Synergistic joint variational neural networks for PET-MR image reconstruction with generative modelling priors

利用生成建模先验进行 PET-MR 图像重建的协同联合变分神经网络

• 批准号: 2269756
• 金额: --
• 依托单位: King's College London
• 依托单位国家: 英国
• 项目类别: Studentship
• 财政年份: 2019
• 资助国家: 英国
• 起止时间: 2019 至 无数据
• 项目状态: 已结题
• 来源: https://gtr.ukri.org/projects?ref=studentship-2269756
• 关键词: Synergistic; joint; variational; neural; networks

This project will redesign PET and MR regularised image reconstruction algorithms as synergistically-connected variational neural networks (VNNs), and furthermore make use of deep generative models as priors for each of the VNNs. Both networks will be connected at each layer in order to permit synergistic reconstruction of both PET and MR images simultaneously. The overall aim is to deliver enhanced, synergistic image quality benefits for both PET and MR reconstruction, and furthermore these benefits have potential to make PET-MR imaging faster, cheaper and even safer (due to reduced radiation doses).

该项目将重新设计PET和MR正则化图像重建算法作为协同连接的变分神经网络（VNN），并进一步利用深度生成模型作为每个VNN的先验。两个网络将在每层连接，以便同时进行PET和MR图像的协同重建。总体目标是为PET和MR重建提供增强的协同图像质量优势，此外，这些优势有可能使PET-MR成像更快，更便宜，甚至更安全（由于减少了辐射剂量）。