Towards semantically steered navigation in shape spaces exemplified by rodent skull morphology in correlation to external attributes

形态计量学在生物学中的一项主要任务是发现外在属性（如遗传、地理和生态数据）与一类生物的特定形状之间的相关性。这需要将数据分析结果从属性空间映射到形状空间的技术，反之亦然，允许用户在交互式反馈循环中改进形状空间和属性空间中的数据分析。本项目建立在计算机科学和生物学之间已有的研究合作基础上，我们为啮齿动物下颌骨的CT数据建立了一个线性形状空间。结合最先进的体可视化技术，将属性空间的分析结果映射到形状空间，可以沿着形状空间中有意义的方向（特征）进行交互式导航。在这个项目中，我们想把这个研究延伸到四个方面。首先，利用当前系统的导航技术检测到的形状变化将被量化，以便进行形状和遗传因素的相关分析。其次，我们的目标是一个分层形状模型，允许分离粗和细尺度形状变化。第三，为了使用户能够将人类知识纳入分析过程，我们计划集成单个形状的交互式变形技术。这将允许在形状空间中沿着相应的方向导航，并使用结果来细化属性空间中的分析。此外，可以通过编辑后的形状有目的地扩展形状空间。最后但并非最不重要的是，我们计划加强形状空间分析从刚性到可变形的形状。为此，我们将啮齿动物尾巴作为相关分析的额外重要数据源，并计划应用谱嵌入方法来提出等距不变形状表示。

A major task of morphometrics in biology is to discover correlations between extrinsic attributes like genetic, geographic and ecological data and the specific shape of a class of organisms. This requires techniques for mapping data analysis results from attribute space to shape space and vice versa allowing the user to refine the data analysis in shape space as well as in attribute space within an interactive feedback loop. This project builds on an already existing research cooperation between Computer Science and Biology in which we established a linear shape space for CT data of rodent mandibles. Together with state-of-the-art volume visualization techniques the mapping of analysis results in attribute space to shape space allows the interactive navigation along meaningful directions (traits) in shape space. In this project we want to extend this research towards four aspects. First of all shape variations detected using the navigation techniques of the current system will be quantified for the purpose of genetic mapping enabling correlation analysis of shape and genetic factors. Second we aim for a hierarchical shape model that allows to separate coarse and fine scale shape variations. Third, to enable the user to incorporate human knowledge into the analysis process we plan to integrate interactive deformation techniques of individual shapes. This will allow navigating along corresponding directions in shape space and using the results to refine the analysis in attribute space. In addition, the shape space can be extended purposefully by the edited shape. Last but not least we plan to enhance the shape space analysis from rigid to deformable shapes. For this aim we include rodent tails as additional important data source for correlation analysis and plan to apply spectral embedding methods to come up with an isometric invariant shape representation.