Automated Model Design and Benchmarking of 3D Deep Learning Models for COVID-19 Detection with Chest CT Scans

Accepted in AAAI-2021.

@article{He2021CovidNet3D, 
  title={Automated Model Design and Benchmarking of 3D Deep Learning Models for COVID-19 Detection with Chest CT Scans}, 
  journal={Proceedings of the AAAI Conference on Artificial Intelligence}, 
  author={He, Xin and Wang, Shihao and Chu, Xiaowen and Shi, Shaohuai and Tang, Jiangping and Liu, Xin and Yan, Chenggang and Zhang, Jiyong and Ding, Guiguang}, 
  year={2021}
}

Dependences

pip install -r requirements.txt

Datasets

• CC-CCII: Zhang, K., Liu, X., Shen, J., Li, Z., Sang, Y., Wu, X., Zha, Y., Liang, W., Wang, C., Wang, K., et al.: Clinically applicable AI system for accurate diagnosis, quan- titative measurements, and prognosis of covid-19 pneumonia using computed tomography. Cell (2020)
• MosMed: Morozov, S., Andreychenko, A., Pavlov, N., Vladzymyrskyy, A., Ledikhova, N., Gombolevskiy, V., Blokhin, I., Gelezhe, P., Gonchar, A., Chernina, V., Babkin, V.: Mosmeddata: Chest ct scans with covid-19 related findings. medRxiv (2020)
• COVID-CTset: Rahimzadeh, M., Attar, A., Sakhaei, S.M.: A fully automated deep learning-based network for detecting covid-19 from a new and large lung ct scan dataset. medRxiv (2020)

Statistics

Dataset	Class	#Patients		#Scans
		Train	Test	Train	Test
CC-CCII	NCP	726	190	1213	302
	CP	778	186	1210	303
	Normal	660	158	772	193
MosMed	NCP	604	255	601	255
	Normal	178	76	178	76
COVID-CTset	NCP	202	42	202	42
	Normal	200	82	200	82

search

bash scripts/search_ct.sh

A logger directory will be created according to the logger.name in config file, with the following structure:

Supporse logger.name=MyExp：

|_output
    |_MyExp
        |_version_0 ()
            |_epoch_0.json
            |_last.pth
            |_best_acc{}_epoch{}.pth
            |_log.txt
            |_search_ct.yaml
        |_version_1()

• epoch_0.json, epoch_1.json, ..., epoch_N.json are the architectures of different epochs.
• last.pth is the latest checkpoint
• best_acc{}_epoch{}.pth is the best checkpoint
• log.txt
• search_ct.yaml is the backup config file, which will be used in the retraining stage

retrain

bash scripts/retrain_ct.sh

The commands in retrain_ct.sh are as follows:

srun -n 1 --cpus-per-task 2 python -m ipdb retrain.py \
--config_file outputs/checkpoint/version_0/search_ct.yaml \
--arc_path outputs/checkpoint/version_0/epoch_0.json  \
input.size [128,128]

You should manually set config_file and arc_path. The image size in the search stage is 64x64. Here, in the retraining stage, you should specify a larger image size.

arc_path indicates which architecture you want to retrain. You can select it based on their perfomance in the search stage.

The following directory will be created:

|_output
    |_MyExp
        |_version_0 (search stage)
            |_epoch_0.json
            |_last.pth
            |_
        |_version_0_retrain_0 (retraining stage)
            |_last.pth
            |_best_acc0.96_epoch13.pth (file name records the best acc and the corresponding epoch)
            |_othe files
        |_version_0_retrain_1 (results of other architectures if you select other architecture json file.)

Q&A

• ModuleNotFoundError: No module named 'sklearn.neighbors._base'

You may need to upgrade your scikit-learn lib.