Robust Detection of Mitral Papillary Muscle from 4D Transesophageal Echocardiography

摘要

Mitral valve (MV) diseases, one of the most common valvular diseases, often require surgical repair to reduce mitral regurgitation and improve cardiac pump function. These procedures however are very complex and require careful planning. In particular, chordae replacement or sub-valvular repair demands a precise assessment of the relative position of the papillary muscles with respect to the leaflets in the beating heart. This can be achieved only before opening the chest through imaging like computerized tomography or transesophageal echocardiography (TEE). Yet, quantitative analysis of the MV structure and dynamics, in particular the papillaries, is still tedious and prone to user variability. This manuscript presents a novel approach to automatically detect and track papillary muscle tips in 4D TEE. The proposed data-driven method combines the Marginal Space Learning method with Random Sample Consensus and Belief Propagation cope with varying image quality and signal drop-offs. Experiments on 30 randomly-selected volumes show that the accuracy of our algorithm falls within inter-rater variability (5.58mm out of 6.94mm for the anterior tip and 5.75mm out of 7.06mm for the posterior tip), while being extremely fast (under 3 seconds). The proposed method could therefore provide the surgeon with quantitative MV evaluation for optimal therapy planning.