ViES: Visual-interactive Exploration for individualized Selection of relevant data regions

点击翻译按钮获取中文摘要

This project aims at supporting the diagnosis of glaucoma induced retinal damage by interactive visual means. The industrial application partner, Heidelberg Engineering GmbH, provides high-resolution data sets that need to be reduced for their assessment. Currently, only a strongly simplified automated data reduction is applied, which can lead to misinterpretations, particularly in the case of small retinal changes. These automated computations will be extended with novel visual interactive methods. Presenting the data at multiple scales, extracting and tracking specific features, as well as visual accentuation and linking will be applied to support the selection of relevant details. In this way, smaller individualized data sets are generated that are easily manageable by the physicians and allow for a comprehensive analysis of retinal sub structures. Within the scope of this project we aim at developing a framework, which supports seamless transitions between these different steps: data selection, reduction and visual analysis of details. To design the framework, we can take advantage of previously developed approaches in the DFG priority program 1335 (Scalable Visual Analytics): 1. A novel technique to visualize large multi-scale data sets. The encoding of heterogeneity values between adjacent scales provides guidance for finding interesting details. 2. A novel technique to visualize data, uncertainties, and related parameter dependencies. 3. A novel technique to extract and track features in 3-dim. data sets. These approaches will be adapted and enhanced with regard to the given application scenario - the diagnosis of retinal damage. This requires the following fundamental investigations: 1. The multi-scale visualization was developed for 1-dim. time series. Although a first extension for representing 2-dim. data exists, handling 3-dim. data requires substantial enhancements. 2. The visualization approach for data, uncertainties and parameter dependencies was designed for representing 1-dim. multi-run simulation data. Dealing with high-resolution 3-dim. data sets raises new issues and challenges. 3. The application scenario requires the specification of domain specific features and the development of novel methods for their extraction and visualization. For the reduction and visual presentation of the retinal data certain quality criteria have to be met. For example, we need to visualize both; the selected data itself, but also the applied selection criteria. For this purpose, appropriate methods will be developed.