Home

Awesome

PF-Net: Pulmonary Fibrosis Segmentation Network with Multi-Scale Guided Dense Attention

This repository provides source code of PF-Net for pulmonary firbrosis segmentation proposed published on IEEE TMI 2021. If you use this code, please cite the following paper:

pfnet_result

pf_net The structure of PF-Net. It combines 2D and 3D convolutions to deal with images with anisotropic resolution. For example, the in-plane resolution is around 4 times of through-plane resolution in our dataset, and we use 2D convolutions for the first two levels and 3D convolutions for the other levels in the encoder. Multi-Scale Guided Dense Attention is introduced in the decoder to deal with lesions with various positions, sizes and shapes.

Requirements

Train and Test

python net_run.py train config/pfnet.cfg
python net_run.py test config/pfnet.cfg