README.md 12.9 KB
Newer Older
1 2
# COVID-Net Open Source Initiative

3 4
**Note: The COVID-Net models provided here are intended to be used as reference models that can be built upon and enhanced as new data becomes available. They are currently at a research stage and not yet intended as production-ready models (not meant for direct clinicial diagnosis), and we are working continuously to improve them as new data becomes available. Please do not use COVID-Net for self-diagnosis and seek help from your local health authorities.**

lindawangg's avatar
lindawangg committed
5 6
**Update 04/14/2020: We released two new models, COVIDNet-CXR Small and COVIDNet-CXR Large, which were trained on a new COVIDx Dataset with both PA and AP X-Rays from Cohen et al, as well as additional COVID-19 X-Ray images from Figure1**

lindawangg's avatar
lindawangg committed
7
<p align="center">
lindawangg's avatar
lindawangg committed
8
	<img src="assets/covidnet-cxr-small-exp.png" alt="photo not available" width="70%" height="70%">
lindawangg's avatar
lindawangg committed
9 10 11
	<br>
	<em>Example chest radiography images of COVID-19 cases from 2 different patients and their associated critical factors (highlighted in red) as identified by GSInquire.</em>
</p>
lindawangg's avatar
test  
lindawangg committed
12

lindawangg's avatar
lindawangg committed
13
**Core COVID-Net team: Linda Wang, Alexander Wong, Zhong Qiu Lin, James Lee, Paul McInnis, Audrey Chung, Matt Ross (City of London), Blake VanBerlo (City of London)**\
lindawangg's avatar
lindawangg committed
14
Vision and Image Processing Research Group, University of Waterloo, Canada\
15
DarwinAI Corp., Canada
16

lindawangg's avatar
lindawangg committed
17
The COVID-19 pandemic continues to have a devastating effect on the health and well-being of the global population.  A critical step in the fight against COVID-19 is effective screening of infected patients, with one of the key screening approaches being radiological imaging using chest radiography.  It was found in early studies that patients present abnormalities in chest radiography images that are characteristic of those infected with COVID-19.  Motivated by this, a number of artificial intelligence (AI) systems based on deep learning have been proposed and results have been shown to be quite promising in terms of accuracy in detecting patients infected with COVID-19 using chest radiography images.  However, to the best of the authors' knowledge, these developed AI systems have been closed source and unavailable to the research community for deeper understanding and extension, and unavailable for public access and use.  Therefore, in this study we introduce COVID-Net, a deep convolutional neural network design tailored for the detection of COVID-19 cases from chest radiography images that is open source and available to the general public.  We also describe the chest radiography dataset leveraged to train COVID-Net, which we will refer to as COVIDx and is comprised of 13,800 chest radiography images across 13,725 patient patient cases from three open access data repositories.  Furthermore, we investigate how COVID-Net makes predictions using an explainability method in an attempt to gain deeper insights into critical factors associated with COVID cases, which can aid clinicians in improved screening.  **By no means a production-ready solution**, the hope is that the open access COVID-Net, along with the description on constructing the open source COVIDx dataset, will be leveraged and build upon by both researchers and citizen data scientists alike to accelerate the development of highly accurate yet practical deep learning solutions for detecting COVID-19 cases and accelerate treatment of those who need it the most.
lindawangg's avatar
lindawangg committed
18

lindawangg's avatar
lindawangg committed
19
For a detailed description of the methodology behind COVID-Net and a full description of the COVIDx dataset, please click [here](assets/COVIDNet_CXR.pdf).
20

21
Currently, the COVID-Net team is working on **COVID-RiskNet**, a deep neural network tailored for COVID-19 risk stratification.  Currently this is available as a work-in-progress via included `train_risknet.py` script, help to contribute data and we can improve this tool.
lindawangg's avatar
lindawangg committed
22

23
If you would like to **contribute COVID-19 x-ray images**, please submit to https://figure1.typeform.com/to/lLrHwv. Lets all work together to stop the spread of COVID-19!
lindawangg's avatar
lindawangg committed
24

lindawangg's avatar
lindawangg committed
25
If you are a researcher or healthcare worker and you would like access to the **GSInquire tool to use to interpret COVID-Net results** on your data or existing data, please reach out to a28wong@uwaterloo.ca or alex@darwinai.ca
lindawangg's avatar
lindawangg committed
26

lindawangg's avatar
lindawangg committed
27
Our desire is to encourage broad adoption and contribution to this project. Accordingly this project has been licensed under the GNU Affero General Public License 3.0. Please see [license file](LICENSE.md) for terms. If you would like to discuss alternative licensing models, please reach out to us at linda.wang513@gmail.com and a28wong@uwaterloo.ca or alex@darwinai.ca
lindawangg's avatar
lindawangg committed
28

lindawangg's avatar
lindawangg committed
29 30 31 32 33 34 35 36
The README contains information about:
* [requirements](#requirements) to install on your system
* how to [generate COVIDx dataset](#covidx-dataset)
* steps for [training](#steps-for-training), [evaluation](#steps-for-evaluation) and [inference](#steps-for-inference)
* [results](#results)
* [links to pretrained models](#pretrained-models)

