Model-based hepatic percutaneous microwaveablation planning. First validation on a clinical dataset

Abstract

A model-based planning tool, integrated in an imaging system, is envisioned for CT-guided percutaneous microwave ablation. This study aims to evaluate the biophysical model performance, by comparing its prediction retrospectively with the actualablation ground truth from a clinical data set in liver. The biophysical model uses a simplified formulation of heat depositionon the applicator and a heat sink related to vasculature to solve the bioheat equation. A performance metric is defined toassess how the planned ablation overlaps the actual ground truth. Results demonstrate superiority of this model predictioncompared to manufacturer tabulated data and a significant influence of the vasculature cooling effect. Nevertheless, vasculatureshortage due to branches occlusion and applicator misalignment due to registration error between scans affects the thermalprediction. With a more accurate vasculature segmentation, occlusion risk can be estimated, whereas branches can be usedas liver landmarks to improve the registration accuracy. Overall, this study emphasizes the benefit of a model-based thermalablation solution in better planning the ablation procedures. Contrast and registration protocols must be adapted to facilitate itsintegration into the clinical workflow.

Figure 1.

Comparison between manufacturer and biophysical model prediction of the ablation volume, influence of the vascular fraction. Histogram of relative distance from the identity line $\frac{(y-x)}{x}$. Tukey 95% confidence interval range test between cases with vascular fraction above 5% and the rest of the cohort. Left column a shows results without vascular cooling effect incorporated in the model. Right column b shows results with vasculature cooling effect incorporated in the model

Figure 2.

Biophysical model performance compared with manufacturer data and related to ablation ground truth. Scatter plot a corresponds to the relative ablation volume difference with ground truth |ΔV|/V_GT. Scatter plot b corresponds to the Dice coefficient difference between model and manufacturer data as a function of the vascular fraction. Lower left corner c shows an example of outlier (patient P006) with overestimated ground truth volume and high vascular fraction. Lower right corner shows two examples of patients with superior model performance and high vascular fraction (d: patient P001, e: patient P003).

Figure 3.

Whisker plots applied to the ablation performance indicators: volume relative difference with ground truth |ΔV|/V_GT, Hausdorff distance from ground truth and Dice coefficient. Comparison between manufacturer and model prediction. Upper row a shows results before excluding cases with overestimated ground truth volume and cases with apparent misaligned applicator (with respect to the ground truth). Lower row b shows results after applying those exclusion criteria on the cohort.

Figure 4.

P006 patient: registration correction based on vasculature landmarks and evidence of vasculature shortage. Upper row a shows a poor match (Dice = 0.47) between the ablation predicted from the biophysical model and the final ground truth, and misaligned intra-operative (dark green) and post-operative (light green) vasculatures. Intermediate row b shows a better match (Dice = 0.57) between the ablation predicted from the biophysical model and the final ground truth, after realignment of the intra-operative (dark green) and post-operative (light green vasculatures). In the lower row, the final ground truth is shown on the postoperative scan (image c), whereas the 3D visuals d of the final ground truth (red), intermediate ground truth (pink) and biophysical model prediction (blue) evidence the vasculature shortage effect.

Figure 5.

Contrast dosage and contrast phase effect on the vasculature segmentation. Upper row shows different contrast dosages for three patients (a: patient P007, pre-ablation, 90 ml, b: patient P007, post-ablation, 70 ml, c: patient P011, 30 ml, d: patient P018, 53 ml). Lower row shows different contrast phases for two patients (e: patient P003, f: patient P008)

Figure 6.

Retrospective analysis pipeline for the biophysical model validation.

Figure 7.

Metric to quantify the thermal ablation performance. Upper row a shows the absolute Hausdorff distance calculation from the predicted ablation to the ground truth. Lower row b shows the vascular fraction estimation.