This is a Pytorch based convolutional neural network for detecting pneumonia in frontal-view chext X-ray images.
The project uses data from the Chest X-Ray Images dataset on kaggle.
There are 5,863 chest X-Ray images (JPEG) and 2 categories (Pneumonia/Normal).
- Python 3.5+
- Pytorch and its dependencies
-
Clone this repository.
-
Download chest X-ray images from kaggle.
-
Run
model_stn.ipynb
-
Base model (resnet 34) without dropout in the fully connected layer achieves a test accuracy of 92.79%.
- run
model_no_dropout.ipynb
- run
-
Model with dropout in the fully connected layers achieves a test accuracy of 95.03%.
- run
model_dropout.ipynb
- run
-
Model that uses a spatial transformer network before the convolutional neural network achieves a test accuracy of 95.51%.
- run
model_stn.ipynb
- run