Repository files navigation

This repository presents an ensemble based transfer learning approach for accurately classifying common thoracic diseases from Chest X-Rays (CXRs)

We use theCheXpert dataset for training and evaluation

Thefinal ensemble consists of the following models

• DenseNet-121
• DenseNet-169
• DenseNet-201
• Inception ResNet v2
• Xception

The ensemble weights are found empirically while the disease-wise optimal prediction thresholds are found by maximizing theYounden's J Statistic

The ROC curves for each individual model and the final ensemble are locatedhere
We achieve a mean area under the curve (AUC) of0.915 on the validation set, that comes close to the SOTA of0.94 (at the time of writing these models, i.e., May 2020)

Harshit Varma,Richeek Das,Ankit Kumar Misra,Prapti Kumar