A method and system for detecting a virtual electrode (VE) on a coronary sinus (CS) catheter in a fluoroscopic image sequence is disclosed. User inputs indicating locations of CS catheter electrodes and a location of a VE are received. A catheter electrode model and a VE part model is initialized in a first frame of the fluoroscopic image sequence. The VE is tracked by detecting electrode position candidates and catheter body point candidates in the subsequent frames of the fluoroscopic image sequence using respective trained detectors, tracking the catheter electrode model in the subsequent frames based on the detected electrode position candidates, generating VE part hypotheses in the subsequent frames based on detection of the most proximal electrode (MPE) in each subsequent frame, calculating a probability score for each of the VE part hypotheses, and selecting an VE part hypothesis with the highest probability score.
