Validation of a Web-Based Planning Tool for Percutaneous Cryoablation of Renal Tumors

Thermal ablative therapies for small renal masses (SRMs; < 4 cm) are an alternative treatment for nephron-sparing surgery (NSS) [1]. With fewer complications reported compared to the laparoscopic approach, the image-guided percutaneous approach for this treatment is established [2]. Despite the availability of larger studies with recurrence-free survival exceeding 85% after longer follow-up, long-term oncological outcomes for percutaneous cryoablation are under debate [3‐5]. Two recent meta-analyses suggest higher local recurrence rates for cryoablation compared to partial nephrectomy; whereas, more recent studies show that cryoablation challenges partial nephrectomy for local control of cT1a renal tumors [6‐8].

Currently, there is limited software available for interventional radiologists to plan the procedure and to predict procedural outcomes [9]. Obtaining full tumor coverage with enough margin is imperative and depends on the type, configuration, and the number of needles used. The physician makes these decisions based on the predicted ice ball formation from a single needle provided by the manufacturer and experience with the equipment used [10, 11]. Without proper support of dedicated software, physiological components such as organ tissue characteristics and blood flow are difficult to take into account [12, 13]. Once the ablation has started and the ice ball is formed, the possibility of needle replacement is obliterated. This emphasizes the importance of pre-procedural planning to ensure radical treatment.

In April 2013, the Go-SMART project started with the aim to build a generic open-source software simulation environment to be used for planning of image-guided percutaneous cancer treatment modalities [14]. The workflows of the environment are designed to enable minimal invasive procedure planning in advance by the interventionalist using a pre-interventional diagnostic scan only. A part of the project was to develop and incorporate a workflow for planning percutaneous cryoablation of renal tumors in the web environment. This included image segmentation and registration tools as well as a treatment simulation model with the ability for validation of this model.

The aim of our study was to validate the simulation model for virtual planning of percutaneous cryoablation of renal tumors. The model was designed based on two variables: an equitation modelling temperature development during ablation and a prediction of tissue response based on physiological properties.

This study was IRB approved. The development of the web-based environment (freely available through (https://​smart-mict.​eu/​)) is extensively described [14]. In short, a simulation tool using a multi-scale physiological model was developed to predict the result of the treatments in terms of ablation zone size and shape. To calculate the first-order effect of the cryoablation, a modified Pennes bioheat equation with added perfusion term was used. This model is based on the density, specific heat capacity, heat conductivity, and temperature of the perfused tissue (renal tumor), heat flux due to the ablation instrument, and the norming effect of tissue perfusion. The norming effect on itself was based on the local fraction of cells considered dead, the perfusion coefficient, material property of renal tumor tissue, standard body temperature, current local temperature, and the density and specific heat capacity of blood [15]. Within the modified Pennes bioheat equation, physical properties change (liquid to solid) due to the expanding ice ball was taken into account. This resulted in the use of adjusted heat capacity and thermal conductivity [14]. For the cell death model under hypothermia, a simple empirical isotherm was used. Extensive discussion of the theory behind the mathematical model used for simulating is published elsewhere [13]. Validation tools were incorporated to verify the predicted treatment result based on true post-operative control images of treated patients. The web-based environment workflow is outlined in Fig. 1 and comprehensively described below.

In 18 patients, 19 MR-guided percutaneous cryoablations were performed. One patient was treated twice for the recurrent disease at different anatomical locations with an interval of 21 months. A median of 3 (range, 2–4) needles per procedure was used. All tumor characteristics and treatment specifications are listed in Table 1.

Mean volume of S (real ablation zone) and Σ (simulated ablation zone) was 14.8 cm³ (SD ± 9.9) and 26.6 cm³ (SD ± 15.4), respectively. The mean diameter of the sphere which circumscribes S was 4.6 cm (SD ± 1.1) and 5.6 cm (SD ± 1.3) for Σ. The AAE had a mean of 3.8 mm (SD ± 2.4). The mean value for DSC was 0.62 (SD ± 0.17). In 9 out of 19 cases (47%), DSC was scored as good or excellent (value ≥ 0.7) (Table 2; Figs. 7, 8).

Mean TO and PPV were 0.88 (SD ± 0.10) and 0.53 (SD ± 0.24), respectively. In 17 cases (89%), TO was scored as good or excellent. For PPV, only 4 cases (21%) were scored as such, 13 cases (68%) were scored poor or inadequate (value < 0.6) (Table 2).

The relatively low values for PPV combined with high TO values indicate that the simulation is overestimating the real ablation zone in the majority of cases (Fig. 9).

In this study, a simulation model for renal cryoablation treatment planning was validated. Results from validation of the first 19 cases showed a tendency of overestimation of the real ablation zone by the simulation, leading to undertreatment when used in the clinical setting.

Based on this result, the simulation tool can be further refined. The inherent limitation to simulation models used for cryoablation is the development based on experimentally derived parameters, especially regarding bioheat transfer, which is generally recorded in an ex vivo setting [14]. Also, the additive effect of using multiple probes during the procedure is challenging to take into account based on the available literature. Because cryoablation modelling heavily depends on the application of accurate parameters, especially thermophysical properties, the use of experimentally derived parameters results in significant uncertainties in these models leading to errors when predicting the cryoablation treatment effect [13]. These errors become apparent during validation of the applied model in a setting as described in this study. Adjustment of the used parameters in the model is required for further refinement of the simulation tool. Due to a lack of available reliable tested parameters, subsequent testing on clinical cases as described in this study should reveal the sensitivity of the simulation model to the adjusted parameters. Optimizing the used physiological parameters should eventually lead to improved simulation results.

