4D Multi object segmentation based on MR image sequences - Medical application for evaluation of myocardial differences in shape and function after infarction

心肌梗死（MI）是西方文明中世界范围内男性和女性的主要死亡原因。心肌梗死患者的生活质量和病程取决于心肌的恢复和避免心脏持续性功能障碍性收缩的发展，这可能导致心脏功能的进行性损害和心脏重塑。早期检测有重塑风险的患者具有临床意义，因为必须尽早启动有效的治疗以避免重塑。这项工作的目的是开发新的自动方法，用于基于时空MRI数据集检测、量化和预测心肌重塑。因此，必须开发一种新的工作流程，以实现数据集的自动预处理、基于模型的分割和4D MRI中的运动场估计以及定量参数提取、分析和可视化。为了开发和评价，存在MI患者的基线和随访MRI数据集的综合数据池。所有数据集均采用标准成像参数采集。也存在具有高时间和/或空间分辨率的健康受试者的MRI数据。本项目共有360多个匿名MRI数据集。它们包含多个MRI序列（例如Cine-MRI、LGE-MRI、T2 w-MRI）和相关心脏结构的手动分割。这项工作的一个中心方面是开发基于模型的方法。该方法使用了一种新的集成的分割和运动估计的左，右心室的方法，并考虑了形状和形状的变化，以及典型的心脏运动的知识。另一方面是临床MRI数据集的定量分析和分类。这里，使用基于学习的分类算法。因此，提取并分析表征心脏的形状和运动的相关参数，从而提供心肌重构的自动检测和预后。

Myocardial infarction (MI) is the leading cause of death for both men and women worldwide in the western civilization. The quality of life and the course of disease for MI patients depend on the revitalization of the myocardium and avoiding the development of a persistent dysfunctional contraction of the heart, which can lead to progressive impairment of the heart function combined with cardiac remodeling. Early detection of patients with risk of remodeling is clinical relevant because effective therapies have to be initiated early to avoid remodeling.The aim of this work is to develop new automatic methods for detection, quantification and prediction of myocardial remodeling based on spatio temporal MRI datasets. Therefore, a new workflow has to be developed that enables automatic pre-processing of the datasets, a model based segmentation and motion field estimation in 4D MRI as well as quantitative parameter extraction, analysis and visualization. For development and evaluation a comprehensive data pool of baseline and follow-up MRI dataset of MI patients exists. All datasets were acquired with standard imaging parameters. MRI data of healthy subjects with high temporal and/or spatial resolution exists as well. Overall more than 360 anonymized MRI datasets are available for this project. They contain multiple MRI sequences (e.g. Cine-MRI, LGE-MRI, T2w-MRI) and manual segmentations of the relevant cardiac structures.A central aspect of this work is the development of a model based approach. This method uses a new integrated segmentation and motion estimation of the left and right ventricle approach and considers knowledge about shape and shape variations as well as the typical motion of the heart. Another aspect is the quantitative analysis and classification of clinical MRI datasets. Here, learning based classification algorithms are used. Therefore, relevant parameters characterizing shape and motion of the heart are extracted and analyzed providing an automatic detection and prognosis of myocardial remodeling.