Low- and Zero-dose Contrast-enhanced MRI Using Deep Learning

项目摘要 研究动机：基于Gd的造影剂(GBCA)被用于大约一年 第三，核磁共振扫描。GBCA独特的松弛参数创造了不可或缺的 图像对比度可广泛应用于临床，如血管造影术和肿瘤 侦测。然而，GBCA的使用与发展 肾源性系统性纤维化(NSF)。NSF可能会很痛苦，会导致严重残疾，而且 甚至是死亡。发生NSF的风险可以预防数百万晚期患者 慢性肾脏疾病(CKD)接受增强MRI检查。最近的 对大脑和身体中的Gd沉积的鉴定增加了额外的 关于使用GBCA的安全问题。研究表明，信号增加 平扫T1加权MR图像上与先前相关的强度 GBCA暴露，这种Gd滞留与肾功能无关。而当 最初的报告集中在线性GBCA上，最近的报告显示Gd 大循环GBCA也会发生沉积，尽管水平较低。FDA已经 最近发布了对比剂增强磁共振成像后Gd滞留的警告， 并要求GBCA制造商进行人体和动物研究，以进一步 评估这些造影剂的安全性。本项目通过以下方式解决这些问题 利用人工智能开发低剂量和零剂量增强磁共振成像 (AI)和深度学习(DL)。 方法：这个快速通道项目有两个阶段和三个目标。目标1(第一阶段)是 开发一种可以合成全剂量对比剂增强磁共振图像的DL方法 预对比度图像和对比度增强图像仅占标准图像的10% GBCA剂量。将构建一个软件基础设施，以无缝集成DL 磁共振扫描仪和PACS之间的软件。目标2(第二阶段)是开发一种动态链接法 可以使用不含GBCA的合成全剂量对比增强的磁共振图像 以不同的图像对比度进行收购。在AIM 3(第二阶段)中，我们将进行临床验证 并对低剂量和零剂量DL方法进行评估，包括对轻度- 至中度CKD。非劣性测试和综合诊断性能 将进行全剂量图像与真实全剂量图像的比较。 意义：这项工作将导致更安全的对比度增强MRI。低剂量和 零剂量对比剂增强MRI方法不仅将使数百万患者受益 晚期CKD，目前不能接受增强MRI检查，但还有更多 肾功能正常的患者，术后有Gd滞留的风险 增强磁共振成像。

Project Summary Motivation: Gadolinium-based contrast agents (GBCAs) are used in approximately a third of all MRI scans. The unique relaxation parameters of GBCAs create indispensable image contrast for a wide range of clinical applications, such as angiography and tumor detection. However, the usage of GBCAs has been linked to the development of nephrogenic systemic fibrosis (NSF). NSF can be painful, cause severe disability, and even death. The risk of developing NSF prevents millions of patients with advanced chronic kidney disease (CKD) from receiving contrast-enhanced MRI exams. The recent identification of gadolinium deposition within the brain and body has raised additional safety concerns about the usage of GBCAs. Studies have demonstrated increased signal intensity on the unenhanced T1-weighted MR images that is correlated with previous GBCA exposure, and this gadolinium retention is independent of renal function. While initial reports focused on linear GBCAs, more recent reports show that gadolinium deposition occurs with macrocyclic GBCAs as well, albeit at lower levels. FDA has recently issued warnings about gadolinium retention following contrast-enhanced MRI, and required GBCA manufacturers to conduct human and animal studies to further assess the safety of these contrast agents. This project addresses these concerns by developing low-dose and zero-dose contrast-enhanced MRI using artificial intelligence (AI) and deep learning (DL). Approach: This fast-track project has two phases and three aims. Aim 1 (Phase I) is to develop a DL method that can synthesize full-dose contrast-enhanced MR images using pre-contrast images and contrast-enhanced images acquired with only 10% of standard GBCA dose. A software infrastructure will be constructed to seamlessly integrate the DL software between MR scanners and PACS. Aim 2 (Phase II) is to develop a DL method that can synthesize full-dose contrast-enhanced MR images using GBCA-free acquisitions with different image contrast. In Aim 3 (Phase II), we will clinically validate and evaluate both low-dose and zero-dose DL methods, including on patients with mild- to-moderate CKD. Non-inferiority tests and diagnostic performance of the synthesized full-dose images compared to the true full-dose images will be performed. Significance: This work will lead to safer contrast-enhanced MRI. The low-dose and zero-dose contrast-enhanced MRI method will benefit not only millions of patients with advanced CKD, who cannot currently undergo contrast-enhanced MRI, but many more patients with normal kidney function, who are at the risk of gadolinium retention after contrast-enhanced MRI.