Prediction of Covid-19 positive from X-rays. The images in the Dataset used are Denoised with Convolutional Variational AutoEncoder (hyperparameterized for Covid-19 Dataset) and a Convolutional Neural Network is used for classification with softmax entropy loss.
This work supports it: https://pubs.rsna.org/doi/10.1148/radiol.2020200432 and https://pubs.rsna.org/doi/10.1148/radiol.2020200642
And this recommends not using it: https://www.acr.org/Advocacy-and-Economics/ACR-Position-Statements/Recommendations-for-Chest-Radiography-and-CT-for-Suspected-COVID19-Infection
Note : This model doesn't claim any diagnostic performance. This is merely an implementation of neural networks for classification.
Read about the problems in deep learning for COVID-19 Detection in https://arxiv.org/abs/2004.12823 and https://arxiv.org/abs/2004.05405
The dataset used is publicly available at https://github.com/ieee8023/covid-chestxray-dataset. The dataset can be used for research purposes and belongs to the owner of the repository.
The dataset contains X-rays of various diseases, but the X-rays of only Covid-19 Positive are extracted. The healthy X-rays are extracted from https://www.kaggle.com/paultimothymooney/chest-xray-pneumonia. The dataset ratio of positive:healthy were kept to 50%-50%
Note : Read the earlier analysis done with this dataset in the above linked papers.
The model achieves 0.968
accuracy, 0.9877
AUC scores on train set and 0.901
accuruacy , 0.92441
AUC scores on test set after 20
epochs of training on Google Colab with GPU for ~30 mins.
The model however performs really poorly on complex structures such as in this research. A Sequential single input and single output architecture with increasing filters outperforms any other complex structures.
from src.data.extract import class_wise, train_test_wise
if __name__ == "__main__":
class_wise('src/dataset/input', 'src/dataset/classwise')
train_test_wise("src/dataset/classwise/", "src/dataset/traintest/" , 0.5, True)
python -m src.model.cnn
python -m src.gui.main