Contributors: Abhilash Kuhikar, Karen Sanchez, Shyam Narasimhan Please see Final Report for implementation details
To build a platform for automatically diagnosing a patient with tuberculosis and detecting the relevant symptoms from CXRs using Machine Learning techniques. We used Grad-cam for visual explanation from Deep Networks to identify the regions of Chest X-ray where the network concentrates for learning the classification.
The source code for grad-cam was used from here: https://github.com/jacobgil/pytorch-grad-cam
We attempt to make the network to learn from CXR from a radiologist's perspective
- Classification using VGGNet
- Classification using Own CNN
- Deriving latent features from a network designed for Semantic Segmentation of lungs and using them for classification
- Annotating the chest regions for abnormalities and detecting them
- Generating clinical readings from the CXRs
- Building Visualization pipeline more robust to different changes in network architectures