Application of Machine Learning in Medical Image Classification

Aim

Use machine learning approaches to help preliminary diagnosis of chest radiographs: classify radiographs into 3 classes – COVID, normal, and viral pneumonia with the highest possible accuracy.

Solutions

• Used transfer learning to train several models including ResNet, DenseNet, InceptionResNet, etc. on a labeled training set.

• Used stacking to ensemble the above models to form a stronger model.

Highlights

Achieved accuracy of more than 0.98. Especially, the model achieved almost 100% precision in the COVID class.