lingyunshao / chestx-ray14_cnn

Chest X-ray diagnosis prediction using Deep Learning

Chest disease classification using Convolutional Neural Networks

This project will investigate a set of convolutional neural networks, to identify diseases on previously unseen chest radiographs. MobileNet (V1), ResNet50, VGG16 and modified versions of the first two models are evaluated, to find and propose the most suitable architecture for the problem. The ChestX-ray14 (Wang et al., 2017) dataset provides a large collection of images in fourteen categories.

Feature maps and class activation maps are generated, to gain further insight into which features are predominant during the classification process. Transfer learning is used as a means to try and improve model performance.

The test results demonstrate, that a shortened version of MobileNet is the most suitable model for the task, and significant diagnostic predictions can be made with the proposed network architecture. Transfer learning proved to be beneficial and offered increased classification performance, regardless of the apparent domain discrepancy.

A full report of the findings is available in the following dissertation: https://drive.google.com/open?id=115NfM_HbA3DD2OePtvNM2nGgdZ93Cfg5

To run the jupyter notebook, which contains the project:

• Download a 64-bit version of Python

• Open console at .ipynb location

• Acquire necessary libraries: "pip install tensorflow keras scikit-learn scipy Pillow pandas matplotlib jupyter"

• Run "jupyter notebook" and open the project

To import the best performing model:

• Download trained models: https://drive.google.com/open?id=1pVtfHhocxyK3h_7b86RKaYe_8lxXkRZL

• Unzip contents to Sessions folder

• Run "Load session model" cell with "_Mobile_transfer" session parameter

• Run "Import session model" cell

.

Dataset from: https://nihcc.app.box.com/v/ChestXray-NIHCC