, mean Volumetric DSC; KLD, Kullback–Leibler Divergence; AUC, Area Under the Curve; t-SNE, t-distributed Stochastic Neighbor Embedding; PCA, Principal Component Analysis; IVPL, Internal Validation Patch Level; EVPL, External Validation Patch Level; IVPAL, Internal Validation PAtient Level; EVPAL, External Validation PAtient Level; AAcc, Average Accuracy; ma, micro-averaging; 1-NN, Nearest Neighbor; N/A, Not applicable; Symbol ’-’ represents that the corresponding information was not provided in the publication