Deep Learning-based Framework for Segmentation and Motion Tracking of Left Ventricle in 3D Echocardiography

项目总结/摘要 冠状动脉疾病仍然是世界各地的主要死因。急性心肌梗死（MI） 导致局部功能障碍，这使心脏的远端区域处于机械劣势， 长期不利的左心室（LV）重塑和并发的充血性心力衰竭（CHF）。应力 超声心动图是目前临床上建立的、具有成本效益的2D成像技术， 通过对静息时和运动后左心室成像来表征心肌损伤， 药理学诱导的应激以揭示局部缺血和/或梗塞。然而，2D的固有局限性 超声心动图使得难以表征缺血/梗塞区的整个3D体积， 定性评估室壁运动异常以表征心肌变形导致变异性 专家之间。虽然3D超声心动图有可能解决2D成像的局限性，但它不是 由于低信噪比（SNR）而在标准临床应用中被广泛接受。随着最近的进步 在深度学习算法中，许多分割和配准任务已经达到了接近专家级的准确度。 此外，以前的工作已经表明，应变分析的效用作为一种方式来量化壁运动的程度 心脏成像模式异常。尽管如此，目前的许多深度学习框架主要集中在 基于强度的特征仍然难以在3D超声心动图数据集上训练，这反过来又导致 菌株分析不良。因此，在这个奖学金，我建议开发新的数据驱动的神经网络模型 专为3D超声心动图定制，以改善左心室的分割和运动跟踪， 以实现完整的3D心脏应变分析。我的第一个目标是开发一个基于多帧注意力的神经 网络，以利用超声心动图数据集的时空特征， 左心室分割该方法将利用帧间时空特征， 增强用于分割的相关特征提取。我的第二个目标是开发一个注册神经 结合强度特征和曲面曲率弯曲的三维超声心动图网络 能量，以改善左心室的运动跟踪。这个神经网络将建立在精确的 从第一个分割的目的是在边界处包括独特的曲率能量特征，以增强 在心肌的所有区域的跟踪精度。改进的运动跟踪将被用于计算应变 用于检测完整的3D缺血/梗塞区。综上所述，本研究将提供一个客观、定量的 在3D超声心动图中使用应变分析表征室壁运动异常的工具。

PROJECT SUMMARY/ABSTRACT Coronary artery disease remains the leading cause of death around the world. Acute myocardial infarction (MI) causes regional dysfunction which places remote areas of the heart at a mechanical disadvantage resulting in long term adverse left ventricular (LV) remodeling and complicated congestive heart failure (CHF). Stress echocardiography is currently the clinically established, cost-effective 2D imaging technique for detecting and characterizing myocardial injury by imaging the left ventricle at rest and after either exercise or pharmacologically-induced stress to reveal ischemia and/or infarct. However, the inherent limitations of a 2D echocardiography make it difficult to characterize the whole 3D volume of ischemic/infarct zone, and the qualitative assessment of wall-motion abnormality to characterize myocardial deformation leads to variability among experts. Although 3D echocardiography has potential to address the limitations of 2D imaging, it is not widely accepted in standard clinical use due to the low signal-to-noise ratio (SNR). With the recent advancements in deep learning algorithms, many segmentation and registration tasks have achieved near expert level accuracy. Also, previous works have shown the utility of strain analysis as a way to quantify the degree of wall-motion abnormality in cardiac imaging modalities. Still, many of the current deep learning frameworks focus largely on intensity-based features which are still difficult to train on 3D echocardiography datasets, which in turn leads to poor strain analysis. Thus, in this fellowship, I propose to develop novel data-driven neural network models specifically tailored to 3D echocardiography to improve segmentation and motion tracking of left ventricle in order to achieve full 3D cardiac strain analysis. My first aim is to develop a multi-frame attention-based neural network to exploit the spatiotemporal features of the echocardiography dataset to improve 3D segmentation of left ventricle. This method will take advantage of the inter-frame spatiotemporal features to augment the relevant feature extractions for segmentation. My second aim is to develop a registration neural network in 3D echocardiography by combining intensity-based features and surface-curvature bending energy to improve the motion tracking of left ventricle. This neural network will build upon the accurate segmentations from the first aim to include unique curvature energy features at the boundaries to enhance tracking accuracy at all areas of the myocardium. The improved motion tracking will be used to calculate strain for detection of full 3D ischemic/infarct zones. In summary, this research will provide an objective, quantitative tools for characterizing wall-motion abnormality with strain analysis in 3D echocardiography.