Contrast-Based Registration Of Left Atria To Fluoroscopic Image Sequences By Temporal Markow Filtering And Motion Regularization

Atrial fibrillation is a common heart arrhythmia which can be treated minimally invasively using catheters. Navigation during the procedure can be performed under fluoroscopic guidance. To overcome the low soft tissue contrast of fluoroscopic images, they can be fused with a pre-interventionally acquired 3-D segmentation of the patient's left atrium (LA). As catheter positioning is crucial to destroy arrhythmogenic tissue of the LA, better navigation by fusing the 3-D LA shape with fluoroscopic images is important. Furthermore, additional graphical information can be visualized together with the LA once the LA segmentation has been registered to the fluoroscopic images. Registration can be performed based on contrast agent which is injected into the LA and then captured as part of an X-ray sequence. Recent publications have shown that an intensity-based registration using the contrasted area leads to good results. However, the dependency between the frames of an X-ray image sequence has not been exploited yet. Since contrast agent is injected over a certain period of time and captured over many frames, we propose a time-dependent registration. To this end, we introduce motion regularization to ensure small motion of the overlay between successive frames. We propose two methods for motion regularization. One method is based on a Markov chain model of the heart motion, the other approach introduces a motion regularizer into the similarity measure used for registration. The Markov chain approach reduces the registration error significantly from 7.9 +/- 6.3mm to 5.7 +/- 4.6 mm. The motion regularizer also reduces the registration error significantly, but less than the Markov approach.