COVID-19 Diagnosis Tool

A COVID-19 Diagnosis Tool for identifying COVID-19 positive patients based on lung CT scans, and determine the area of infections on patient's lung CT scans.

Dataset

The raw data are in .nii format with masks. The .nii files are converted to .png files using the MATLAB Script under data_preprocessing. The description of the raw data used can also be found inside the same folder.

Models

Classification Model

A pre-trained ResNet18 + 3-layer Fully-connected layers ANN is used for testing whether the given CT scan belongs to a COVID-19 positive patient. The test accuracy is 92%.

Segmentation Model

A U-net is used for labelling the area of infections on the CT scans.

Training

Setup the environment and install the dependencies required:

python setup.py

Baseline Model

python baseline.py

Model

python train.py

Some Segmentation Examples

Left: Model Prediction

Middle: Ground Truth

Right: Original CT Scan