The objective of this paper is to provide a stepforward towards the per procedural visualisation of the electric field distribution during a clinical irreversible electroporation (IRE) procedure. To this end, an automated workflow is needed to compute the electric field distribution on a single Cone Beam Computed Tomography (CBCT) scan.The aim of the current paper is to propose a deep learning strategy for the automatic segmentation of the needles. In particular, a novel coarse-to-fine approach is proposed to extract relevant needle information from the CBCT scan, despite inherent artefacts generated during capture. The obtained needle information is subsequently fed into a standard static linear model for the electric field computation. Since the set-up is performed in the medical image framework, the electric field distribution and the region of interest are visible to provide to the radiologist a visual evaluation of the treatment.The segmentation results are evaluated on 8 of the 16 patients of the dataset using the Dice coefficient to compare the predicted segmentation with the ground truth. The proposed segmentation method is fast (around 2 minutes are needed with a commodity hardware), allowing its use in a clinical setting.