In this project, a large dataset that contains chest X-rays are analyzed. The dataset can be found here. The architecture shown here is used to improve the accuracy of previous work by other researchers.

Here is a sample from the dataset.

We can evaluate the snapshot models by looking at the saliency maps in this figure. The red points highlighted on the chest X-ray images are the points with greater gradients thus the model puts a focus on during prediction. Our L-Vit model achieved a precision rate of 0.8825 and a recall rate of 0.815. The code repositories for this project can be found here.