Previous Page  19 / 26 Next Page
Show Menu
Previous Page 19 / 26 Next Page
Page Background

The paper they are presenting at MICCAI on Thursday (see their poster

PS5-21 from 10:30-12:00) is about a co-registration and segmentation

framework that combines ideas from these fields towards producing

automatic annotations of sub-cortical structures in brain MR images. These

annotations are of great importance for diagnosis and characterization of

different neurodegenerative and neuropsychiatric disorders, including

schizophrenia, Alzheimer's disease, attention deficit, and sub-types of


What I found most interesting about their method in comparison with

previous multi-atlas segmentation approaches, is that instead of using

expert manual annotations (which are difficult to obtain and highly time-

consuming), they showed that it is possible to use semantic priors learned

with any machine learning technique to guide both, groupwise registration

and segmentation processes. For that, they contacted Stavros Tsogkas, a

deep learning and computer vision researcher currently working at

University of Toronto, to design a convolutional neural network

architecture that was finally used to produce the semantic priors.

Their work was not only a nice example of science collaboration, but it also

produced really promising results which may lead to future better

understanding of structural alterations related to different brain disorders.


MICCAI Daily: Thursday

Mahsa had experience on brain image

segmentation techniques, and during the

last year she was particularly interested in

multi-atlas segmentation