In this Image processing project, I used a CNN classifier to classify COVID-19 Infected Lung Xray images from Healthy Lung Xray images.
I selected my dataset from three different sources:
- Cohen's COVID Chest X-ray Dataset
- Paul Mooney's Chest X-ray Dataset (Pneumonia)
- COVID-19_Radiography_Dataset
Train Dataset contains 850 images by the following distribution:
-
Pre-trained DenseNet-121 is used as the core here for our Deep Learning Model (More details here).
-
I used pre-trained weights as a means to Transfer Learning. To learn and achieve higher accuracy on our model faster.
-
Instead of freezing CNN Layers and training only the Fully Connected Layer (like most people do in Classification Task), I trained all the layers, including CNNs and Classification layer.
Test Dataset contains 200 images from each group.
Classification results:
precision recall f1-score support
0 0.73 0.95 0.83 200
1 0.94 0.66 0.77 200
accuracy 0.81 400
macro avg 0.84 0.80 0.80 400
weighted avg 0.84 0.81 0.80 400
Many thanks to Arun Pandian R for his useful tutorial.