A Modular Framework for Shape- and Image-prior-based Anatomical Structure Contour Extraction

Abstract

The automatic extraction of anatomical structure contours from medical images is a challenging task in the presence of missing or unrelated parts, occlusions caused by other structures, and image noise. Employing prior information about the anatomical structures has been one of the most popular ways of addressing these challenges. This paper presents a novel framework that incorporates both shape and image priors into the contour extraction process with deformable contours. The framework handles the deformable contour evolution and the prior information integration separately by stopping the evolution of the deformable model and regularly re-initializing the expert-delineated contour with the most similar image and shape properties. The method can be used with any deformable model without complicating the deformable model functional. An explicit training phase is not required for the construction of the prior model. Moreover, it can be applied to any medical shape contour extraction task with simple modifications. The system is tested on echocardiographic images and cardiac magnetic resonance imaging (MRI) slices for left ventricle border extraction under the level set framework. The visual and numerical results show the effectiveness of the proposed method.