This repository contains the relevant core demo for fetal brain age estimation and anomaly detection based on the center slice using deep ensembles with uncertainty.
- Python 3.6
- Tensorflow >=1.10.0
- Keras >=2.2.4
To train the network, make sure you have the following files serialized using pickle and located in the path below. The default shape of our network input is (N_subject,192,192,1) while output shape is (N_subject,). Or just revise the code wherever you want to fit your data format.
./data/train_data.p
./data/train_label.p
./data/validation_data.p
./data/validation_label.p
Please use the command below to train the network. The model will be automatically saved in the default path./save_models/. Search for an optimal learning rate and it won't take much time to train.
python main.py --action train
To predict the fetal brain age based on the deep ensembles, please use the command below. The trained networks named as 'demo_0.h5', 'demo_1.h5', etc. are preferably loaded from ./models/ .
python main.py --action predict
Use --help to see other usage of main.py.
@article{SHI2020117316,
title = "Fetal brain age estimation and anomaly detection using attention-based deep ensembles with uncertainty",
journal = "NeuroImage",
volume = "223",
pages = "117316",
year = "2020",
issn = "1053-8119",
doi = "https://doi.org/10.1016/j.neuroimage.2020.117316",
url = "http://www.sciencedirect.com/science/article/pii/S1053811920308028",
author = "Wen Shi and Guohui Yan and Yamin Li and Haotian Li and Tingting Liu and Cong Sun and Guangbin Wang and Yi Zhang and Yu Zou and Dan Wu"
}
Please feel free to contact me or open an issue if you have any question. E-mail: allard.w.shi at gmail.com