Abstract
Infant brain MRI reconstruction presents unique challenges due to the rapid development and small size of infant brains, often requiring thin-section imaging that can result in noisy or incomplete data. Traditional reconstruction methods often struggle with the specific characteristics of infant brain imaging. In this work, we propose a novel deep generative adversarial network approach for thinsection infant MR image reconstruction.
Our method leverages the power of generative adversarial networks to produce high-quality reconstructed images from limited or noisy infant brain MRI data. The framework consists of three main components: (1) a generator network that learns to reconstruct high-quality images, (2) a discriminator network that ensures realistic output, and (3) a specialized loss function that preserves infant brain characteristics.
We evaluate our approach on a comprehensive dataset of infant brain MRI scans with various imaging protocols. Experimental results demonstrate that our GAN-based method significantly improves image quality compared to traditional reconstruction approaches. The method shows particular effectiveness in handling the unique challenges of infant brain imaging, including rapid developmental changes and small anatomical structures.
The proposed framework represents a significant advancement in infant MRI reconstruction, providing higher quality images that could improve diagnostic accuracy and reduce the need for additional imaging procedures.
BibTeX
@article{gu2019deep,
title={Deep Generative Adversarial Networks for Thinsection Infant MR Image Reconstruction},
author={Gu, Jiaqi and Li, Zeju and Wang, Yuanyuan and Yang, Haowei and Qiao, Zhongwei and Yu, Jinhua},
journal={IEEE Access},
year={2019},
publisher={IEEE},
doi={10.1109/ACCESS.2019.2918926}
}