Here, we validate and adopt our deep CNN approach, called Decompose, Transfer, and Compose (DeTraC), for the classification of COVID-19 chest X-ray images. DeTrac has achieved a high accuracy of 95.12% (with sensitivity of 97.91%, specificity of 91.87%, and precision of 93.36%) in the detection of COVID-19 X-ray images from normal, and severe acute respiratory syndrome cases.
We used 80 samples of normal CXRs from the Japanese Society of Radiological Technology and the following open source chest radiography datasets, which contains 105 and 11 samples of COVID-19 and SARS (with 4248×3480 pixels).
open source for chest radiography datasets:
https://github.com/ieee8023/covid-chestxray-datasetMatlab R2019a - window 8 or later version
- Consist of three parts :
- Run (Training_original_classes.m) matlab file on the original classes (dataset A),
- check the validation accuracy to get the significant classification performence ,
- using the learned weights to get the features for each class separetly (extract_features.m).
- Run (Pca_CXR.m) matlab file to reduce the dimension features space for each original class.
- Run (construct_dataset_B.m) to apply the K-means clustering algorithm.
- Run (Training_after_decompose.m) to apply the DeTraC model.
DeTraC_COVID19 achieved high accuracy of 95.12% which proved that CNNs have an effective and robust solution for the detection of the COVID-19 cases from CXR images and as a consequence this can be contributed to control the spread of the disease.
Table 1: COVID-19 classification obtained byDeTraC-ResNet18on chest X-rayimages.
| Accuracy | Sensitivity | Specificity | Precision |
|---|---|---|---|
| 95.12% | 97.91% | 91.87% | 93.36% |
Fig: the learning curve accuracy and loss between training and test sets.
Please do not hesitate to contact us if you have any question. asmaa.abbas@science.aun.edu.eg
If you used DeTraC and found it useful, please cite the following papers:
• Abbas A, Abdelsamea MM, Gaber MM. DeTraC: Transfer Learning of Class Decomposed Medical Images in Convolutional Neural Networks. under review 2020.
• Abbas A, Abdelsamea MM, Gaber MM. Classification of COVID-19 in chest X-ray images usingDeTraC deep convolutional neural network. under review 2020.