ChestX-ray-Prediction

Prediction on diseased chest based on the NIH dataset.

Raw data source

This released dataset, chestX-ray, contains 14 kinds of chest diseases.

Box: https://nihcc.app.box.com/v/ChestXray-NIHCC images refers to the images (after downloading, you need to extract them first) Data_Entry_2017.csv contains the necessary information related to this method, like the disease distribution

Baselines

(Original paper of ChestX-ray is: http://openaccess.thecvf.com/content_cvpr_2017/papers/Wang_ChestX-ray8_Hospital-Scale_Chest_CVPR_2017_paper.pdf)

Three baselines (AlexNet, LeNet, VGG-16) can be found at /models

Running data_extract.py

This method is to make the trainSet.tfrecords and testSet.tfrecords

Python data_extract.py image_path csv_path train_test_choice dataset_choice

image_path: the path of your downloaded images

csv_path: the path of Data_Entry_2017.csv file

train_test_choice: if you plan to make trainSet, you need to set it as 1; if you plan to make testSet, you need to set it as 2

dataset_choice: if you don't plan to make the dataset with gender as label, you only need to leave it as 1; otherwise, you should set it as 2

Output files( train_test_choice is 1 or 2, dataset_choice is 1):

a. tfrecord files (14 diseaseName_train_healthyordisease.tfrecords or 14 diseaseName_test_healthyordisease.tfrecords) b. log file, which contains the disease name and the number of its diseased & healthy images. (train_log.txt or test_log.txt)

Running model.py

This method contains three baselines: AlexNet, LeNet, VGG-16

Python modelName.py train_disease_images train_healthy_images train_disease_images test_healthy_images disease_name path

train_disease_images: (can be found in log file) The number of diseased images of the selected trainSet

train_healthy_images: (can be found in log file) The number of diseased images of the selected trainSet

test_disease_images: (can be found in log file) The number of healthy images of the selected testSet

test_disease_images: (can be found in log file) The number of healthy images of the selected testSet

disease_name: selected one disease from 14 kinds of diseases

path: the path of all the tfrecord datasets generated