A new computational framework for anatomically consistent 3D statistical shape analysis with clinical imaging applications

摘要

A computational approach is presented for the quantitative comparison and statistical decomposition of sets of 3D closed genus-0 surfaces obtained from untagged medical images. The core of the analysis procedure consists in a two-step surface parameterisation technique followed by the proper orthogonal decomposition method. The capabilities of the proposed approach are demonstrated through analysis and sensitivity tests considering a human right ventricle endocardial surface captured at several time frames over the cardiac cycle, as well as decomposition and classification tests considering several sets of artificially generated 3D surfaces.