We implemented a visualization application supporting interventional radiologists during analysis of simulated minimally invasive cancer treatment.

Visualization-guided evaluation of simulated treatment

The current clinical practice employs only rudimentary, manual measurement tools.
Our system provides visual support throughout three evaluation stages:

Stage 1
Starting with determining prospective treatment success of the simulation parameterization.
In case of insufficiencies,
Stage 2
includes a simulation scalar field for determining a new configuration of the simulation.
For complex cases, where Stage 2 does not lead to a decisive strategy,
Stage 3
reinforces analysis of interdependencies of scalar fields via bivariate visualization.

We conducted a user study with 9 interventional radiologists, which revealed the disadvantages of manual analysis in the measurement stage of evaluation and highlight the demand for computer-support through our system.

The analysis consists of three stages. Hence, we provide a visual metaphor for each subsequent step. Visualizations of the patient image and outlines of tumor and predicted coagulated region reflect the basic framework on which we build our advanced methods. Completely failed treatment is comparatively easy to spot, since the outline of the coagulation zone does not fully enclose the tumor. Hence, we focus our considerations on cases which cover the tumor entirely, but the safety margin may be violated.

A typical use-case of our visual analysis technique for evaluating simulations of minimally invasive cancer treatment. An interventional radiologist evaluates the safety margin of dead tissue around a tumor after simulated treatment (a) and encounters critical areas (orange and blue segments). Zooming in (b) unveils underlying patient data via a levels of detail-based approach and reveals a potential vessel (bright area). The radiologist decides to include the tissue temperature field from simulation via iso-contours (c) to further examine the finding. For final decision, the radiologist then analyses the dependence of blood temperature and tissue vulnerability (d). We provide a categorized, texture-based iso-contour representation, in this case categorizing into low and high values. After the radiologist identifies the cause of the issue, she updates the simulation parameters accordingly and tries to destroy the problematic vessel.

Stage 1 | Initially, we aid the process of evaluating success of a simulation configuration by removing the need for manual measurements. As previously stated, MICT demands destroying the tumor plus a rim of healthy tissue around it. However, severe over-treatment, i.e., coagulating large healthy portions of the organ, should be avoided. Therefore, we visualize the distance between simulated coagulation area and tumor in relation to the respective outlines.

Stage 2 | In Stage 1, the user possibly identifies regions of the predicted coagulated region, which do not satisfy the safety margin. In Stage 2, we additionally employ a customizable approach based on colored iso-contours for visualizing a single scalar field from the simulation. For example, emitted power or temperature are often informative during analysis of such failed regions. In combination with the Stage 1 visualization, the user can either immediately infer a new parameterization of the simulation or proceed with determining more complex interactions in Stage 3.

Stage 3 | Parameter interdependencies often make the behavior of energy-based MICT hard to predict, and a single scalar field is often insufficient for analyzing the potential reasons for failure. In the final stage, we add a second scalar field from the simulation for concurrent analysis of both variables and their mutual interaction. We introduce textural elements to achieve a categorization of the second input variable into multiple ranges.

Analysis of uncertainty using visualization

Many of the input parameters for simulating RFA treatment cannot be measured exactly. Further, inaccuracies in segmentation and registration may lead to slight deviations between simulation and reality. To alleviate this issue, we implemented means for sampling the parameter space of the simulation domain. This includes, but is not limited to, slightly varying the single needle tips or perfusion values. However, simple methods for presenting the results to the user are difficult to analyze. Consecutively, we implemented a visualization technique that presents the results of statistical analysis in a more elaborate way.

The user might be interested in, e.g., average values, which parameterizations surpass a vessel, or whether the ensemble of simulations splits up into groups locally. Therefore, we perform statistical tests to determine parameters such as medians and standard deviations, and further analyze these to present outliers and grouping.

Showing the result of multiple simulations resulting from parameters pace sampling with standard techniques. It is difficult for a user to determine the overall behavior of certain parameterizations, reason about which parameterization may be most effective, or estimate the influence of single parameters for a given case.

Simplification of many simulation results leads to a much better overview for the user. In this case, a vessel is embedded into the simulation domain and, as per our results, difficult to ablate. The black line shows the median of all simulations (50 in this case). The saturated blue region depicts the 1-sigma deviation range from the median, whereas the bright blue region represents the 3-sigma range. In green, we show local outliers.

Upon zooming in, we switch the representation to group mode. Here, we deepen the analysis of the visualization ensemble and determine whether distinct groups of simulations exist, based on medians and standard deviations. Besides showing the borders of these groups, we also present additional information, such as how many parameterizations fall within a group.