Non-invasive prediction of optimal targets for efficient radio-frequency ablation is a major challenge in the treatment of ventricular tachycardia. Most of the related modelling work relies on magnetic resonance imaging of the heart for patient-specific personalized electrophysiology simulations. In this study, we used high-resolution computed tomography images to personalize an Eikonal model of cardiac electrophysiology in seven patients, addressed to us for catheter ablation in the context of post-infarction arrhythmia. We took advantage of the detailed geometry offered by such images, which are also more easily available in clinical practice, to estimate a conduction speed parameter based on myocardial wall thickness. We used this model to simulate a propagation directly on voxel data, in similar conditions to the ones invasively observed during the ablation procedure. We then compared the results of our simulations to dense activation maps that recorded ventricular tachycardias during the procedures. We showed as a proof of concept that realistic re-entrant pathways responsible for ventricular tachycardia can be reproduced using our framework, directly from imaging data.