(A-20) Rapid Electroporation-Dependent Tissue Property Prediction in Canine Lung Tumors

Introduction::
IRE is used clinically to treat tumors situated near sensitive structures, where surgical resection and thermal ablation methods are contraindicated; this includes locally advanced and unresectable liver, kidney, prostate, and pancreatic malignancies. The efficiency of electroporation treatments is contingent on the application of a critical electric field over the targeted tissue volume[1].






The distribution of the electric field and thermal heating depends on the tissue-specific electrical properties, which both differ between patients in healthy and malignant tissues and change in an electric field-dependent manner from the electroporation process itself[2]. Therefore, tissue property estimations are paramount for accurate treatment planning with electroporation therapies. Ex vivo methods to find electrical tissue properties often misrepresent the targeted tissue, especially when translating results to heterogenous tumors. Therefore, in situ method for tissue characterization are greatly needed to find patient-specific electric tissue properties. Here, we developed an accurate and robust deep neural network (DNN), trained on realistic finite element model simulations, to directly predict tissue properties from in situ voltage ramps (VR).




Materials and Methods::
A voltage ramp (VR) is an in situ method that applies a series of increasing electric potentials across the treatment electrodes ranging from well below to above the electroporation threshold (25 - 2250 V/cm). Both voltage and current are measured, then finite element modeling (FEM) is used to calculate tissue impedance effects from different electric field strengths (Fig 1A). The conductivity of tissue as a function of the applied electric field (Fig 2B).






To generate the dataset which the DNN was trained on, finite element modeling was employed in COMSOL^TM Multiphysics 6.1 (COMSOL, Stockholm, Sweden) The model incorporated electroporation and thermal effects to simulate applied current during a voltage ramp. A parametric sweep was performed on the geometry, applied potentials, and tissue properties (Fig 1C).






We chose to use a regression DNN (2-hidden layers), trained with supervised learning. The input to the model would be the voltage and current data read from a VR and the output would be the dynamic tissue properties (Fig 1D).




Results, Conclusions, and Discussions:: The trained DNN was validated with simulated test data that it was not explicitly trained on. We were able to predict the bulk conductivity (𝜎0), change in conductivity (𝜎a), and intercept fitting parameter (E) with a low percent error. The quartiles for the percent error were < 1% with R2 >0.94. The slope fitting parameter (A), showed no correlation. However, there is not a strong relationship between measured current and A; varying A for the range of electroporation applications (0.002 cm/V - 0.01 cm/V), only led to a 10.4% change in applied current.

Our model was able to correctly predict the dynamic conductivity curve in a previously characterized ex vivo porcine liver model (R2 > 0.93, p < 0.0001) within 100 seconds from probe insertion, showing great utility for a clinical application. We then used the validated DNN to predict the in situ conductivity curve for 5 canine patient lung tumors. Since the explicit tumor property values are unknown, we simulated the applied current using the predicted conductivity and compared it with the measured current during the electroporation treatment. We found agreement with (R2 > 0.9930, p < 0.0001) and max error of 25%.

We believe this platform can be incorporated prior to treatment to quickly ascertain electroporation-dependent electrical tissue properties required for treatment planning models or real-time treatment prediction algorithms. Further, this method can be used over traditional ex vivo methods for in situ or patient-specific tissue characterization with clinically relevant geometries.


Acknowledgements (Optional): : This research was funded in part by the NIH RO1 (RO1CA24047) and the Veterinary Memorial Fund. We would like to acknowledge the Surgical Oncology clinical team at the Virginia-Maryland College of Veterinary Medicine for their assistance in gathering the voltage and current data during their treatment of canine lung tumor patients.


References (Optional): : [1] RV Davalos et al., “Tumor Ablation with Irreversible Electroporation,” Ann Biomed Eng, vol. 33, no. 2, 2005.
[2] N Beitel-White et al. "Multi-tissue Analysis on the Impact of Electroporation on Electrical and Thermal Properties," IEEE Trans Bio Med, vol 68, no. 3, 2021.