3D Cardiac Motion in Experimental Models of Pulmonary Arterial Hypertension

通过人工智能方法建模MRI数据：右心室（RV）功能是多种心肺疾病（包括肺动脉高压（PAH））生存的独立预测因子。心脏磁共振成像（CMR）已被推荐为临床评估右心室质量、体积和功能的金标准。然而，由于右心室的复杂形态和几何形状，从CMR图像中手动分割心脏具有挑战性，并且可重复性差。我们的临床团队正在开发人工智能（AI）技术来预测PAH患者的生存，并研究常见和罕见的遗传变异如何影响右心生理。同时，本博士项目提出建立人工智能工具来生成肺动脉高压啮齿动物模型心脏的三维运动模型。在我们使用2D CMR数据进行初始机器训练的基础上：1)我们将获得高分辨率3D和2D CMR图像，以跟踪单核碱（MCT）和缺氧（SuHx）大鼠的PAH疾病进展。手动注释将用于训练用于自动分割大鼠心脏的深度学习模型和创建3D高分辨率地图集；2)我们将在图谱上代表大鼠特定的心脏运动，以确定区域和时间心脏功能；3)我们计划利用这些模型跟踪和了解适应的遗传修饰因子（如BMPR2），并检测干预措施（如代谢调节剂）对RV重塑的影响；4)最后，我们将使用该管道验证在包括UK Biobank在内的大型人群中使用并行计算框架的发现。

Modelling MRI Data by an Artificial Intelligence Approach:Right ventricular (RV) function is an independent predictor of survival in a broad variety of cardiopulmonary diseases including pulmonary arterial hypertension (PAH). Cardiac magnetic resonance imaging (CMR) has been recommended as the gold standard for RV assessment in clinic, e.g. RV mass, volume and function. However, manual cardiac segmentation from CMR images is challenging and poorly reproducible due to the complex morphology and geometry of the RV. Our clinical team is developing Artificial Intelligence (AI) techniques for the prediction of PAH patient survival and investigating how common and rare genetic variants influence right heart physiology. In parallel, this PhD project proposes to build up the AI tools to generate three-dimensional (3D) motion models of the hearts of pulmonary hypertension rodent models. Building on our initial machine training using 2D cine CMR data: 1) we will acquire high-resolution 3D and 2D cine CMR images to follow the PAH disease progression in the monocrotaline (MCT) and Sugen-hypoxia (SuHx) rats. Manual annotations will be performed for training a deep learning model for automated segmentation of rat hearts and for creating a 3D high-resolution atlas; 2) we will represent rat-specific cardiac motion on the atlas to determine regional and temporal heart function; 3) we plan to use the models to follow and understand the genetic modifiers of adaptation (e.g. BMPR2) and detect the effects of interventions (e.g. metabolic modulator) on RV remodelling; 4) lastly, we will use this pipeline to validate discoveries made using a parallel computational framework in large human populations that include UK Biobank.