The outcome of thermal ablation treatment of malignant liver tumors is highly variable (1). Small and spherical tumors (≤ 3–5 cm in diameter) can benefit the most from this treatment, since larger tumors are often associated with reduced technical efficacy and higher local tumor progression rates (2–5). Moreover, for irregularly shaped tumors and tumors in proximity to major hepatic vessels and central bile ducts, the therapeutic effectiveness and safety are often compromised (6, 7). Unwanted effects of large ablation areas can manifest in an excessive treatment area with potential thermal injury to neighboring organs or major intrahepatic structures, or an unnecessary loss of surrounding healthy liver parenchyma (8). The present treatment limitations arise mainly from technical constraints, since predominantly ellipsoidal ablation volumes up to 5 cm in diameter are available with commercial ablation devices ( Figure 1 ). This set of ablation shapes does not allow precise tumor ablation in critical locations in the liver nor in the case of irregularly shaped tumors, which may restrict the applicability of thermal ablation treatment in these situations.

Traditionally, the choice of treatment strategy for malignant liver tumors depends on factors such as tumor size, number of nodules and proximity to vascular and biliary structures, rather than tumor shape (9, 10). However, the value of preserving healthy liver parenchyma has been shown through certain liver-directed therapies. For instance, non-anatomical liver resection, stereotactic body radiation therapy and irreversible electroporation can all treat tumors closely to their shape with sufficient margins (11–13). In single-probe thermal ablation, two techniques have been routinely used in the clinical practice to create more complex ablation shapes. The overlapping ablation technique (OAT) aims to provide complete coverage of larger liver tumors by sequential or simultaneous ablation of overlapping ablation volumes. The sequential approach results in a smaller, but more controllable ablation volume, whereas the simultaneous approach can achieve a large, but unpredictable ablation volume (14, 15). Furthermore, in other organs such as the thyroid, the moving-shot technique (MST) has been accepted over the past decade for treatment of thyroid nodules (16). This technique was defined as unit-by-unit ablation of thyroid nodules achieved by manual retraction of the ablation probe. In contrast to the fixed-electrode technique (FET), most commonly used for ablation of liver tumors, the MST emerged as a safer alternative aiming to prevent thermal injury to neighboring critical structures in the thyroid gland (17, 18).

The aim of this work was to remove the limitations imposed by the current state of ablation technology. The main objective was to extend the potential of thermal ablation to non-spherical and critically located tumors, by providing a comprehensive, personalized, and configurable treatment model. The model was described in a proof-of-concept study conducted by the authors in a tissue-mimicking liver model (19). It is based on a technique of configuring ablation volume shapes by robotically retracting the ablation probe along its axis while modulating the power over time and distance. The present experimental study aimed to compare the repeatability of ablation volumes created with the configurable ablation technique against the standard ablation technique, which produces mainly ellipsoidal ablation volumes. The hypothesis was that the repeatability of configurable ablation shapes is noninferior to the repeatability of the standard shapes in the tissue-mimicking model.

All 48 samples of standard ablation shapes, and all 35 samples of configurable ablation shapes were successfully prepared, ablated and post-processed. Three additional samples had to be discarded since the ablation probe had to be replaced, mainly due to issues related to the cooling system (tubing tear or clog). Segmented ablation contours from the sample cross-sections were used for the statistical analysis ( Figure 4 ). Visually, all the ablation shapes showed similar amount of repeatability around their contours. The standard ablation shapes ( Figure 4 , OVAL) appeared to be rather heart-shaped than ellipsoidal, as suggested in the manufacturer’s catalogue. The configurable ablation shapes appeared as simulated ( Figure 4 , LONG–PEAR). An explanatory video showcasing the process of configurable ablation for one of the shapes is provided in the Supplementary Material .

For the configurable shapes, the applied ablation profiles ( Figure 5 , left), the resulting ablation shapes ( Figure 5 , middle), and the measured ablation widths ( Figure 5 , right) are shown. The recorded distances travelled by the ablation probe were step functions due to profile discretization. However, the average speed per individual distance interval ( Figure 5 , marked between the dashed lines) corresponded to the planned probe speeds ( Table 1 , distance over time). The recorded output power values varied within 85–90% of the planned power values ( Table 1 , power). Sample illustrations of the resulting ablation shapes are shown in the tissue-mimicking liver model, where the ablated area was stained in magenta color, and non-ablated area in off-white color. Ablation probe marks were visible along the centerline of the samples, where the white dots correspond to the measurement points (spaced in 5 mm increments). The letters, H for head and T for tail, indicate the areas outside of the planned probe trajectories. It should be noted that the tail areas were always greater than zero, even in cases where the probe was constantly in motion ( Figures 5A, D ). On the contrary, the head areas were negligible in those cases, and only prominent when the ablation probe was static for certain amount of time ( Figures 5B, C ). Regarding median ablation shape width, the elongated shape measured 22 mm at the widest point, the hourglass shape 26 mm at the widest point and 13 mm at the narrowest point, the teardrop shape 29 mm at the widest point and 7 mm at the tip, and the pear ablation shape 24 mm at widest point and 11 mm at the tip. The width spread (interquartile range, IQR) at each measurement point maintained roughly an even trend in all ablation shapes, it ranged 1–2 mm for the elongated shape, 1 mm for the hourglass shape, 1–3 mm for the teardrop shape, and 0–2 mm for the pear ablation shape ( Figures 5A–D ). Similarly, this also applied to the length spread (IQR) at the head and tail, where the measurements ranged 0–3 mm, 1–2 mm, 3 mm, 1–2 mm respectively.

