Pneumonia Classification in Pytorch

This is a Pytorch based convolutional neural network for detecting pneumonia in frontal-view chext X-ray images.

Dataset

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).

Prerequisites

• Python 3.5+
• Pytorch and its dependencies

Usage

1. Clone this repository.

2. Download chest X-ray images from kaggle.

3. Run model_stn.ipynb

Results

• Base model (resnet 34) without dropout in the fully connected layer achieves a test accuracy of 92.79%.

  • run model_no_dropout.ipynb

• Model with dropout in the fully connected layers achieves a test accuracy of 95.03%.

  • run model_dropout.ipynb

• Model that uses a spatial transformer network before the convolutional neural network achieves a test accuracy of 95.51%.

  • run model_stn.ipynb