This repository contains code to train and test MGI-CNN. (https://doi.org/10.1016/j.neunet.2019.03.003)
For data processing: SimpleITK, Scipy
pip install SimpleITK scipy
This code requires unzipped LUNA16 dataset. (https://luna16.grand-challenge.org/Download/)
For training: Ubuntu 16.04, Python 3.6, Tensorflow 1.10
(Optional) GPUtil
pip install GPUtil
Each fold takes about 12 hours to run 100 epochs using Nvidia GTX 1080 ti. Note that all experiments in our paper are based on 40th epoch.
For training:
python main.py --data_path=PATH --summ_path_root=PATH --fold=0 --maxfold=5 --multistream_mode=0 --model_mode=0 --train
For testing:
python main.py --data_path=PATH --summ_path_root=PATH --fold=0 --maxfold=5 --multistream_mode=0 --model_mode=0 --test --tst_model_path=PATH --tst_epoch=40
Example
--data_path=/home/jsyoon/MGICNN/dataset/
/home/jsyoon/MGICNN/dataset/raw/1.3.6.1.4.1.14519.5.2.1.6279.6001.100225287222365663678666836860.mhd
/home/jsyoon/MGICNN/dataset/raw/1.3.6.1.4.1.14519.5.2.1.6279.6001.100225287222365663678666836860.raw
...
/home/jsyoon/MGICNN/dataset/raw/candidates_V2.csv
We participated in the competition and got the following CPMs:
https://luna16.grand-challenge.org/Results/
Bum-Chae Kim, Jee Seok Yoon**, Jun-Sik Choi, and Prof. Heung-Il Suk*
Corresponding author: hisuk@korea.ac.kr
*\ For code inquiries: wltjr1007@korea.ac.kr
Machine Intelligence Lab.,\ Dept. Brain & Cognitive Engineering.\ Korea University, Seoul, South Korea.