Model-Based Segmentation of the Left Atrium in CT and MRI Scans

摘要

Ablation is a minimal invasive interventional method used in cardiac electrophysiology. It is one option for the treatment of patients suffering from paroxysmal or persistent atrial fibrillation through pulmonary vein isolation. During the intervention endocardial surface potentials from a tracked mapping catheter are recorded with respect to a static patient specific surface geometry. The purpose of the presented work is to compare two different automatic segmentation methods working on both CT and MRI volumes. Segmentation of the left atrium is challenging because the shape variability is high. The use of statistical shape models initialized by means of affine image registration was explored as first method. The second method was non-parametric and based on atlas registration and statistical region growing. Segmentation results were validated and compared using a leave-one-out cross validation on the volumes provided with segmentation results achieved manually by experts. The Dice's coefficient was used as error measure. The method based on statistical region growing performed better than statistical shape models. A Dice's coefficient of 0.87 was achieved on both imaging modalities.