Automated detection of left atrium boundary in intracardiac echocardiography during atrial fibrillation ablation

Intracardiac echocardiography (ICE) is used during radiofrequency ablation (RFA) of atrial fibrillation to visualize and reconstruct left atrium (LA) anatomy and catheter placement, as well as to monitor potential complications, including the most harmful esophageal injuries. However, esophagus´ dynamic interface with LA posterior wall (LAPW) can be only roughly depicted by any available echo-based method. In clinical practice, at best, LAPW contour is manually traced (MT) on ICE images by the cardiologist, a procedure which is cumbersome and time-consuming. Accordingly, the present study aimed detecting LAPW position automatically by ICE during RFA. To identify LA borders a procedure based on level-set method and pixel clustering was applied. Algorithm was tested on 9 ICE acquisitions. Two series of LA contours were compared with MT LA boundaries by linear regression, Bland-Altman analysis, Dice coefficient and Hausdorff distance. Mean analysis time was 1 sec/frame. Detected contours were fairly concordant with MT ones. The developed technique permits automated and accurate detection of LAPW and may be the preliminary step towards additional analysis including dynamic quantification of esophagus real-time position´ changes and its distance from the LAPW.