This is an implementation of CheXNet based on pytorch.
- If there is any errors in my code, please contact me, thanks!
- You can download the chest x-ray images from NIH
- fine_tune_densenet.py
- test_densenet.py
- For a fair comparison, we follow the publicly available data split list.
| Pathology | Orignial CheXNet | Our Implemented |
|---|---|---|
| Atelectasis | 0.8094 | 0.7756 |
| Cardiomegaly | 0.9248 | 0.8873 |
| Effusion | 0.8638 | 0.8311 |
| Infiltration | 0.7345 | 0.6980 |
| Mass | 0.8676 | 0.8219 |
| Nodule | 0.7802 | 0.7640 |
| Pneumonia | 0.7680 | 0.7166 |
| Pneumothorax | 0.8887 | 0.8517 |
| Consolidation | 0.7901 | 0.7519 |
| Edema | 0.8878 | 0.8469 |
| Emphysema | 0.9371 | 0.9042 |
| Fibrosis | 0.8047 | 0.8280 |
| Pleural Thickening | 0.8062 | 0.7693 |
| Hernia | 0.9164 | 0.9314 |
| Avg | 0.8414 | 0.8127 |
- python 3
- pytorch 0.4.0