Abstract: The aim is to provide an online estimation of the electric field applied during an electroporation ablation, following the numerical workflow in (1). This minimally invasive and non-thermal ablation technique can be used on deep-seated tumors, where traditional techniques may affect vital structures. However, it requires thorough planning and evaluation, due to its inherent complexity.
To this end, we propose a novel coarse-to-fine algorithm for the extraction of needles delivering the electric field, from a single Cone Beam Computed Tomography. It is a crucial step in the computation of the electric field to evaluate the procedure (2). A coarse segmentation is obtained by a modified U-Net (3), trained with a patch optimization strategy and a well-suited loss function. The analytical representation is computed by a Hough transform (4), completed with a voting procedure. Finally, the electric field is obtained with a standard linear static model.
The results are evaluated on 8 of 16 patients: for the coarse segmentation, we compare to the groundtruth using the Dice coefficient and for the analytical representation, the distance between the estimated and real coordinates is computed. Under two minutes on a commodity hardware, the needles are extracted with a more precise and stable algorithm than the previous coarse segmentation with thresholding.
Keywords: Deep Neural Network, Fine-object Segmentation, CBCT, Electric field distribution

(1) O. Gallinato, B. Denis de Senneville, O. Seror, C. Poignard (2019). Numerical workflow of irreversible electroporation for deep-seated tumor. Physics in Medicine and Biology.
(2) O. Gallinato, B. Denis de Senneville, O. Seror, C. Poignard (2020). Numerical Modelling Challenges For Clinical Electroporation Ablation Technique of Liver Tumors. Math. Model. Nat. Phenom.
(3) F. Isensee, et al. (2018). nnU-Net: Self-adapting Framework for U-Net-Based Medical Image Segmentation. arXiv:1809.10486
(4) L. Shapiro, G. Stockman (2001). Computer Vision. Prentice-Hall, Inc.