We present CovidAID
(Covid AI Detector), a PyTorch (python3) based implementation, to identify COVID-19 cases from X-Ray images. The model takes as input a chest X-Ray image and outputs the probability scores for 4 classes (NORMAL
, Bacterial Pneumonia
, Viral Pneumonia
and COVID-19
).
It is based on CheXNet (and it's reimplementation by arnoweng).
Please refer to INSTALL.md for installation.
CovidAID
uses the covid-chestxray-dataset for COVID-19 X-Ray images and chest-xray-pneumonia dataset for data on Pneumonia and Normal lung X-Ray images.
Chest X-Ray image distribution
Type | Normal | Bacterial Pneumonia | Viral Pneumonia | COVID-19 | Total |
---|---|---|---|---|---|
Train | 1341 | 2530 | 1337 | 115 | 5323 |
Val | 8 | 8 | 8 | 10 | 34 |
Test | 234 | 242 | 148 | 30 | 654 |
Chest X-Ray patient distribution
Type | Normal | Bacterial Pneumonia | Viral Pneumonia | COVID-19 | Total |
---|---|---|---|---|---|
Train | 1000 | 1353 | 1083 | 80 | 3516 |
Val | 8 | 7 | 7 | 7 | 29 |
Test | 202 | 77 | 126 | 19 | 424 |
Please refer our paper paper for description of architecture and method. Refer to GETTING_STARTED.md for detailed examples and abstract usage for training the models and running inference.
We present the results in terms of both the per-class AUROC (Area under ROC curve) on the lines of CheXNet
, as well as confusion matrix formed by treating the most confident class prediction as the final prediction. We obtain a mean AUROC of 0.9738
(4-class configuration).
3-Class Classification | 4-Class Classification | |||||||||||||||||||||||||||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
|
| |||||||||||||||||||||||||||||||||||||
ROC curve | ||||||||||||||||||||||||||||||||||||||
Confusion Matrix |
To demonstrate the results qualitatively, we generate saliency maps for our model’s predictions using RISE. The purpose of these visualizations was to have an additional check to rule out model over-fitting as well as to validate whether the regions of attention correspond to the right features from a radiologist’s perspective. Below are some of the saliency maps on COVID-19 positive X-rays.
This work was a part of guided project done by Harshvardhan Munda.
If you have any question, please file an issue or contact the author:
Harshvardhan Munda
[email protected]
[email protected]
- Add support for
torch>=1.0
- Support for multi-GPU training