| 摘要 |
<p>The invention concerns shape discrimination analysis, especially for two class discrimination. For instance, statistical shape discrimination analysis in medical image analysis of anatomical shapes, such as organs. The invention includes determining whether there is a significant difference between shapes of two classes of objects. This includes scalarizing (44) the multivariate data by projecting the multivariate data in the "informative" direction, determining (46) a scalar test statistic of different partitions of the scalarized multivariate data. The invention also includes identifying the difference between shapes of two classes of objects by determining a discriminative direction of the two classes. This includes using a kernel function, mapping (72) multivariate data describing the shape in an input space into a space of a higher dimension, (determining 74) an "informative" direction that best discriminates the two classes, and moving (76) the multivariate data in the "informative" direction, and finally using preimage techniques to project (78) the moved multivariate data from higher dimension space to the input space. Aspects of the invention include methods, software and a computer system.</p> |
