Update from 14/09/22: Published in IEEE Intelligence Systems, May-June 2022, volume 37, pages 54-64

Update from 05/12/21: To appear in IEEE Intelligent Systems

Citation of preprint on medRxiv:

@article {Ter-Sarkisov2021.02.16.21251754,
	author = {Ter-Sarkisov, Aram},
	title = {One Shot Model For COVID-19 Classification and Lesions Segmentation In Chest CT Scans Using LSTM With Attention Mechanism},
	year = {2021},
	doi = {10.1101/2021.02.16.21251754},
	publisher = {Cold Spring Harbor Laboratory Press},
	journal = {medRxiv}
}

Citation of journal publication:

@article{9669072,
  author={Ter-Sarkisov, Aram},
  journal={IEEE Intelligent Systems}, 
  title={One Shot Model for COVID-19 Classification and Lesions Segmentation in Chest CT Scans Using Long Short-Term Memory Network With Attention Mechanism}, 
  year={2022},
  volume={37},
  number={3},
  pages={54-64},
  doi={10.1109/MIS.2021.3135474}}

One Shot Model Using LSTM with Attention

Segmentation And Classification Results From The Paper:

Segmentation Results (CNCB-NCOV Segmentation Dataset, (http://ncov-ai.big.ac.cn)

# Affinities	AP@0.5	AP@0.75	mAP@[0.5:0.95:0.05]
One Shot +LSTM+Attention	0.605	0.497	0.470
Mask R-CNN	0.502	0.419	0.387

Classification Results (CNCB-NCOV Classification Dataset, (http://ncov-ai.big.ac.cn)

# Model	COVID-19	CP	Normal	F1 score
One Shot + LSTM+Attention	95.74%	98.13%	99.27%	98.15%
ResNet50	91.04%	97.64%	98.97%	96.88%
ResNeXt50	91.94%	88.45%	84.30%	87.31%
ResNeXt101	91.58%	92.13%	94.02%	92.86%
DenseNet121	92.64%	96.16%	98.98%	96.15%

Classification Results (iCTCF-CT Classification Dataset, (http://ictcf.biocuckoo.cn)

# Model	COVID-19	Normal	F1 score
One Shot + LSTM + Attention	97.73%	98.68%	98.41%
VGG16(Ning et al, 2020)	97.00%	85.47%	-

The Model:

Attention Layer:

LSTM + Attention:

Due to the size of the backbone (ResNext101+FPN), we provide the second-best model, with ResNext50+FPN backbone.

To train the model, simply run

python train.py

on the CNCB-NCOV data. You need both segmentation and classification splits, see https://github.com/AlexTS1980/COVID-Single-Shot-Model for details. To evaluate the provided model, change the path in eval.py before running:

python eval.py

AlexTS1980/COVID-LSTM-Attention