A framework combining window width-level adjustment and Gaussian filter-based multi-resolution for automatic whole heart segmentation

Heart diseases are prevalent among the general population. These diseases can be diagnosed in their early stages through a quantitative evaluation of cardiac functions. In a typical procedure, heart segmentation is initially performed. Quantitative information is then obtained from a 3D reconstructed image of the heart. However, manual segmentation is time-consuming and prone to inter- and intra-observer variations. As such, automatic methods must be developed to assess cardiac functions quantitatively. In this study, an automatic algorithm for whole heart segmentation was established through window width-level adjustment and Gaussian filter-based multi-resolution methods. The proposed algorithm preprocesses the image by adjusting the window width and the centre to acquire cardiac images with clear anatomical structures. The cardiac image is then decomposed into several resolution layers by using a Gaussian filter to eliminate discontinuity associated with traditional pyramid down-sampling and decomposition. A registration-based segmentation algorithm is applied to the cardiac image. The proposed segmentation algorithm was validated with a clinical dataset of 14 cardiac dual-source computed tomography images. Results show that the proposed methods improve the registration accuracy of the epicardium and the endocardium. The volume of the manual segmentation standard is not significantly different from that of the proposed segmentation and the accuracy of the method reaches almost 1 mm in most areas. Thus, the proposed method can be used to perform a high-precision segmentation of the whole heart. (C) 2016 Elsevier B.V. All rights reserved.