Abstract
Automatic left ventricle segmentation from cardiac MRI is crucial for quantitative assessment of cardiac function. Traditional approaches often use single-scale analysis, which may miss important features at different spatial scales. In this work, we propose a multi-scope convolutional neural network approach that leverages information at multiple scales for improved left ventricle segmentation.
Our method introduces a multi-scope architecture that processes cardiac images at different spatial resolutions simultaneously. The framework consists of three key components: (1) multiple scope modules that capture features at different scales, (2) a feature fusion mechanism that combines multi-scale information, and (3) a segmentation network that leverages the fused features for accurate delineation.
We evaluate our approach on cardiac MRI datasets with expert annotations. Experimental results demonstrate that our multi-scope approach significantly improves segmentation accuracy compared to single-scale methods. The method shows excellent performance in capturing both fine details and global structure of the left ventricle.
The proposed framework represents a significant advancement in cardiac image analysis, providing more accurate segmentation that could improve cardiac function assessment and patient outcomes.
BibTeX
@inproceedings{li2018multi,
title={A Multi-Scope Convolutional Neural Network for Automatic Left Ventricle Segmentation from Magnetic Resonance Images: Deep-Learning at Multiple Scopes},
author={Li, Xinyi and Wang, Yuanyuan and Yan, Wenjun and Van der Geest, Rob J and Li, Zeju and Tao, Qian},
booktitle={International Congress on Image and Signal Processing, BioMedical Engineering and Informatics (CISP-BMEI 2018)},
year={2018},
doi={10.1109/CISP-BMEI.2018.8633185}
}