Hierarchical cardiac MRI segmentation using shaped region-interface descriptors

Presented is a generalised hierarchical image segmentation framework for medical image segmentation. The framework represents an anatomy region based image description model comprising of a knowledge base of interface conditions between anatomy regions. This paper describes the implementation and application of this framework for Cardiac MRI dataset segmentation. We describe the construction of the knowledge base from manually delineated contours and the possible automation of this learning process. Demonstration software provides accurate segmentation of a CMRI data consisting of 12 slices and 2 phases in less than 5 min. The user is provided with quantification of blood volume and myocardial mass for each phase thus providing a blood volume curve.