Abstract
Magnetic Resonance Imaging (MRI) super-resolution is crucial for improving the quality of brain imaging data, particularly when high-resolution scans are unavailable or costly to acquire. Traditional super-resolution methods often fail to preserve the complex structural relationships and spatial connections inherent in brain anatomy. In this work, we propose DeepVolume, a novel deep learning framework that incorporates brain structure and spatial connection awareness for brain MRI super-resolution.
Our approach leverages anatomical knowledge by integrating brain structure priors and spatial connectivity information into the super-resolution process. The framework consists of three key components: (1) a structure-aware encoder that captures brain anatomical features, (2) a spatial connection module that preserves inter-regional relationships, and (3) a high-resolution decoder that generates detailed brain images while maintaining structural integrity.
We evaluate DeepVolume on multiple brain MRI datasets, including T1-weighted, T2-weighted, and FLAIR sequences. Experimental results demonstrate that our method significantly outperforms existing super-resolution approaches in terms of both quantitative metrics and visual quality. The generated high-resolution images preserve fine anatomical details and maintain the spatial relationships between brain structures.
DeepVolume represents a significant advancement in brain MRI super-resolution, providing a more anatomically accurate and clinically relevant approach to improving image resolution while preserving the complex structural characteristics of brain tissue.
BibTeX
@article{li2019deep,
title={DeepVolume: brain structure and spatial connection-aware network for brain MRI super-resolution},
author={Li, Zeju and Yu, Jinhua and Wang, Yuanyuan and Zhou, Hanzhang and Yang, Haowei and Qiao, Zhuowei},
year={2019},
journal={IEEE Transactions on Cybernetics},
doi={10.1109/TCYB.2019.2933633}
}