zhiqin1998 / covid19-detection

COVID-19 Detection with Radiography Images

This is a project assignment for course WID3008 Image Processing. The objective of the project is as follow:

1. To classify whether a given chest x-ray image is infected by COVID-19 or is a normal chest x-ray image
2. To evaluate the accuracy of COVID-19 classification for chest x-ray images

More detailed information at the Jupyter Notebooks

Requirements

pip install -r requirements.txt or conda env create -f conda_env.yml

Datasets

Datasets are obtained from Kaggle

Below are a few training examples:

COVID-19

NORMAL

Data Preprocessing

The preprocessing steps are defined as follow:

Below is a few images comparison showing the output of CLAHE:

Model Definition

We use the InceptionV3 base model and add our own classifier dense layer to it to predict our binary image class. Training steps are as follow:

1. Freeze base InceptionV3 model and train network head with Adam optimizer
2. Unfreeze a few inception block and train using SGD with low learning rate

Results

Epoch Plots

Confusion Matrix

ROC Curve

Guided GradCAM

References

1. https://www.kaggle.com/tawsifurrahman/covid19-radiography-database
2. https://keras.io/api/applications/
3. https://github.com/ismailuddin/gradcam-tensorflow-2