OpenCovidDetector: Open source algorithm for detecting COVID-19 on Chest CT

Important: please refer to this hub https://github.com/ChenWWWeixiang/diagnosis_covid19 for newset virsion.

OpenCovidDetector is an opensource COVID-19 diagnosis system implementing on pytorch, which is also as presented in our paper: Development and Evaluation of an AI System for COVID-19 Diagnosis (https://www.medrxiv.org/content/10.1101/2020.03.20.20039834v2)

About the Project

Early detection of COVID-19 based on chest CT will enable timely treatment of patients and help control the spread of the disease. With rapid spreading of COVID-19 in many countries, however, CT volumes of suspicious patients are increasing at a speed much faster than the availability of human experts. Here, we propose an artificial intelligence (AI) system for fast COVID-19 diagnosis with an accuracy comparable to experienced radiologists.

Performance

A large dataset was constructed by collecting 970 CT volumes of 496 patients with confirmed COVID-19 and 260 negative cases from three hospitals in Wuhan, China, and 1,125 negative cases from two publicly available chest CT datasets. Trained using only 312 cases, our diagnosis system, which is based on deep convolutional neural network, is able to achieve an accuracy of 94.98%, an area under the receiver operating characteristic curve (AUC) of 97.91%, a sensitivity of 94.06%, and a specificity of 95.47% on an independent external verification dataset of 1,263 cases. In a reader study involving five radiologists, only one radiologist is slightly more accurate than the AI system.

Visualization

Guidance to Use

Environment

The code has been succesfully run in Ubuntu 16.04, python 3.6.1,CUDA 10.0.

Packages requirements: matplotlib==3.1.2 six==1.13.0 torch==1.3.1 scikit_image==0.16.2 imageio==2.6.1 scipy==1.3.3 numpy==1.15.3 opencv_python==4.1.1.26 pandas==0.23.4 torchvision==0.4.2 Pillow==6.2.1 pydicom==1.4.2 pyradiomics scikit_learn==0.22.2.post1 seaborn==0.10.0 SimpleITK==1.2.4 skimage toml==0.10.0 xlrd==1.2.0

run pip install -r requirements.txt to install all above packages.

Use Trained Model for Inference

1. Data Preparation : A demo covid-19 data can be made at https://cloud.tsinghua.edu.cn/f/365e7f81e4b443eb9fab/?dl=1. When using your own data, make sure that

   • Data and lung segmentations are in pair and in .nii format.
   • Negative cases should named begin with 'c'. We use names to as classification groundtruths.

2. Download Trained Weight: a trained model is available at https://cloud.tsinghua.edu.cn/f/ba180ea9b2d44fdc9757/?dl=1

3. Test: The input data files and input lung segmentation files should be in pair.

python testengine.py -p <path to trainedmodel> -m <list of paths for lung segmentation> -i <list of paths for image data> -o <path to save record> -g <gpuid>

Train on Your Own Data

1. Data Preparation : The datasets from Wuhan Union Hospital, Western Campus of Wuhan Union Hospital, and Jianghan Mobile Cabin Hospital were used under the license of the current study and are not publicly available. Applications for access to the LIDC-IDRI database can be made at https://wiki.cancerimagingarchive.net/display/Public/LIDC-IDRI. ILD-HUG database can be accessed at http://medgift.hevs.ch/wordpress/databases/ild-database/.

2. Volumes to Images: We suggest that test data should be in ".nii" format (any formats that SimpleITK can work on is OK with small changes in codes) and training data should be in ".jpg" format (any formats that opencv-python can work on is OK with small changes in codes). A script "data/test_hu.py" is used to cut volumes into images.

3. Lung Segmentation : using Deeplabv1 (https://github.com/DrSleep/tensorflow-deeplab-resnet) or any other segmentation methods.

4. Split Dataset:

python data/get_set_seperate_jpg.py -p <list of paths to jpgs for seperate> -t <train list output path> -v <validation list output path>

5. Begin Training: training parameters are listed on options_lip.toml. Run python main.py to train the model.

6. Metrics: to evaluate the performances, run python testengine.py -p <path to trainedmodel> -m <list of paths for lung segmentation> -i <list of paths for image data> -o <path to save record> -g <gpuid> and then the script python ploc_roc.py -i <list of paths for recording npys> -o <path to save metrics>

More Research Tools

• Model Visualization: A script to show Grad-CAM/CAM result is available. Input images should be in jpg formats and should be concatenated with lung mask as Red channel. The input raw jpgs and input masked jpgs should be in pair:

python models/gradcam.py --image_path <raw jpg img path> --mask_path <jpg img with mask path> --model_path <path to trained model> --output_path <path to output>

You can also use our volume cam script, which demands raw data volumes and lung segmentation volumes as inputs.

python models/grad_volume.py --image_path <raw data nii path> --mask_path <lung mask nii file path> --model_path <path to trained model> --output_path <path to output>

• Fractal Dimension Features :

  • Extract Fractal Dimension: extract fractal dimension of a region.python fractal-dimension/fractals.py -i <binary nii file determined regions> -o <output txt path>
  • Extract 3D mesh Fractal Dimension: extract fractal dimension of gray level mesh. The input data and input region should be in pair.python fractal-dimension/fractal.py -m <binary nii file determined regions> -r <nii file of raw data> -o <output txt path>

• Extract Radiomics Features : parameters of radiomics are listed in radiomics/RadiomicsParams.yaml. Run python get_r_features.py m <binary nii file determined regions> -r <nii file of raw data> -o <output csv path> to get radiomics features of a region. The input data and input region should be in pair.

• LASSO Analysis : this script analysis radiomics features using LASSO. python plot_lasso_mse.py -i <input csv file >

• Abnormal Locating : we fine-tune the trained model in slices from only COVID-19 positives in order to train a model to locate abnormal slices in COVID-19 positive volumes. Test the model using python multi_period_scores/get_abnormal_scores.py and visualize the results using python analysis_mp.py. Some of our patients have multi-period CTs, the visualization of abnormal slices shows the changes of lesion with time.

Citation

If you find this project helpful, please cite our paper:

@article {OpenCovidDetector,
	author = {Jin, Cheng and Chen, Weixiang and Cao, Yukun and Xu, Zhanwei and Zhang, Xin and Deng, Lei and Zheng, Chuansheng and Zhou, Jie and Shi, Heshui and Feng, Jianjiang},
	title = {Development and Evaluation of an AI System for COVID-19 Diagnosis},
	year = {2020}, doi = {10.1101/2020.03.20.20039834},journal = {medRxiv}}