If there are any technical questions after the README and past/current issues have been read, please post an issue or contact:
Desmond Lin's avatar
Desmond Lin committed
37 38 39
* desmond.zq.lin@gmail.com
* paul@darwinai.ca
* jamesrenhoulee@gmail.com
lindawangg's avatar
lindawangg committed
40
* linda.wang513@gmail.com
lindawangg's avatar
lindawangg committed
41

lindawangg's avatar
lindawangg committed
42 43 44 45 46 47 48 49 50 51 52 53 54
If you find our work useful, can cite our paper using:

```
@misc{wang2020covidnet,
    title={COVID-Net: A Tailored Deep Convolutional Neural Network Design for Detection of COVID-19 Cases from Chest Radiography Images},
    author={Linda Wang and Alexander Wong},
    year={2020},
    eprint={2003.09871},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}
```

lindawangg's avatar
lindawangg committed
55
## Requirements
56 57 58

The main requirements are listed below:

lindawangg's avatar
lindawangg committed
59 60
* Tested with Tensorflow 1.13 and 1.15
* OpenCV 4.2.0
lindawangg's avatar
lindawangg committed
61
* Python 3.6
62 63 64 65 66 67
* Numpy
* Scikit-Learn
* Matplotlib

Additional requirements to generate dataset:

68
* PyDicom
69 70
* Pandas
* Jupyter
lindawangg's avatar
lindawangg committed
71

lindawangg's avatar
lindawangg committed
72
## COVIDx Dataset
lindawangg's avatar
lindawangg committed
73
**Update 04/14/2020: Released new dataset with 152 COVID-19 train and 31 COVID-19 test samples. There are constantly new xray images being added to covid-chestxray-dataset and Figure1 covid dataset so we included train_COVIDx2.txt and test_COVIDx2.txt, which are the xray images we used for training and testing of CovidNet-CXR models.**
74 75

The current COVIDx dataset is constructed by the following open source chest radiography datasets:
lindawangg's avatar
lindawangg committed
76
* https://github.com/ieee8023/covid-chestxray-dataset
lindawangg's avatar
lindawangg committed
77
* https://github.com/agchung/Figure1-COVID-chestxray-dataset
78
* https://www.kaggle.com/c/rsna-pneumonia-detection-challenge (which came from: https://nihcc.app.box.com/v/ChestXray-NIHCC)
79

lindawangg's avatar
lindawangg committed
80
We especially thank the Radiological Society of North America, National Institutes of Health, Figure1, Dr. Joseph Paul Cohen and the team at MILA involved in the COVID-19 image data collection project for making data available to the global community.
81 82

### Steps to generate the dataset
lindawangg's avatar
lindawangg committed
83

84 85
1. Download the datasets listed above
 * `git clone https://github.com/ieee8023/covid-chestxray-dataset.git`
lindawangg's avatar
lindawangg committed
86
 * `git clone https://github.com/agchung/Figure1-COVID-chestxray-dataset`
87 88
 * go to this [link](https://www.kaggle.com/c/rsna-pneumonia-detection-challenge/data) to download the RSNA pneumonia dataset
2. Create a `data` directory and within the data directory, create a `train` and `test` directory
lindawangg's avatar
lindawangg committed
89
3. Use [create\_COVIDx\_v3.ipynb](create_COVIDx_v3.ipynb) to combine the three dataset to create COVIDx. Make sure to remember to change the file paths.
90
4. We provide the train and test txt files with patientId, image path and label (normal, pneumonia or COVID-19). The description for each file is explained below:
lindawangg's avatar
lindawangg committed
91 92
 * [train\_COVIDx2.txt](train_COVIDx2.txt): This file contains the samples used for training COVIDNet-CXR.
 * [test\_COVIDx2.txt](test_COVIDx2.txt): This file contains the samples used for testing COVIDNet-CXR.
lindawangg's avatar
lindawangg committed
93

