• The primary goal of ClinicIMPPACT is conducting a multi-center clinical study involving Radio-Frequency Ablation (RFA). Due to many factors, such as legislative regulation or site-specific circumstances, workflows vary between single medical institutions.


• Capturing the uncertainties arising from such deviations is crucial in user-centered development of a planning application responsible for guiding the interventional radiologist (IR) during the technical processes involved in the study.


• We thoroughly analyze the clinical workflows of all involved institutions and identify a general procedure, while capturing possible uncertainties and exceptions as optional steps in the resulting planning system.

• We develop an integrated software package which comprises all necessary sub-modules for planning and evaluation of RFA in liver malignancies. The application optimally exploits the resources of a single PC for all inherent computations.


• We aim at using widely available consumer hardware, such as high-performance GPUs, while concurrently achieving best performance.


• Sophisticated algorithms allow us to cut down on the time required to execute complex tasks, while optimized user interfaces decrease the manual effort for the IRs. This strategy leads to easy integration into the clinical routine, since only a standard PC needs to be set up in clinical sites.

• Operating the planning system should interfere with the clinical procedure as little as possible. However, predicting the outcome of RFA treatment requires many parameters.


• We require a patient-specific anatomical model comprising organ boundaries, vessel trees and tumors segmented from multiple scans. Additionally, accurate identification of the RFA probe from intraoperative scans increases the prediction of the induced lesion.


• The resulting issues are manifold. For minimal interference, we aim at a high degree of automation of the involved algorithms. Furthermore, patient motion and breathing deformation between stages of the clinical workflow is inevitable.


• Hence, we develop fast registration and segmentation toolchains to guarantee smooth operation of our software during all stages of the workflow.

• Patient-specific parameters such as tissue perfusion and needle position strongly influence the accuracy at which we can predict the lesion location and extent. The clinical study intends to accompany interventions.


• As a result, we need to optimize the performance of the simulation algorithms not only in terms of accuracy, but also computational speed. Overall, simulating a lesion should, in no case, require more time than actual treatment so we can immediately compare the result of actual treatment with our prediction.

Visual guidance and validation of results are core parts of the project. For fast diagnosis and planning, we implement high-performance algorithms for concurrently visualizing multiple datasets. Additionally, data such as simulation results contain multiple arrays of values, e.g. temperature or cell death probability.

Our research focuses on customizable representation of such data for analysis of the inter-dependence of patient-specific parameters. Understanding these connections allows for a reasoning process during retrospective analysis of cases, ultimately resulting in improved understanding of the bio-mechanical processes involved in RFA treatment. Further, prospective exploration of the parameter space of RFA becomes possible by visual analysis of simulation results. Due to the fast simulation process, comparing the results of different parameterizations and picking the optimal subset possibly leads to improved results of the procedure.

Our techniques encompass improved 2D visualization in connection with radiological viewers and novel 3D techniques for visual analysis of treatment parameters. The proposed methods provide novel 3D rendering schemes for overview analysis, and enhance the detailed validation process by reducing tedious manual measurements to visual analysis at-a-glance Figure 1, and introduce new techniques for multivariate analysis of the results of simulated treatment Figure 2.