Real-time control of radiofrequency ablation using three-dimensional ultrasound echo decorrelation imaging in normal and diseased ex vivo human liver

Abstract

Objective. Ultrasound echo decorrelation imaging can successfully monitor and control thermal ablation of animal liver and tumor tissueex vivoandin vivo. However, normal and diseased human liver has substantially different physical properties that affect echo decorrelation. Here, effects of human liver tissue condition on ablation guidance by three-dimensional echo decorrelation imaging are elucidated in experiments testing closed-loop control of radiofrequency ablation (RFA) in normal and diseased human liver tissueex vivo. Approach. Samples of normal, steatotic, and cirrhotic human liver tissue underwent RFA, targeting a 20 mm-diameter spherical ablation zone. For each tissue condition, RFA was controlled by echo decorrelation inN> 14 trials, automatically ceasing if average cumulative decorrelation within the targeted ablation zone surpassed a predetermined threshold (successfully controlled trials), or otherwise completing a standard ablation cycle of the RFA generator (unsuccessfully controlled). For comparison,N= 14 RFA trials for each tissue condition followed the RFA generator's standard algorithm without echo decorrelation feedback (uncontrolled). Receiver operating characteristic (ROC) and precision-recall curve analyses compared 3D echo decorrelation maps to segmented ablation zones. To assess effects of closed-loop control and liver condition on treatment reliability, ablation volumes, rates, and Dice coefficients for measured vs. targeted ablation zones were statistically compared among control conditions and liver types.Results. ROC curves showed effective prediction of local ablation by echo decorrelation across all liver types and control conditions (0.876 ⩽AUROC ⩽ 0.953). Successful control was significantly more frequent, ablated volumes were generally larger, and optimal echo decorrelation thresholds were smaller for normal compared to diseased liver.Significance. This study validates three-dimensional echo decorrelation imaging for monitoring and control of RFA in healthy and diseased human liver while elucidating the dependence of RFA and echo decorrelation outcomes on liver condition and resulting implications for clinical applications.

Keywords: Three-dimensional echo decorrelation imaging; cirrhosis; human liver tissue; real-time radiofrequency ablation control; steatotis; thermal tumor ablation.

Figure 1.

Experimental setup and echo decorrelation $\left(\Delta \right)$ feedback control algorithm for ex vivo human liver RFA control. (a) Experimental setup: liver tissue placed inside the white cuvette, RFA needle inserted through a guide built into the orange lid, and Z6Ms transesophageal ultrasound array fixed in a Tegaderm-covered window using the yellow 3D-printed window cover. Azimuth $\left(y\right)$, range $\left(z\right)$, and elevation $\left(x\right)$ directions are shown in yellow Cartesian coordinates. (b) Control algorithm flow chart.

Figure 2.

B-mode and echo decorrelation images and corresponding tissue histology for nine representative trials for the three control conditions and three liver types. Each row corresponds to a liver tissue condition, normal (Nor.), steatotic (Ste.), or cirrhotic (Cir.). Groups arranged from left to right represent uncontrolled (UC), successfully controlled (SC), and unsuccessfully controlled (USC) trials respectively. The color bar at the top left sub-figure ranges from −4 to −1 (log₁₀-scaled decorrelation per ms). White scale bars indicate 20 mm. For each trial, a grayscale B-mode image cross-section is overlaid with an echo decorrelation (‘hot’ colormap) at elevation zero, the corresponding tissue section closest to this elevation is shown with its manually segmented ablation zone (cyan boundaries), and 3D echo decorrelation iso-surfaces are plotted at 10^−3.2 (log₁₀-scaled decorrelation per ms) in red with the 3D ablation zone in cyan, rotated for convenience of 3D visualization. The Cartesian coordinate axes denote elevation $\left(x\right)$, azimuth $\left(y\right)$, and range $\left(z\right)$. The 20 mm-diameter control ROI boundary is plotted as a dotted green circle in the cross-sectional image panels.

Figure 3.

ROC and PR curves representing the predictive performance of 3D echo decorrelation for RFA in healthy and diseased ex vivo human liver. (a) ROC curves for uncontrolled (UC) trials. (b) ROC curves for successfully controlled (SC) trials. (c) ROC curves for unsuccessfully controlled (USC) trials. (d) ROC curves for all three liver tissue conditions across all three control conditions. (e) PR curves for the nine groups resulting from the combination of three control condition and three liver tissue conditions, with legends matching panels (a)–(c). * $p<0.05$, ** $p<0.01$, *** $p<0.001$.

Figure 4.

Ablation outcomes for uncontrolled (UC), successfully controlled (SC), and unsuccessfully controlled (USC) trials in normal (red), steatotic (blue) and cirrhotic (black) liver tissue. (a) Ablation volume (ml). (b) Ablation rate (ml min⁻¹). (c) Dice coefficient comparing measured versus targeted ablation zones. (d) Final cumulative decorrelation averaged within the spherical control ROI with diameter 20 mm. *: $p<0.05$, **: $p<0.01$.

Figure 5.

Scatter plots of measured ablation volume (top) and ablation rate (bottom) vs. log₁₀-scaled average cumulative decorrelation in the spherical control ROI with 20 mm diameter. The control threshold is shown with a gray vertical line, cutoff thresholds corresponding to 90% specificity of local ablation prediction for each liver tissue condition are marked using red, blue, and black vertical dotted lines for normal, steatotic, and cirrhotic livers, respectively, and the targeted ablation volume corresponding to a 20 mm-diameter sphere is plotted as a black horizontal line in the top panel. Lines of best fit, obtained from linear regression all plotted data points, are shown with black dashes in both panels.

Figure 6.

Scatter plots of measured ablation volume (top) and ablation rate (bottom) vs. output of a multiple linear regression model incorporating liver tissue condition as well as log10-scaled average cumulative decorrelation in the spherical control ROI. Lines corresponding to equality between predicted and measured outcomes are shown with black dashes in both panels.