Author: Wei Dong (wdong@wdong.org)
This repository contains the dockerized version of my best single (non-ensembled) lung cancer model for Data Science Bowl 2017 (cross validation loss around 0.425).
You need nvidia-docker to run the program.
Usage:
./process.sh input output
input: an input directory containing dicom files (input/*.dcm).
output: output directory where results (HTML) are written to.Output format is designed to be viewed with a web browser on the local machine. Because the output contains all dicom files and a 3D lung mesh model and is quite large, it is not suitable to be served on a web server.
More models and training code are upcoming.
(Images and samples from the SPIE dataset.)