Robust Morphological Averages in Three Dimensions for Anatomical Atlas Construction

摘要

We present original methods for obtaining robust, anatomical shape-based averages of features of the human head anatomy from a normal population. Our goals are computerized atlas construction with representative anatomical features and morphopometry for specific populations. A method for true-morphological averaging is proposed, consisting of a suitable blend of shape-related information for N objects to obtain a progressive average. It is made robust by penalizing, in a morphological sense, the contributions of features less similar to the current average. Morphological error and similarity, as well as penalization, are based on the same paradigm as the morphological averaging.