Automated segmentation of midbrain structures using quantitative susceptibility mapping images
Benjamn Garzn 1 , Grgoria Kalpouzos 1 , and Rouslan Sitnikov 2
Aging Research Center, Karolinska Institute
and Stockholm University, Stockholm, Sweden,
Research Centre, Karolinska University Hospital,
We present a fully automated algorithm for segmentation
of the red nucleus, substantia nigra and subthalamic
nucleus from a pair of T1w and quantitative
susceptibility (QSM) images, aimed at providing
accurate, objective and reproducible segmentations. The
algorithm produces spatial probabilistic maps via a
multi-atlas label fusion scheme by combining global
(T1w) and local (QSM) non-linear registrations. These
probabilistic maps are employed as priors in a model
representing QSM intensities as a Gaussian mixture.
Manual segmentations were obtained for 16 subjects and
used to train and validate the model. Cross-validated
Dice scores ranged between 0.66 (subthalamic nucleus)
and 0.85 (red nucleus).
This abstract and the presentation materials are available to members only;
a login is required.