The tool aims to speed up the diagnosis done by a doctor and to point out X-ray images that need to be double-checked. This tool is not to be used as a stand-alone diagnostic model.
The model was tested on limited data. It shouldn’t be used as a final diagnostic tool, but a tool for helping doctors with the diagnosis.
Information on training data
The model is trained on these datasets:
1. NIH Chest X-ray Dataset – https://www.kaggle.com/nih-chest-xrays/data
2. Pediatric Chest X-ray Pneumonia – https://data.mendeley.com/datasets/rscbjbr9sj/2
3. Shenzhen Hospital X-ray Set – https://lhncbc.nlm.nih.gov/publication/pub9931
4. PadChest – https://arxiv.org/abs/1901.07441
Model performance metrics
The model’s performance is 0.93 F1-score if given one frontal chest X-ray for detecting whether a person has a condition or not.
To clone the model from GitHub, open the link below: