A Comparison of Methodologies for Pulmonary Veins Segmentation in High Definition Voltage Maps of Patients with Atrial Fibrillation

This paper compares three methodologies used to segment images extracted from the atria of patients with atrial fibrillation (AF). We collected voltage maps of 120 patients' left atrium before being treated for AF with tissue ablation. The high-definition voltage maps (HDVM) were extracted with the Rythmia software system from Boston Scientific and subsequently analyzed offline in MATLAB. During the analysis, the atrium is segmented into three main structures: body, pulmonary veins (PVs), and mitral valve (MV). Method I is based on a manual elimination of the PV by the operator. Method II uses a semi-automatic method based on geometric considerations coupled with a threshold for voltage value. Method III, in addition, uses geometric features, i.e., the geometric curvature, to eliminate the maps' PVs. We quantitatively compare the two first methods' efficiency and ease of use. In particular, the values of two electrical biomarkers associated with the maps are computed before and after the PV's elimination process. This allows us to classify them according to their sensitivity with respect to the cleaning of the PVs. Finally, we discuss which methodology is more adequate to perform the PVs and MV segmentation in light of improving the precision of the resulting maps.