Unsupervised Graph Domain Adaptation for Neurodevelopmental Disorders Diagnosis
Lists
Only you see this
Abstract
Unsupervised domain adaptation (UDA) methods aim to reduce annotation efforts when generalizing deep learning models to new domains. UDA has been widely studied in medical image domains. However, UDA on graph domains has not been investigated yet. In this paper, we present the first attempt of unsupervised graph domain adaptation in medical imaging, with application to neurodevelopmental disorders (NDs) diagnosis, i.e. differentiating NDs patients from normal controls. It is of great importance to developing UDA methods for NDs because acquiring accurate diagnosis or labels of NDs can be difficult. In our work, we focus on Autism spectrum disorder and attention-deficit/hyperactivity disorder which are the two most common and frequently co-occurred NDs. We propose an unsupervised graph domain adaptation network (UGDAN) that consists of three main components including graph isomorphism encoders, progressive feature alignment, and un-supervised infomax regularizer. The progressive feature alignment module is designed to align graph representations of the source and target domains progressively and effectively, while the unsupervised infomax regularizer is introduced to further enhance the feature alignment by learning good unsupervised graph embeddings. We validate the proposed method with two experimental settings, cross-site adaptation and cross-disease adaptation, on two publicly available datasets. The experimental results reveal that the proposed UGDAN can achieve comparable performance compared to supervised methods trained on the target domain.
BibTeX
@inproceedings{wang2020unsupervised,
author = {Bomin Wang and Zhi Liu and Yujun Li and Xiaoyan Xiao and Ranran Zhang and Yankun Cao and Lizhen Cui and Pengfei Zhang},
title = {{Unsupervised Graph Domain Adaptation for Neurodevelopmental Disorders Diagnosis}},
booktitle = {International Conference on Medical Image Computing and Computer-Assisted Intervention},
year = {2020},
url = {https://openreview.net/forum?id=F9wv34alER},
}Loading related content...