AAPM DL-Sparse-View CT 挑战赛提交报告:为几何未知扇束 CT 设计迭代网络
机器学习
2021-06-02 v1 数值分析
图像与视频处理
数值分析
医学物理
摘要
本报告简要阐述我们参与 AAPM DL-Sparse-View CT 挑战赛(队名:“robust-and-stable”)的动机与方案描述。任务是利用数据驱动重建技术从有限视角扇束测量中恢复乳腺模型体模图像。该挑战的独特之处在于:参与者获提供真实图像及其无噪声、欠采样正弦图(以及相关的有限视角滤波反投影图像)的集合,但未获提供实际前向模型。因此,我们的方法首先在数据驱动的几何校准步骤中估计扇束几何。在随后的两步流程中,我们设计了一个迭代端到端网络,可计算近乎精确的解。
引用
@article{arxiv.2106.00280,
title = {AAPM DL-Sparse-View CT Challenge Submission Report: Designing an Iterative Network for Fanbeam-CT with Unknown Geometry},
author = {Martin Genzel and Jan Macdonald and Maximilian März},
journal= {arXiv preprint arXiv:2106.00280},
year = {2021}
}
备注
This is a technical report of a method participating in a not yet finished challenge. Therefore, it does not contain any final results. In particular, the reported reconstruction errors are only with respect to our own validation split of the provided training data. Once the official challenge report is released, these values will be updated with the results from the actual test set