REGISTRATION-BASED PROPAGATION FORWHOLE HEART SEGMENTATION FROM COMPOUNDED 3D ECHOCARDIOGRAPHY

Whole heart segmentation of 3D ultrasound (US), also re-ferred to as echocardiography or simply echo, is useful in car-diac functional analysis to achieve quantitative diagnostic in-formation of the heart. However, characteristics of US imag-ing such as limited field-of-view, artifacts and inconsistent in-tensity distribution makes automated approaches a challenge. In this paper, we present a framework for automatic whole heart segmentation from 3D echo. This work is motivated by the new technology of compounding 3D echo from 2D matrix array transducers. We propose to use the registration-based segmentation framework and adopt a new similarity measure combining local phase, intensity information and local geom-etry for registration. The experimental results demonstrated the proposed method had achieved an accuracy of 6.4 % vol-ume difference against the gold standard for the left ventricle segmentation and an average accuracy of 14 % for segmenta-tion of all four chambers and myocardium. Index Terms—Whole heart segmentation, ultrasound, echocardiography, image registration 1