Marginal Space Learning for Efficient Detection of 2D/3D Anatomical Structures in Medical Images

摘要: Recently, marginal space learning (MSL) was proposed as a generic approach for automatic detection of 3D anatomical structures in many medical imaging modalities [1]. To accurately localize object, we need to estimate nine pose parameters (three position, three orientation, and anisotropic scaling). Instead exhaustively searching the original nine-dimen-sional parameter space, only low-dimensional spaces are searched MSL improve speed. In this paper, apply 2D object perform thorough comparison between alternative full (FSL) approach. Experiments on left ventricle MRI images show outperforms FSL both speed accuracy. addition, propose two novel techniques, constrained nonrigid MSL, further efficiency real applications, strong correlation may exist among same spaces. For example, large have scaling values along all directions. Constrained exploits speed-up. The estimates rigid transformation an image, therefore cannot under deformation. directly deformation localization experiments liver 226 abdominal CT volumes demonstrate effectiveness methods. Our system takes less than second detect volume.