COVID-19

CNN model that diagnoses using Chest X-ray medical images

DISCLAIMER: This model was designed and implemented for educational purposes. it MUST NOT be used for medical diagnosis as it was NOT tested by a field expert.

Introduction:

CoVID19 is a disease caused by the newly identified virus SARS-CoV-2 of the Corona Viruses family¹. The disease's incubation period is 14 days, with most cases being between 4-5 days². The clinical presentation starts with coughing, fever, dyspnea, and bilateral infiltrates on chest imaging. The presentation can be severe in a lot of cases, especially in old people³. Radiological features on chest x-ray of confirmed COVID-19 cases were parenchymal abnormalities, specifically peripheral consolidations (accumulation of fluids)⁴; although Chest X-rays weren't so inclusive.

1. World Health Organization. Director-General's remarks at the media briefing on 2019-nCoV on 11 February 2020. Here
2. Li Q, Guan X, Wu P, Wang X, Zhou L, Tong Y, et al. Early Transmission Dynamics in Wuhan, China, of Novel Coronavirus–Infected Pneumonia. New England Journal of Medicine. 2020Jan29;
3. Wu Z, Mcgoogan JM. Characteristics of and Important Lessons From the Coronavirus Disease 2019 (COVID-19) Outbreak in China. Jama. 2020Jan24;
4. Yoon SH, Lee KH, Kim JY, Lee YK, Ko H, Kim KH, et al. Chest Radiographic and CT Findings of the 2019 Novel Coronavirus Disease (COVID-19): Analysis of Nine Patients Treated in Korea. Korean Journal of Radiology. 2020Jan26;21(4):494.

I would like to thank Dr.Mohammed for helping me understand this disease and participating in this paragraph. Also pyimagesearch blog for inspired me.

Dataset:

1. The posstive cases of COVID-19 X-ray images taken from Dr. Cohen in his repo
2. The normal cases taken from dataset that's published in Kaggle's dataset Here.

Model:

I designed 2 different models of CNN, one of them contains 4 layers with padding and relu as an activation function after every 2 layers there is a max-pooling layer, then two fully connected layers with relu and softmax as an activation function. the second one contains 2 Convolution layers with relu as an activation function follows by the max-pooling layer. The optimizer for both models is Adam and binary_crossentropy as a loss function.

Result & Predection

The accuracy for both models is above 95% on the validation set. 1st Model and 2ed Model:

• Predection using 2nd model:

To Do:

• Testing the model.
• Calculate the confusion matrix.
• Try on CT scan imeages.
• Try ResNet model on this problem.

Contributing

Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change.