Fully Automated 4D Echocardiographic Mitral Valve Analysis for Surgical Repair

描述（由申请人提供）：二尖瓣反流（MR）影响2.4%的美国成年人，即使是轻微的，也会增加死亡率，严重程度与生存率降低之间存在强烈的分级关系[2,3]。二尖瓣小环成形术修复已成为功能性和退行性MR的首选手术治疗方法。然而，最近的几项长期研究表明，术后显著MR的复发率出乎意料地高。[5-9]提高瓣膜修复的临床效果需要术前对患者特异性体内瓣膜形态进行彻底的分析，以预测哪些患者将从瓣膜修复中获益，并确定针对患者特异性瓣膜几何形状扭曲的修复策略。实时三维超声心动图（rt-3DE）是在手术室检查活体瓣膜形态和功能最实用的工具。然而，目前检查rt-3DE图像数据的方法既效率低下，又限制了向手术团队传达的信息量。因此，本提案的目标是开发并验证一种全自动四维时空分割方法，该方法可以从rt-3DE图像数据中自动生成二尖瓣的动态几何模型。假设这些模型可以准确而稳健地捕捉活体二尖瓣的动态形态。为了研究这一假设，将在Specific Aim 1中构建专家标记的二尖瓣rt-3DE图谱数据库。这些地图集编码了正常和患病受试者群体中动态瓣膜形态的信息，并作为4D自动分割的训练数据。在Specific Aim 2中，将使用Specific Aim 1中构建的参考地图集实现全自动二尖瓣小叶的4D分割。该算法结合互补概率分割和形状建模技术，从rt-3DE图像中自动生成患者特定的四维二尖瓣形状模型。在具体目标3中，将使用手动图像分析作为金标准来验证4D分割算法。将评估该方法的准确性和再现性，并优化算法以提高性能效率。成功完成本项目后，外科医生将拥有有效的术前瓣膜评估工具，为指导二尖瓣修复手术提供前所未有的可视化和定量数据。

DESCRIPTION (provided by applicant): Mitral regurgitation (MR) affects 2.4% of adult Americans [1] and increases mortality even when mild, with a strongly-graded relationship between severity and reduced survival.[2,3] Mitral valve repair with undersized ring annuloplasty has become the preferred surgical treatment for both functional and degenerative MR. However, several recent long-term studies have documented unexpectedly high recurrence rates of significant MR after surgery.[5-9] Improving the clinical outcome of valve repair requires a thorough pre-operative analysis of patient-specific in vivo valve morphology, to predict which patients will benefit from valve repair over replacement and to identify repair strategies that target patient-specific distortions in valve geometry. Real-time 3D echocardiography (rt-3DE) is the most practical tool for inspection of in vivo valve morphology and function in the operating room. However, the current methods for examining rt-3DE image data are both inefficient and limited in the amount of information conveyed to the surgical team. Therefore, the goal of this proposal is to develop and validate a fully automated 4D spatiotemporal segmentation method that automatically generates dynamic geometric models of the mitral valve from rt-3DE image data. It is hypothesized that these models can accurately and robustly capture the dynamic morphology of the in vivo mitral valve. To investigate this hypothesis, a database of expert-labeled rt-3DE atlases of the mitral valve will be constructed in Specific Aim 1. These atlases encode information about dynamic valve morphology in a population of normal and diseased subjects and serve as training data for 4D automatic segmentation. In Specific Aim 2, a fully automatic 4D segmentation of the mitral leaflets will be implemented using the reference atlases constructed in Specific Aim 1. The proposed algorithm integrates complementary probabilistic segmentation and shape modeling techniques to automatically generate 4D patient-specific shape models of the mitral valve from rt-3DE images. In Specific Aim 3, the 4D segmentation algorithm will be validated using manual image analysis as a gold standard. The accuracy and reproducibility of the method will be assessed, and the algorithm will be optimized for performance efficiency. Upon successful completion of this project, surgeons will have an efficient tool for pre-operative valve assessment that will provide unprecedented visual and quantitative data for guidance of mitral valve repair surgery.