Uncertainty in Medical Visualization: Bringing Imaging and Simulation Uncertainty Together

可视化方法已经成为支持诊断、治疗计划和术中辅助的临床常规的组成部分。医学可视化是基于某些假设来创建的，并且通常不会使医学专家意识到这些假设，这些假设可能导致所显示的图片与实际情况的潜在偏差。因此，医学专家经常将这种可视化理解为真实的图像，至少在一定程度上是基于这种图像做出决定的。然而，从配准和分割任务的实际图像采集到最终渲染的医学可视化管道包括许多潜在的误差源。为了进行更有教养的决策，需要定量地捕捉这些错误概率的影响，并将其直观地传达给医学专家。这就是不确定性可视化的任务。在第一个资助期内，我们开发了对出现的不确定性进行严格建模并对其进行可视化编码的方法，从而形成了用于不确定性感知决策的交互式可视化分析系统。除了使用来自医学成像数据的信息外，还可以通过使用来自模拟生物物理或医疗过程的特定于患者的信息来增强系统，从而进一步改进决策。另一方面，这种模拟也是基于不确定的成像数据，甚至做出进一步的假设。因此，他们的结果也包含可能与临床相关的不确定性。我们的目标是将成像不确定性和模拟不确定性结合在一个视觉分析系统中。利用第一个资助期的不确定性感知图像分割，我们建议使用概率输入开发不确定性感知的患者特定模拟，使用深度学习方法预测模拟结果，并通过交互可视化从分割和模拟中对患者特定数据进行不确定性感知分析。这些方法应在可视化发挥关键作用的医疗应用中使用。

Visualization methods have become an integral part of clinical routine supporting diagnosis, treatment planning, and intraoperative assistance. The medical visualizations are created based on certain assumptions and typically do not make the medical experts aware of those assumptions which may result in potential deviations of the shown picture from the actual situation. Hence, the medical experts often perceive andinterpret the visualization as a true image, on which decisions are based, at least, in part. However, the medical visualization pipeline ranging from the actual image acquisition over registration and segmentation tasks to the final rendering includes many potential sources of errors. To allow for a more educated decision making, the impact of those error probabilities need to be quantitatively captured and visually conveyed to themedical expert. This is the task of uncertainty visualization. Within the first funding period, we developed a rigorous modeling of the appearing uncertainties and methods for visually encoding them, leading to interactive visual analysis systems for uncertainty-aware decision making. Beside using information derived from medical imaging data, decision making can further be improved by enhancing the system with patient-specific information from simulating biophysical or medical processes. Such simulations, on the other hand, are also based on the uncertain imaging data and even make further assumptions. Thus, their results also contain uncertainty potentially bearing clinical relevance. Our goal is to bring together imaging uncertainty and simulation uncertainty in a visual analysis system. Using the uncertainty-aware image segmentation of the first funding period, we propose to develop uncertainty-aware patient-specific simulations using probabilistic inputs, predictions of the simulation outcomes using deep-learning methods, and interactive visualizations for an uncertainty-aware analysis of patient-specific data from segmentations and simulations. The methods shall be employed within medical applications, where visualization plays a crucial role.