Reducing magnetic resonance image spacing by learning without ground-truth

作者：

Highlights：

• MR images are usually acquired with large inter-slice spacing in clinical practice.

• A deep-learning-based algorithm to reduce slice spacing for better rendering.

• No real high-resolution ground-truth is required for training neural-networks.

• Outperform current self super-resolution algorithms for MR images.

摘要

•MR images are usually acquired with large inter-slice spacing in clinical practice.•A deep-learning-based algorithm to reduce slice spacing for better rendering.•No real high-resolution ground-truth is required for training neural-networks.•Outperform current self super-resolution algorithms for MR images.

论文关键词：Generative adversarial network,Magnetic resonance imaging,Super-resolution,Variational auto-encoder

论文评审过程：Received 19 June 2020, Revised 15 February 2021, Accepted 4 June 2021, Available online 13 June 2021, Version of Record 1 July 2021.

论文官网地址：https://doi.org/10.1016/j.patcog.2021.108103