Skip to main content

Table 1 A comparison of SSL pretraining studies that investigated chest X-ray classification using the Chest X-ray14 dataset for pretraining, training, and testing using the official splits. The table gives class-wise average test AUC as reported by the authors, when training using all available labels

From: A survey of the impact of self-supervised pretraining for diagnostic tasks in medical X-ray, CT, MRI, and ultrasound

Method

Initialization

First Author [ref]

Identifier

Extractor

Random

ImageNet

SSL

Zhou [32]

C2L

ResNet-18

-

0.8150

0.8350

Zhou [32]

C2L

DenseNet-121

-

0.8290

0.8440

Ma [39]

SimMIM

ViT-B

0.7169

-

0.7955

Ma [39]

SimMIM

Swin-B

0.7704

-

0.8195

Liu [40]

S2MTS2

DenseNet-121

-

-

0.8250

Haghighi [41]

DiRA

ResNet-50

0.8031

0.8170

0.8112

Pang [42]

POPAR

Swin-B

0.7429

0.8132

0.8181