Table 1 An Overview of the Deep learning Models in Digital Breast Tomosynthesis

[41]

Caffe

DBT¹

2D mamo = 1864, 3D mamo = 339,

Mean ROI sesitivity, suspicous lesions(conventional methods = 0.8320 + \(-0.040\),

Suspicious lesions = 328, malignant lesions = 115

DL = 0.893 + \(-0.003\)), malignant lesions(conventional methods = 0.852 + \(-0.065\),

DL = 0.930 + \(-0.046\))

[43]

AlexNet/DCNN

DM/DBT

Dataset = 2192

AUC (before pruning = 0.88, after pruning = 0.90)

[44]

VGG19

SM/DM/DBT

Dataset patients = 76, lesions = 78

AUC = 0.89 + \(-0.04\) classification of malignant and benign

[45]

ResNet

DM/DBT

Patients = 62,417, exams = 198,201, images = 830,450

ROC AUC = 0.9

[46]

VGG16

DM/DBT

Patients = 441, views = 927, CC = 460, ML = 4, MLO = 463

Malignant classification (AUC = 0.91, ACC = 95.1%, SEN = 70.8%, SPE = 98.9%)

[47]

3D-DCNN

DBT

Patients = 40, reconstructed volume = 160

Avg AUC = 0.847 + \(-0.012\)

[48]

DCaRBM/DCNN

DBT

Images = 87, breast/volume = 87, image slices = 5040

AUC = 0.87, ACC = 86.81, SPE = 87.5, SEN = 86.6

[50]

AlexNet (2D-CNN)

DM/DBT

Data = 3705

auROC = 0.854

[49]

CNN (AlexNet)

DM/DBT

Data = 3290

auROC = 0.73

[51]

CNN (ImageNet)

DM/DBT

Patients = 1124

ACC = 0.91, F1 = 0.91, Precision = 0.93, Recall = 0.88 AUC = 0.97

[52]

ResNet-34

SM/DM/DBT

Exams = 207,776

Four class acc = 82.2, four class macro AUC = 0.95,

Binary acc = 91.1, binary AUC = 0.971

[53]

EMPIRE/FBP

DBT

Patients = 374

pAUC = 0.880

[54]

ResNet-50

DM/DBT

Cases = 63,798

AUC = 0.95

[55]

Faster RCNN/DCNN

DBT

Cases = 89

ROC AUC = 0.96

[56]

3D-Mask-RCNN

DBT

Cases = 364

Lesion based mass detection (sensitivity = 90% with 0.8 FPs),

Breast based mass detection (sensitivity = 90% with 0.83 FPs)

[57]

U-Net

DBT

Data = 87

SEN = 0.869, ACC = 0.871, AUC = 0.859

[58]

DenseNet

DBT

Patients = 5060, studies = 5610, DBT volumes = 22,032

Sensitivity = 65% at 2 FPS

[60]

Faster RCNN

DBT

Patients = 68, DBT volume = 265

mean true positive fraction, typical AD 0.6 + \(-0.05\)