The repeatability of configurable ablation shapes was noninferior to standard ablation shapes for all ablation shape pairs (one standard, one configurable) ( Figure 6A ). The median (maximal) repeatability measured as width variability in 95% of observed shapes ranged 0.57–1.50 mm (1.35–2.95 mm) in standard and 0.75–1.20 mm (2.50–2.96 mm) for configurable shapes. Furthermore, the repeatability was observed to be noninferior in configurable (median 1.00 mm, IQR 0.70–1.44 mm) than in standard shapes (median 1.22 mm, IQR 1.00–1.50 mm), as determined by the Mann-Whitney test (p < 0.001, 95% CI ≤ -0.05 mm, Δ = -0.22 mm) ( Figure 6B ). Lastly, a linear regression analysis indicated that higher probe speeds were associated with a decrease in repeatability of the observed shapes (p < 0.001, β = -1.00 mm/s, 95% CI = [-1.33, -0.67] mm, R² = 0.08). However, a low R² value showed that other factors aside from probe speed may also have a significant impact on the repeatability ( Figure 6C ).

Standard ablation shapes deviated from an ideal ellipsoid. The shapes displayed extrusions along the lateral edges in four cases and noticeable indents in the central proximal areas in all cases ( Figures 7A, B ). Similarly, extrusions along the contour edges were also noted in six configurable ablation shapes ( Figures 7C, D ). Furthermore, standard ablation profiles produced smaller ablation volumes in the tissue-mimicking model compared to ablation volumes reported by the ablation device manufacturer in ex-vivo animal tissue ( Figure 8 ). On average, the relative differences were 21% for the ablation width, 25% for the ablation length, and 52% for the estimated ablation volume, where the ablation volume was estimated from the revolved cross-sections. Lastly, the relative difference of the sphericity indices was 1–11% across different tissue-mimicking ablation volumes compared to the ex-vivo measurement catalogue.

This study investigates the repeatability of the novel robotic approach for configurable ablation tailored to distinct tumor shapes in the liver (19). The approach was introduced as a potential solution to treatment of irregularly shaped tumors, including those in challenging intrahepatic locations, with minimal thermal damage outside the ablative margin. The uncustomary technique of dynamic and moving thermal ablation employed for shape configuration has not yet been introduced in clinical treatment of liver tumors.

The repeatability of configurable ablation shapes was observed to be noninferior to the standard ablation shapes in the tissue-mimicking liver model by less than the predefined noninferiority margin of 1 mm. Furthermore, the 95% confidence interval for the differences in median repeatability between the two groups was significantly below zero, indicating that the proposed technique is as consistent or more consistent than the standard technique. Regarding safety, the maximal shape repeatability of 3 mm for both groups was below the reported safety distance required to prevent thermal injury (5–10 mm) (24).

Regarding irregularities in created ablation shapes ( Figure 7 ), while lateral extrusions could be attributed to inhomogeneous gel formation of the tissue-mimicking model, proximal indents were likely caused by ablation probe ineffectiveness. Similarly to the heat-sink effect observed near vascular structures, the cooling liquid running through the ablation probe was presumably the cause of energy dissipation, creating indents on the top of the ablation contours. This effect was more pronounced in the standard than in the configurable shapes. When the ablation probe was static for a longer period, the thermal zone was localized around the probe. The influence of heating, however, weakened further away from the energy source. Consequently, it is presumed that the cooling of the probe had a pronounced effect on the ablation shape in the area around the probe and further away from the source. On the other hand, when the probe was moving, the effect of cooling did not play a role on the ablation shape formation, since the energy source was in motion. Evidently, the effect of applied ablation profile on ablation shape configuration was multifold. Slow and controlled retraction speeds, continuous and stable energy delivery and adequate probe cooling were essential to create the desired shapes. Moreover, it was possible to configure the shapes without continuous movement of the probe, i.e. in a stepwise motion, likely due to heat conduction which was propagating upwards in the direction of the probe and therefore connecting the “steps”.

