Semi-automatic segmentation and landmark identification in cardiac 4D flow MRI

在这个项目中，我们探索了分割和识别心脏解剖的重要组成部分和执行流量测量所需的先进计算和数学方法，这对于设计软件工具来处理、分析、可视化和探索4D流量MRI数据集是必不可少的。MRI技术提供了一种非侵入性的工具，用于量化和可视化心脏和周围血管的解剖和生理，有助于心脏病的诊断。传统上，MRI数据集以2D图像堆栈的形式提供，代表3D体积。心脏4D血流MRI是一项相对较新的技术，它提供时变的体积数据集，呈现一系列3D图像。先前获取4D流MRI数据集的技术需要更长的扫描时间（20至40分钟），这使得它们在许多情况下令人望而却步。随着新的4D流MRI数据采集技术的出现，我们的工业合作伙伴可以访问，这些数据集的扫描时间已经减少到大约6分钟。这项尖端技术为无创检测心脏异常提供了极为有用的工具性数据集，增加了心血管疾病的早期诊断机会，心血管疾病是全世界的主要死亡原因（世界卫生组织，2013年3月）。心脏解剖结构的复杂性使得对其重要组成部分如瓣膜、心室尖顶等的自动分割和识别困难且不可靠。另一方面，人工分割和识别这些心脏成分是一项耗时的任务，必须由熟练的心脏成像仪执行。因此，本计划旨在提供有用的互动工具和合适的技术来分析和解释这些数据集，这将有助于减少相关放射科医生的工作量。

In this project, we explore advanced computational and mathematical methods required for segmenting and identifying important components of cardiac anatomy and performing flow measurements, which are essential for designing software tools to process, analyze, visualize, and explore 4D flow MRI datasets. MRI techniques provide a noninvasive tool for quantifying and visualizing the anatomy and physiology of heart and peripheral vessels, helping in the diagnosis of heart diseases. Traditionally, MRI datasets have been made available as stacks of 2D images, representing 3D volumes. Cardiac 4D flow MRI is a relatively new technology that provides time-varying volumetric datasets, presenting a series of 3D images.Prior techniques for acquiring 4D flow MRI datasets required longer scan-times of 20 to 40 minutes, making them prohibitive in many cases. With the advent of new 4D flow MRI data acquisition technology, accessible by our industrial partner, the scan-time for such datasets has been reduced to around six minutes. This cutting edge technology provides extremely useful and instrumental datasets for noninvasive detection of anomalies in the heart, increasing early chances of diagnosis for cardiovascular diseases, which comprise the leading cause of death worldwide (World Health Organization, March 2013).The complexity of cardiac anatomy makes the automatic segmentation and identification of its important components, such as the valves, ventricular apexes, etc., difficult and unreliable. On the other hand, the manual segmentation and identification of such cardiac components is a time-consuming task that must be performed by a skilled cardioimager. Consequently, this project aims to provide useful interactive tools and suitable techniques to analyze and interpret such datasets, which will contribute to reducing the workload of involved radiologists.