Uncertainty in Medical Visualization
Gordan Ristovski · 2017
The many steps in the medical visualization pipeline introduce some amount of uncertainty based on errors or assumptions. The rendered images presented to the user, typically a medical expert, most often omit this information. Since medical experts rely on these visualizations to bring conclusions about the existence and severity of a potential disease or to come up with a treatment plan, conveying the uncertainty that is inherent to the displayed information becomes a crucial part towards a better, more informed decision making. To do so, we describe different types of uncertainties mathematically, which allows us to abstract them from the underlying source and application. We then develop a taxonomy for uncertainty types and identify the biggest challenges when conveying uncertainty through the visualization methods. We convey the uncertainty information in one of the uncertainty types by rendering a single, non-obstructive, most likely surface and encode the variability around it by using color or textures. We then couple this with interactive investigation methods, which allow for a detailed analysis of the uncertainties. Our visualization methods are applied to a concrete medical application - stenoses assessment in vascular structures and are evaluated with medical experts, who showed a clear preference for our solution. We also investigated using probabilistic ensembles of different segmentation outputs to reduce the final uncertainty. We then looked into the uncertainty coming in from the simulation parameters in radiofrequency ablation. We developed methods to address the uncertainty in the simulation output to two different target groups - medical experts and simulation experts. To the former, we show the uncertainty using perceptually uniform, intuitive uncertainty glyphs. For the latter, we integrated tools in our system that allow for an interactive analysis of the impact of different parameters and different parameter ranges on the simulation result.