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

Table 6 The public X-ray datasets referenced in this review, including links to request or download the data

Name [Citation]	Description	Examples	Patients
CheXpert [6]	A fully manually annotated 14-class dataset of chest X-rays.	\({224\,316}\)	\({65\,240}\)
ChestX-ray14 [33]	A 14-class dataset of chest X-rays with labels extracted from radiology reports.	\({112\,120}\)	\({30\,805}\)
ChestMNIST [34]	Identical to ChestX-ray14. Part of MedMNIST [54].	\({112\,120}\)	\({30\,805}\)
COVIDx CXR-2 [56]	chest X-rays labelled for the presence or absence of COVID-19.	\({19\,203}\)	\({16\,656}\)
MIMIC-CXR [65]	Chest X-rays, metadata, and free text reports. Same label categories as CheXpert. Some labels were manually determined, and others were automatically assigned using the reports.	\({371\,920}\)	\({65\,079}\)
RSNA Pneumonia [183]	Chest X-rays with bounding box labels for bacterial and viral pneumonias	\({30\,000}\)	\({12\,274}\)
PneumoniaMNIST [184]	Paediatric chest X-rays labelled for the presence or absence of pneumonia. Part of MedMNIST.	5856	5856

ISSN: 1471-2342