An open-source AI algorithm that automatically detects 37 radiological findings on chest X-rays in real-time, trained on over 2.5 million studies from the NHS.
Try X-Raydar Online
GitHub
HuggingFace
X-Raydar is a deep learning system trained on over 2.5 million chest X-ray studies from six NHS hospitals across the UK, spanning 13 years of clinical data. Given a chest X-ray, it automatically screens for 37 radiological findings and returns a probability for each — in real-time.
It can be used to prioritise urgent cases, flag abnormalities for radiologist review, or as an automated second-reader.
X-Raydar is open-source and freely available to the research community for non-clinical evaluation. Researchers can use our online platform to upload chest X-rays and receive automated reports in real-time, or download the model weights and source code to run X-Raydar on their own infrastructure.
This project was funded by a Wellcome Trust Innovator Award.
37 radiological findings detected by X-Raydar, based on the Fleischner glossary and RadLex terminology.
| Finding | Glossary Term | Examples in Free Text |
|---|---|---|
| Lungs | ||
| Parenchymal opacification | Consolidation, Airspace | Consolidation; airspace opacification; batwing shadowing; pulmonary oedema |
| Interstitial opacification / Fibrosis | Reticular pattern, Honeycombing | Interstitial shadowing; fibrosis; Kerley B lines; septal thickening |
| Ground-glass opacity | Ground-glass opacity | Ground glass opacity; ground glass change |
| Parenchymal mass or nodule | Mass, Nodule, Opacity | Lung mass; pulmonary nodule; lung metastases |
| Cavity | Cavity | Cavitation; cavity; cavitating lesion |
| Bulla | Bulla | Bulla; bullae; lung lucencies |
| Emphysema | Emphysema | Emphysema; emphysematous change |
| Hyperexpanded lungs | Air trapping | Hyperexpanded lungs; hyperinflation; large lung volume |
| Bronchial changes | Bronchiectasis | Bronchiectasis; bronchial wall thickening |
| Collapse / Volume loss | Collapse | Lung collapse; decreased lung volume |
| Atelectasis | Atelectasis | Atelectasis; atelectatic bands; linear atelectasis |
| Apical changes | Apical cap | Apical thickening; apical fibrosis |
| Pulmonary blood flow redistribution | Pulmonary blood flow redistribution | Upper lobe diversion; prominent upper lobe vessels |
| Mediastinum & Hila | ||
| Pneumomediastinum | Pneumomediastinum | Air in mediastinum |
| Mediastinum, displaced | Mediastinal compartments | Mediastinal shift |
| Mediastinum, widened | Mediastinal compartments | Mediastinal widening |
| Hilum / Paratracheal changes | Hilum lymphadenopathy | Hilar enlargement; hilar lymphadenopathy |
| Pleural Space | ||
| Pneumothorax | Pneumothorax | Pneumothorax; tension pneumothorax |
| Pleural effusion | Pleural space | Pleural effusion; pleural fluid; blunting of costophrenic angle |
| Pleura, abnormality | Pleural plaque | Pleural plaques; pleural thickening; pleural scarring |
| Diaphragm | ||
| Diaphragm, abnormality | Diaphragm | Elevated hemidiaphragm; eventration of diaphragm |
| Cardiovascular | ||
| Cardiomegaly | Enlargement Heart | Cardiomegaly; heart enlarged; large heart |
| Cardiac calcification | Calcification Heart | Coronary artery calcification |
| Dextrocardia | — | Dextrocardia; situs inversus |
| Aortic calcification | Calcification Aorta | Calcification of the aorta |
| Aortic tortuosity | Tortuous Aorta | Unfolding of the thoracic aorta |
| Osseous Structures & Chest Wall | ||
| Fracture, rib | Fracture Rib | Rib fracture |
| Fracture, clavicle | Fracture Clavicle | Clavicle fracture |
| Bone, lesion | Bone-forming neoplasm | Bone metastasis |
| Scoliosis | Scoliosis | Scoliosis; kyphoscoliosis; kyphosis |
| Mass, paraspinal | Mass Paraspinal | Paraspinal density |
| Skin & Soft Tissue | ||
| Emphysema, subcutaneous | Emphysema, Subcutaneous | Subcutaneous emphysema; surgical emphysema |
| Mass, soft tissue | Mass, Soft tissue density | Axillary mass; axillary lymphadenopathy |
| Abdomen | ||
| Bowel, dilated | Bowel Dilated | Dilated bowel loops; distended stomach |
| Pneumoperitoneum | Pneumoperitoneum | Pneumoperitoneum |
| Hernia | Hernia | Hiatus hernia |
| Other | ||
| Abnormal, not clinically important | Normal variant | Granuloma; pectus deformity; anatomical variant |
| Medical object | Medical Object | — |
| Normal anatomy | Normal anatomy | — |
Based on the Fleischner Society glossary and RadLex terminology.
X-Raydar comprises two models: a chest X-ray image classifier that predicts the probability of 37 radiological findings from a frontal chest X-ray, and an NLP model that extracts radiological findings from free-text radiology reports. Both sets of pre-trained weights are available for download on HuggingFace.
The source code required to run inference with either model is available on GitHub. The repository includes instructions for setting up the environment, loading the pre-trained weights, and running predictions on your own data.
If you use X-Raydar in your research, please cite:
YD Cid, M Macpherson, L Gervais-Andre, Y Zhu, G Franco, R Santeramo, et al. Development and validation of open-source deep neural networks for comprehensive chest x-ray reading: a retrospective, multicentre study. The Lancet Digital Health, 6(1), e44–e57, 2024.
