We present an algorithm for passing from dense noisy neuroanatomical segmentations, directly to a complexity-reduced representation with respect to a deformed smooth template surface, bypassing the need for triangulation of any target data. We demonstrate the utility of this algorithm toward improving reproducibility of hippocampal definitions, using a dataset containing 4 MR images per subject, two within the same visit on each of two dates, with dense segmentations provided by unedited longitudinal Freesurfer analysis. We quantify reproducibility of intra-visit and inter-visit variability through L2 distances and Hausdorff distances between pairs of segmentations, and show that our method results in a statistically significant improvement by a factor of 1.63 to more than 3-fold.
|Number of pages||8|
|Journal||Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)|
|Publication status||Published - 2014|
- Complexity reduction
ASJC Scopus subject areas
- Computer Science(all)
- Theoretical Computer Science