Nonparametric mixture models for supervised image parcellation

Mert R. Sabuncu, B. T. Thomas Yeo, Koen Van Leemput, Bruce Fischl, Polina Golland

Research output: Chapter in Book/Report/Conference proceedingArticle in proceedingsResearchpeer-review

780 Downloads (Pure)

Abstract

We present a nonparametric, probabilistic mixture model for the supervised parcellation of images. The proposed model yields segmentation algorithms conceptually similar to the recently developed label fusion methods, which register a new image with each training image separately. Segmentation is achieved via the fusion of transferred manual labels. We show that in our framework various settings of a model parameter yield algorithms that use image intensity information differently in determining the weight of a training subject during fusion. One particular setting computes a single, global weight per training subject, whereas another setting uses locally varying weights when fusing the training data. The proposed nonparametric parcellation approach capitalizes on recently developed fast and robust pairwise image alignment tools. The use of multiple registrations allows the algorithm to be robust to occasional registration failures. We report experiments on 39 volumetric brain MRI scans with expert manual labels for the white matter, cerebral cortex, ventricles and subcortical structures. The results demonstrate that the proposed nonparametric segmentation framework yields significantly better segmentation than state-of-the-art algorithms.
Original languageEnglish
Title of host publicationProceedings of the Medical Image Computing and Computer Assisted Intervention (MICCAI) workshop on Probabilistic Models for Medical Image Analysis (PMMIA)
PublisherMedical Image Computing and Computer Assisted Intervention
Publication date2009
Pages301–313
Publication statusPublished - 2009
Externally publishedYes
EventMedical Image Computing and Computer Assisted Intervention (MICCAI) workshop on Probabilistic Models for Medical Image Analysis (PMMIA) - London, United Kingdom
Duration: 20 Sep 2009 → …

Conference

ConferenceMedical Image Computing and Computer Assisted Intervention (MICCAI) workshop on Probabilistic Models for Medical Image Analysis (PMMIA)
CountryUnited Kingdom
CityLondon
Period20/09/2009 → …

Bibliographical note

Creative Commons Attribution-Noncommercial-Share Alike 3.0

Fingerprint Dive into the research topics of 'Nonparametric mixture models for supervised image parcellation'. Together they form a unique fingerprint.

Cite this