Extraction of cardiac structures through the incorporation of a priori knowledge in a multiscale approach

The task of simultaneous segmentation of complex anatomical structures such as the cardiac organs is not trivial. The authors have divided this task in two phases. (A) extraction of the desired structures, that includes a rough segmentation, (B) a post-processing phase for segmentation refinement. In this work the authors are addressing the extraction phase using an image representation based on a scale space approach. The Scale Space Primal Sketch representation is based on image structures called blobs, that represent regions in which image counts are higher or lower than their surrounding background. The linking of these structures across scale space allows the analysis of the characteristics and behavior of these blobs. Stable blobs are supposed to represent important features of the image and significance measurement rates based on it allows the classification of these structures. This methodology was implemented and applied to MRI thoracic images. It allowed the extraction of the cardiac structures, but also extracted other undesired features. In order to overcome this problem, the authors refined the criteria for the extraction of significant blobs from the Scale Space Primal Sketch by including a priori information on the significance measurement rate