nivation / Chexnet

Image classification: binary classification of Lung X-ray grayscale images using ChexNet

Training CheXNet Model for Lung X-ray Grayscale Images

About CheXNet

Reference code can be downloaded from here
This is a PyTorch implementation of the CheXNet algorithm for pathology detection in frontal chest X-ray images.
More details can be found in here.

Dataset (From NIH Clinical Center)

Database of chest X-ray images.
Download from: https://nihcc.app.box.com/v/ChestXray-NIHCC/folder/37178474737
Unpack archives into separate folders
images_001.tar.gz -> images_001

Usage

Train

To Train your model, remember to change class according to your own class in Main.py. In our case, it's only a binary task, so

nnClassCount = 1

You can choose whether or not change the training parameter. The default is listed:

trBatchSize = 640
trMaxEpoch = 30

Run terminal :

python3 Main.py

It should start working like:

preparing txtfile for cross validation...
fold: 1 complete
fold: 2 complete
fold: 3 complete
fold: 4 complete
fold: 5 complete

training for fold 1

Batch size: 640
Total epoch: 30

Train: 443730
Train Acc: 75280
Val: 18818
------- epoch: 1 ------------------------------------------------

[save] [2020_07_10-18:53:13]
train acc = 0.932 train loss = 0.454
val acc = 0.929 val loss = 0.214

epoch: 2
[save] [2020_07_10-21:02:08]
train acc = 0.927 train loss = 0.411
val acc = 0.925 val loss = 0.211 ...

Test

To test your model, remember to change class according to your own class in Main_testwhentraining.py. In our case, it's only a binary task, so 1 class

nnClassCount = 1

You can choose whether or not change the training parameter. The default is listed:

trBatchSize = 1

Set your own model name:

pathModel = "your model's path.tar"

Run terminal:

python3 Main_testwhentraining.py

It should start woring like:

Testing default
Batch size: 1
model loaded: your_path/model/2020_07_20-14:52:07_fullset.pth.tar
Test: 18822
Start Testing...

and you may find confusion matrix and result csvfile output at your_path/plot/

Heatmap

Open jupyter notebook file heamap2.ipynb in your_path/plot/ and rerun the code