Visuelle Analyse in der Epidemiologie

公共卫生研究从问卷调查和临床测量中产生了大量描述人们生活方式和健康状况的异构数据。在格赖夫斯瓦尔德大学，开展了一项长期研究，以进行流行病学研究，并调查正常和病理变异的患病率和类型。数据采集包括来自大约9000个受试者的图像和非图像数据。这些数据中还包括SPP项目第一阶段调查的动态医学图像数据。为了充分利用这些数据在预防医学方面的潜力，需要复杂的分析和探索技术。由于可能的分析问题的范围很大，目前还不知道其全部程度，分析必须灵活，高效和直观，以实现长时间的可重复性。此外，可视化分析必须包括特征生成的一部分，而不是分析特征空间。将制定程序，使来自医学图像和非图像数据以及人口数据的数据可视化和分析。对于同一数据池的不同子集上的不同研究问题，方法应具有良好的可扩展性，因为这是来自不同临床学科的研究人员从事其项目的通常情况。我们将扩展在前一个资助期开发的工具，以分析人群的模式，而不是个体的病理模式。我们将让用户参与分析的早期阶段，包括特征的生成，这通常是预处理的一部分。

Research in Public Health generates a huge amount of heterogeneous data characterizing the lifestyle and health status of people from questionnaires and clinical measurements. At the University of Greifswald, a long-term study is carried out for epidemiological studies and for investigation of prevalence and type of normal and pathological variation. Data acquisition comprises image and non-image data from approximately 9000 subjects. Among these data are also dynamic medical image data which were investigated in the first stage of the SPP project. To fully exploit the potential of these data in preventive medicine, sophisticated analysis and exploration techniques are necessary.Since the range of possible analysis questions is large and at present unknown to its full extent, analysis must be flexible, efficient and intuitive to enable repeatability over a long period of time. Furthermore, visual analytics must include part of the feature generation instead of analyzing the feature space.Procedures will be developed for visualizing and analyzing data from medical image and non-image data as well as from demographical data. The methods shall scale well with respect to different research questions on different subsets of the same data pool, since this is the usual scenario if researchers from different clinical disciplines pursue their projects. We will extend tools developed in the previous grant period to analyze patterns in the population as opposed to individual patterns of pathology. We will involve the user at an early stage of analysis including the generation of features which is usually part of the preprocessing.