94
### COVIDx data distribution
lindawangg's avatar
readme  
lindawangg committed
95

lindawangg's avatar
lindawangg committed
96
Chest radiography images distribution
97 98
|  Type | Normal | Pneumonia | COVID-19 | Total |
|:-----:|:------:|:---------:|:--------:|:-----:|
lindawangg's avatar
lindawangg committed
99 100
| train |  7966  |    5451   |   152    | 13569 |
|  test |   100  |     100   |    31    |   231 |
lindawangg's avatar
stats  
lindawangg committed
101

lindawangg's avatar
lindawangg committed
102
Patients distribution
103 104
|  Type | Normal | Pneumonia | COVID-19 |  Total |
|:-----:|:------:|:---------:|:--------:|:------:|
lindawangg's avatar
lindawangg committed
105 106
| train |  7966  |    5440   |    107   |  13513 |
|  test |   100  |      98   |     14   |    212 |
lindawangg's avatar
readme  
lindawangg committed
107

lindawangg's avatar
lindawangg committed
108
## Training and Evaluation
109 110 111 112
The network takes as input an image of shape (N, 224, 224, 3) and outputs the softmax probabilities as (N, 3), where N is the number of batches.
If using the TF checkpoints, here are some useful tensors:

* input tensor: `input_1:0`
113
* logit tensor: `dense_3/MatMul:0`
114 115 116 117 118 119
* output tensor: `dense_3/Softmax:0`
* label tensor: `dense_3_target:0`
* class weights tensor: `dense_3_sample_weights:0`
* loss tensor: `loss/mul:0`

### Steps for training
120 121 122 123 124
TF training script from a pretrained model:
1. We provide you with the tensorflow evaluation script, [train_tf.py](train_tf.py)
2. Locate the tensorflow checkpoint files (location of pretrained model)
3. To train from a pretrained model, `python train_tf.py --weightspath models/COVIDNetv2 --metaname model.meta_train --ckptname model-2069`
4. For more options and information, `python train_tf.py --help`
125 126 127 128 129

### Steps for evaluation

1. We provide you with the tensorflow evaluation script, [eval.py](eval.py)
2. Locate the tensorflow checkpoint files
130
3. To evaluate a tf checkpoint, `python eval.py --weightspath models/COVID-Netv2 --metaname model.meta_eval --ckptname model-2069`
131
4. For more options and information, `python eval.py --help`
132

133
### Steps for inference
134 135 136 137
**DISCLAIMER: Do not use this prediction for self-diagnosis. You should check with your local authorities for the latest advice on seeking medical assistance.**

1. Download a model from the [pretrained models section](#pretrained-models)
2. Locate models and xray image to be inferenced
lindawangg's avatar
lindawangg committed
138
3. To inference, `python inference.py --weightspath models/COVIDNet-CXR-Large --metaname model.meta_eval --ckptname model-8485 --imagepath assets/ex-covid.jpeg`
139
4. For more options and information, `python inference.py --help`
lindawangg's avatar
lindawangg committed
140

141 142 143 144 145 146 147
### Steps for Training COVIDNet-Risk

COVIDNet-Risk uses the same architecture as the existing COVIDNet - but instead it predicts the *"number of days since symptom onset"\** for a diagnosed COVID-19 patient based on their chest radiography (same data as COVIDNet). By performing offset stratification, we aim to provide an estimate of prognosis for the patient. Note that the initial dataset is fairly small at the time of writing and we hope to see more results as data increases.

1. Complete data creation and training for COVIDNet (see Training above)
2. run `train_risknet.py` (see `-h` for argument help)

148
*\* note that definition varies between data sources*
149

150
## Results
lindawangg's avatar
lindawangg committed
151
These are the final results for COVIDNet-CXR Small and COVIDNet-CXR Large.
lindawangg's avatar
lindawangg committed
152

lindawangg's avatar
lindawangg committed
153
### COVIDNet-CXR Small
154
<p align="center">
lindawangg's avatar
lindawangg committed
155
	<img src="assets/cm-covidnetcxr-small.png" alt="photo not available" width="50%" height="50%">
156
	<br>
lindawangg's avatar
lindawangg committed
157
	<em>Confusion matrix for COVIDNet-CXR Small on the COVIDx test dataset.</em>
158
</p>
lindawangg's avatar
lindawangg committed
159

