Pytorch

• The original dataset has 930 images with around 66% of the images being covid positive.
• However, the dataset was not very clean and many fields were empty.
• After cleaning up the data (detailed explanation is given in the comments of the code), I was left with 668 images which I split into a training set (90%) and validation/testing set (10%).

In order to account for meta-data such as age, gender, view position, I designed my own classifier.

Steps

Feature extractor from VGG16 -> Flattened the output -> Concatenated with the meta-data -> Used custom classifier.

A detailed explanation of model and results is given in "Report_Results.pdf"