Construction of a statistical atlas of the whole heart from a large 4D CT database

We present in this work an efficient and robust framework for the construction of a high-resolution and spatio-temporal atlas of the whole heart from a database of 138 CT 4D images, the largest sample to be used for cardiac statistical modeling to date. The data is drawn from a variety of pathologies, which benefits its generalization to new subjects and physiological studies. In the proposed technique, spatial and temporal normalization based on non-rigid image registration are used to synthesize a population mean image from all CT image. With the resulting transformation, a detailed 3D mesh representation of the atlas is warped to fit all images in each subject and phase. The obtained level of anatomical detail (a total of 13 cardiac structures) and the extendability of the atlas present an advantage over most existing cardiac models published previously.