. Author manuscript; available in PMC: 2022 Jun 15.

Published in final edited form as: Domain Adapt Represent Transf Afford Healthc AI Resour Divers Glob Health (2021). 2021 Sep 21;12968:3–13. doi: 10.1007/978-3-030-87722-4_1

Table 1.

Benchmarking transfer learning for seven common medical imaging tasks, spanning over different label structures (binary/multi-label classification and segmentation), modalities, organs, diseases, and data size.

Code^†	Application	Modality	Dataset
ECC	Pulmonary Embolism Detection	CT	RSNA PE Detection [2]
DXC₁₄	Fourteen thorax diseases classification	X-ray	NIH ChestX-Ray14 [37]
DXC₅	Five thorax diseases classification	X-ray	CheXpert [22]
VFS	Blood Vessels Segmentation	Fundoscopic	DRIVE [5]
PXS	Pneumothorax Segmentation	X-ray	SIIM-ACR [1]
LXS	Lung Segmentation	X-ray	NIH Montgomery [24]
TXC	Tuberculosis Detection	X-ray	NIH Shenzhen CXR [24]

^†

The first letter denotes the object of interest (“E” for embolism, “D” for thorax diseases, etc); the second letter denotes the modality (“X” for X-ray, “F” for Fundoscopic, etc); the last letter denotes the task (“C” for classification, “S” for segmentation).