160 161 162 163 164 165 166 167 168 169
<div class="tg-wrap" align="center"><table class="tg">
  <tr>
    <th class="tg-7btt" colspan="3">Sensitivity (%)</th>
  </tr>
  <tr>
    <td class="tg-7btt">Normal</td>
    <td class="tg-7btt">Pneumonia</td>
    <td class="tg-7btt">COVID-19</td>
  </tr>
  <tr>
lindawangg's avatar
lindawangg committed
170 171 172
    <td class="tg-c3ow">97.0</td>
    <td class="tg-c3ow">90.0</td>
    <td class="tg-c3ow">87.1</td>
173 174
  </tr>
</table></div>
lindawangg's avatar
lindawangg committed
175

lindawangg's avatar
lindawangg committed
176
<div class="tg-wrap" align="center"><table class="tg">
177 178 179 180 181 182 183 184 185
  <tr>
    <th class="tg-7btt" colspan="3">Positive Predictive Value (%)</th>
  </tr>
  <tr>
    <td class="tg-7btt">Normal</td>
    <td class="tg-7btt">Pneumonia</td>
    <td class="tg-7btt">COVID-19</td>
  </tr>
  <tr>
lindawangg's avatar
lindawangg committed
186 187 188
    <td class="tg-c3ow">89.8</td>
    <td class="tg-c3ow">94.7</td>
    <td class="tg-c3ow">96.4</td>
189 190 191
  </tr>
</table></div>

lindawangg's avatar
lindawangg committed
192 193

### COVIDNet-CXR Large
lindawangg's avatar
lindawangg committed
194
<p align="center">
lindawangg's avatar
lindawangg committed
195
	<img src="assets/cm-covidnetcxr-large.png" alt="photo not available" width="50%" height="50%">
lindawangg's avatar
lindawangg committed
196
	<br>
lindawangg's avatar
lindawangg committed
197
	<em>Confusion matrix for COVIDNet-CXR Large on the COVIDx test dataset.</em>
lindawangg's avatar
lindawangg committed
198 199
</p>

lindawangg's avatar
lindawangg committed
200
<div class="tg-wrap" align="center"><table class="tg">
lindawangg's avatar
lindawangg committed
201
  <tr>
202
    <th class="tg-7btt" colspan="3">Sensitivity (%)</th>
lindawangg's avatar
lindawangg committed
203 204 205
  </tr>
  <tr>
    <td class="tg-7btt">Normal</td>
206 207
    <td class="tg-7btt">Pneumonia</td>
    <td class="tg-7btt">COVID-19</td>
lindawangg's avatar
lindawangg committed
208 209
  </tr>
  <tr>
lindawangg's avatar
lindawangg committed
210 211 212
    <td class="tg-c3ow">99.0</td>
    <td class="tg-c3ow">89.0</td>
    <td class="tg-c3ow">96.8</td>
lindawangg's avatar
lindawangg committed
213 214 215
  </tr>
</table></div>

lindawangg's avatar
lindawangg committed
216
<div class="tg-wrap" align="center"><table class="tg">
lindawangg's avatar
lindawangg committed
217
  <tr>
218
    <th class="tg-7btt" colspan="3">Positive Predictive Value (%)</th>
lindawangg's avatar
lindawangg committed
219 220 221
  </tr>
  <tr>
    <td class="tg-7btt">Normal</td>
222 223
    <td class="tg-7btt">Pneumonia</td>
    <td class="tg-7btt">COVID-19</td>
lindawangg's avatar
lindawangg committed
224 225
  </tr>
  <tr>
lindawangg's avatar
lindawangg committed
226 227 228
    <td class="tg-c3ow">91.7</td>
    <td class="tg-c3ow">98.9</td>
    <td class="tg-c3ow">90.9</td>
lindawangg's avatar
lindawangg committed
229 230 231
  </tr>
</table></div>

lindawangg's avatar
lindawangg committed
232
## Pretrained Models
Linda Wang's avatar
Linda Wang committed
233
**Information on these new models to be released soon**
234

lindawangg's avatar
lindawangg committed
235 236
|  Type | COVID-19 Sensitivity | # Params (M) | MACs (G) |        Model        |
|:-----:|:--------------------:|:------------:|:--------:|:-------------------:|
lindawangg's avatar
lindawangg committed
237 238
|  ckpt |         87.1         |     117.4    |   2.26   |[COVIDNet-CXR Small](https://bit.ly/CovidNet-CXR-Small)|
|  ckpt |         90.0         |     127.4    |   3.59   |[COVIDNet-CXR Large](https://bit.ly/CovidNet-CXR-Large)|