The tissue-mimicking model used in this study represented a somewhat ideal medium due to its homogenous property and simplicity of analysis. The standard tissue model, applied in most cases of commercial ablation devices was an ex-vivo animal liver (porcine, bovine), as it is more representable of the living human tissue. Nonetheless, it was presumed that the complexity of the model could only affect the scale of repeatability of both techniques (standard vs. configurable), rather than their relative difference. Differences in material characteristics, such as thermal conductivity, diffusivity, and coagulation properties, could be the cause of discrepancy in ablation dimensions and sphericity between the models ( Figure 8 ). Unknown history of the ex-vivo samples, including factors such as the temperature, freshness, and size, may also play a role in explaining the observed variations.

In comparison to the proposed configuration model, in clinical practice, there are two existing techniques for ablation shape configuration, overlapping ablation technique (OAT) and moving-shot technique (MST), which have been successfully deployed. The OAT primarily serves to ablate larger tumors, if they are not possible to eliminate with a single ablation volume. Custom ablation shapes can be created in the process; however, they require thorough planning and accurate probe positioning, as many ablation volumes are needed to be tailored to a tumor shape (25). However, this technique is not suitable for intricate patient cases as it does not allow precise ablation control. On the other hand, the MST has been introduced with the specific reason of uncompromised safety from collateral thermal injury. Nonetheless, this technique has only been effective for treatment of thyroid nodules since the thyroid gland is easier to access with an ablation probe than the liver and can be ablated using ultrasound guidance (18). This quality particularly enables the physicians to monitor thermal damage and prevent possible injury. Furthermore, due to convenient position close to the surface of the skin and small size of the tumor nodules, the ablation probe can be retracted manually with adequate dexterity. If MST were to be applied to a larger organ such as the liver, effective ablation control could not be achieved. The reason is that in the liver, thermal energy needed to be deposited in a single ablation shot would be much higher than in the thyroid, since the targeted volumes would be larger. Hence, manual execution would not be feasible because in order to deliver that higher amount of energy, the probe would need to move at very slow retraction speeds (< 1 mm/s).

Lastly, the absence of relevant clinical literature on the topic of treatment strategy in terms of different tumor shapes limits at times the decision-making process to technical specifications rather than patient-specific solutions. Current technical solutions of commercial thermal ablation devices narrow the applicability of thermal ablation treatment for liver tumors. Namely, the maximal ablation volume across all available microwave ablation devices measures 56 mm in width and 78 mm in length. Therefore, complete tumor coverage can be effectively achieved only on small spherical tumors with a sufficient 5–10 mm ablative margin.

Potential limitations of the current study include the unbalanced sample size and the assumption of independent measurements in the configurable shapes group. While the unbalanced sample size was considered in the sample size calculation, the use of a robust variability estimator aimed to mitigate the impact of underlying correlations among the measurements. Furthermore, it is estimated that the measurement uncertainty of ablation width and length in the tissue-mimicking model was between 1–2 mm, as there was no singular cut-off color intensity between ablated (magenta color) vs. non-ablated areas (off-white color). Unless the ablated area was differentiated by the mapped temperature values (21), it was not possible to set a clear-cut segmentation boundary corresponding to the area that reached the necrotic temperatures above 60°C. Further, the tissue-mimicking model lacked realistic depiction of the human physiology, thus more suitable models such as in-vivo animal or perfused ex-vivo animal models would be required as the following step of the preclinical evaluation phase to test the safety and the efficacy of the proposed solution. Nonetheless, the model provided a valuable preclinical canvas to test the feasibility and the repeatability in a reproducible and rigid experimental manner. Lastly, potential clinical limitations in terms of safety and accuracy related to the breathing motion would be compensated with high-frequency jet ventilation (HFJV), a state-of-the-art technique advisably used with the aforementioned decision support system (26–28).

This study concludes that the repeatability of configurable ablation shapes was observed to be noninferior to the standard ablation shapes, within the predefined noninferiority margin of 1 mm. Moreover, the effectiveness of the proposed robotic platform and configuration model for thermal ablation underscores promising advancements towards enhanced treatment standardization and personalization tailored to individual clinical scenarios. This may broaden the applicability of thermal ablation to distinct tumor cases, which are deemed challenging by the standard ablation techniques. In this perspective, a fully integrated platform, with direct feedback from the ablation system to the robotic navigation system, could provide automated control of ablation power, time, and probe speed. In addition to the existing solutions for treatment planning, automatic segmentation, and direct quantitative assessment of ablation margins, the proposed treatment model could provide a complete solution for a standardized thermal ablation treatment even in complex tumor cases. As a next step, in-vivo validation is needed for evaluation of the clinical implications of the proposed treatment model.