Contrast-free Deep Myocardial Tissue Characterization with Cardiac MR Fingerprinting

心血管疾病（CVD）是西方世界发病率和死亡率的主要单一原因，在英国每年造成超过65.000人死亡。磁共振成像（MRI）是一种重要的非侵入性工具，用于CVD的风险评估、治疗指导和治疗监测。磁弛豫特性（例如T1和T2弛豫时间）的定量映射已经被开发，目的是使心肌组织特性的定量测量标准化，从而实现患病和健康组织的非侵入性表征和区分。几项临床研究表明，组织特异性参数（如T1、T1 rho、T2和T2* 弛豫时间以及细胞外容积（ECV）和脂肪分数（FF））可改善CVD的评估。然而，定量心脏MRI仍然受到一些挑战。一个主要的局限性是，尽管有前途的定量组织定征，这些地图通常是网站和供应商特定的，由于几个模型简化和MR系统相关的混杂因素。这些标测图使用不同的MRI序列（造影剂注射前/后）顺序采集，可能处于不同的运动状态（由于生理运动）。此外，由于参数间依赖性引入的误差，一次映射单个参数可能导致不准确的量化。所有上述结果在长扫描时间（限制切片和参数估计的数量），并产生负面影响的再现性，分析和解释的参数map.Cardiac磁共振指纹（MRF）最近出现作为一种方法，快速，同时量化多个组织的属性（例如T1和T2）。然而，还需要几个发展，使强大的和可重复的无造影剂心肌组织表征的多个参数与心脏MRF。当前心脏MRF方法的局限性包括：1）仅量化T1和T2（以及最近的FF），但是大量额外的心肌组织信息（例如T1 rho、T2*）可以进一步了解潜在的CVD，2）加速心脏MRF所需的图像重建方法导致计算时间较长，目前阻碍了临床转化。3)MRF中所需的字典生成和匹配的计算负担随着定量参数的数量呈指数级增加，因此目前只有少数同时参数用心脏MRF进行定量。4)在体内已经观察到T1和T2相对于传统标测技术的偏差，这可以通过几个混杂因素来解释，这些因素目前不包括在心脏MRF模型中，以及5）可重复性和再现性研究有限，这是为定量心脏MRI提供标准化框架的基本步骤。强大而全面的多参数定量心脏MRF方法，可通过单次高效扫描实现可再现的同时T1、T2、T1 rho、T2* 和FF标测。此外，我们将研究所提出的方法是否提供了获得全面的心肌组织定征的可能性，而不需要额外的造影后成像。将研究基于深度学习（DL）的运动校正、重建、字典生成和匹配，以实现在约15- 18秒/切片内采集多个准确的标测图，以及考虑MRF框架中的多个参数和混淆因素所需的计算可扩展性。所提出的方法将在两个不同的临床研究机构的标准化模型、健康受试者和CVD患者中进行验证。

Cardiovascular disease (CVD) is the leading single cause of morbidity and mortality in the Western world, causing over 65.000 deaths every year in England. Magnetic Resonance Imaging (MRI) is an important non-invasive tool for risk assessment, guidance of therapy and treatment monitoring of CVD. Quantitative mapping of magnetic relaxation properties (such as T1 and T2 relaxation times) have been developed with the aim of standardizing the quantitative measurement of myocardial tissue properties, enabling non-invasive characterization and differentiation of diseased and healthy tissue. Several clinical studies have shown the potential of tissue specific parameters such as T1, T1rho, T2 and T2* relaxation times as well as extracellular volume (ECV) and fat fraction (FF) to improve the assessment of CVD. However, quantitative cardiac MRI still suffers from several challenges. A major limitation is that despite promising quantitative tissue characterization these maps are usually site- and vendor-specific due to several model simplifications and MR system-related confounding factors. These maps are acquired sequentially with different MRI sequences (before/after contrast injection) and potentially at different motion states (due to physiological motion). Furthermore, mapping a single parameter at a time can lead to inaccurate quantification due to errors introduced by inter-parameter dependencies. All of the above results in long scan times (limiting the number of slices and parameters estimated) and negatively affects reproducibility, analysis and interpretation of the parametric maps.Cardiac Magnetic Resonance Fingerprinting (MRF) has recently emerged as an approach to rapidly and simultaneously quantify multiple tissue properties (e.g. T1 and T2). However, several developments are yet needed to enable robust and reproducible contrast-free myocardial tissue characterization of multiple parameters with cardiac MRF. Limitations of current cardiac MRF approaches include: 1) quantification of only T1 and T2 (and more recently FF), however a wealth of additional myocardial tissue information (e.g. T1rho, T2*) could enable further understanding of the underlying CVD, 2) image reconstruction methods required to accelerate cardiac MRF result in long computational times, currently impeding clinical translation. 3) The computational burden of dictionary generation and matching required in MRF increases exponentially with the number of quantitative parameters, thus only few simultaneous parameters are currently quantified with cardiac MRF. 4) Biases in T1 and T2 with respect to conventional mapping techniques have been observed in-vivo, which may be explained by several confounding factors, which are currently not included in the cardiac MRF model, and 5) repeatability and reproducibility studies are limited, which is a fundamental step to provide a standardised framework for quantitative cardiac MRI.The proposed project will overcome these problems by developing a novel, robust and comprehensive multiparametric quantitative cardiac MRF approach to enable reproducible simultaneous T1, T2, T1rho, T2* and FF mapping from a single and efficient scan. Furthermore, we will investigate whether the proposed approach offers the possibility of deriving comprehensive myocardial tissue characterization without the need of additional post-contrast imaging. Deep-learning (DL) based motion correction, reconstruction, dictionary generation and matching will be investigated to enable the acquisition of multiple accurate maps in ~15-18s/slice as well as the computational scalability needed to account for several parameters and confounding factors in the MRF framework. The proposed approach will be validated in standardised phantoms, healthy subjects and patients with CVD in two different clinical research Institutions.