The complexity of predicting cryoablation effect underlines the importance of a simulation tool. The ablation effect is dependent on numerous physiological (e.g. blood perfusion, metabolic heat, thermophysical properties) and treatment protocol (e.g. duration freeze–thaw cycle) properties. The extent to which these properties affect the ablation effect is impossible to take into account purely by subjective evaluation of the treating interventionalist. Using a simulation tool taking these properties into account can facilitate treatment planning to ensure oncological safe ablation margins omitting unnecessary ablation of the healthy renal parenchyma.

Although the computational prediction of ice ball formation is complex, and mathematical models for prediction are continuously being improved, several planning tools for clinical use have been developed [9]. Boas et al. developed and validated a planning tool for multiple-probe cryoablation [20]. In this study, simulations with different numbers and configuration of needles were performed using the Pennes bioheat equation. The simulated ice-ball sizes, measured along 3 perpendicular axes, were validated using 26 gel experiments and 42 clinical kidney and liver cases. The surface deviation between the simulated and real ice-ball was used for validation, and showed an absolute average error of 4 mm in the clinical cases. This is comparable to our study (3.8 mm SD ± 2.8). In the treatment planning workflow presented by Boas et al., the desired ice-ball measurements are provided by the interventionalist after which the environment provides a number and configuration of needles resulting in an ice ball with the closest match. Compared to our planning tool, this model lacks a quantification of volumetric coverage between the predicted ablation zone and the tumor. Another limitation is that despite the implementation of a wide variety of needle configurations, the provided advice for needle configuration by this model may clinically be unfeasible. In the environment presented in this study, more flexibility is provided by enabling simulations with up to nine needles with an infinite number of configurations that can be evaluated for adequate tumor coverage.

Torricelli et al. described the development of an algorithm for planning the number and configuration of needles for cryoablation based on the spherical-shaped ice ball formation from one needle as provided by the manufacturer [21]. A stepwise approach was used starting by computing the initial number of ice balls necessary, followed by simulating the configuration of ice balls and optimizing this (i.e. changing needle position) to cover the complete tumor. Finally, an extra ice ball can be added to reach full tumor coverage. Limitation of this approach is the assumption that multiple probe usage only gives an additive effect on ice ball volume and the ablated ablation zone. However, the use of multiple needles has a synergistic effect, resulting in a larger effect than only additive [22, 23]. Moreover, the predicted ice ball size as provided by the manufacturer (tested in gels) tends to overestimate the ablation zone after treatment in vivo [24]. Also, this model has not been validated.

Treatment planning using a simulation model can have multiple purposes. The primary goal is to facilitate pre-procedural planning of image-guided percutaneous ablative therapies. This is done by enabling the interventionalist to virtual test the optimal needle type, number of needles used, and needle configuration in order to obtain complete tumor coverage. Second, the model can be used for training purposes by untrained interventionalists to become acquainted with the treatment effects. Third, it would be beneficial to have an environment enabling the comparison of several ablative treatment modalities, such as microwave ablation (MW), radiofrequency ablation (RFA), and irreversible electroporation (IRE) to choose the optimal treatment modality for individual treatments. Within the Go-Smart project, of which the development of the cryoablation planning tool was a part, also simulation tools for microwave (MWA)- and radiofrequency ablation (RFA) and irreversible electroporation (IRE) for several organs were developed simultaneously. However, preliminary testing during the development phase showed the most favorable results for cryoablation. Most challenging was the development of MWA due to the complexity of electromagnetic modelling, and IRE due to the challenging correct determination of an ablation zone [14]. Improvement of the simulation tools for the other treatment modalities and subsequent validation is awaited.

Artificial intelligence algorithms can directly test model sensitivity to adjustment of thermophysical parameters to evaluate what parameter adjustment optimizes the simulation results can be of great benefit during model validation. Also, algorithms for automated image registration and segmentation would be helpful to increase the speed and accuracy of the model, but moreover would facilitate rapid treatment effect evaluation in clinical use once the ablation is performed.

Some limitations were present that influenced our simulation results. An important limitation is heterogeneity between histological tumor types, e.g. perfusion characteristics, which were not adjusted for in the simulation model. Also the simulation interface suggests a fully rigid needle position. A wide variety of factors, such as breathing or slight needle repositioning between ablation cycles, can lead to needle movements. In a retrospective analysis, it is impossible to account for these movements. Furthermore, deviations in needle alignment between needle position and registered needle position may occur in a millimeter fashion due to artifacts on imaging, errors in needle identification, and registration inaccuracy. Although validation results were not optimal so far, sources of error were investigated and identified. Adjustment of the simulation model for these errors will benefit future simulation results.

In this study, we validated a simulation model used for renal tumor cryoablation treatment planning within a web-based environment. Based on the first validation results of the simulation model, we conclude that refinement of the simulation model is needed to reduce overestimation of the ablation effect. Model parameters adjustment to improve simulation performance and evaluation of the adjustment effect are possible within the environment. Until more accurate results are obtained, the simulation model is not suitable for use in clinical practice.