Prof. Frederik Maes KU Leuven ESAT/PSI

advertisement
ImagO Colloquium
Medical Imaging
Iconic shape models: joint registration,
segmentation and atlas construction
Prof. Frederik Maes
KU Leuven ESAT/PSI - Medical Imaging Research Center
In this talk I will give an overview of 20 years of research at the Medical Image
Computing group of KU Leuven dealing with iconic models, i.e. image-based
representations of photometry, shape and context, such as atlases. Such models
are constructed from a set of pre-annotated images and fitted to new images by
non-rigid image registration, driven by a suitable similarity measure and
regularisation mechanism. Maximisation of mutual information of corresponding
voxel intensities proved to be a breakthrough in the field, but in the context of
image segmentation, use of more specific object labels in the similarity measure
is to be preferred. This naturally leads to a joint segmentation-based
registration and registration-based segmentation approach. When extending this
idea from the analysis of a single image to multiple images, the construction of a
group-specific optimal atlas can be integrated in the solution. By considering
that different images in the group may belong to different subgroups (e.g.
healthy versus diseased), stratified atlases per subgroup can be constructed
that can be interpreted as summaries of morphological differences between
subjects at the group level (i.e. imaging biomakers). Our current work focuses on
extending this scheme towards longitudinal analysis and construction of multitemporal atlases. The various concepts in this talk are mathematically
consistently formulated and solved within a probabilistic framework using the
EM algorithm and illustrated for brain MR images.
Tuesday 20 January 2015, 13.30 – 14.30 hrs
Auditorium, new Q-building
(AZU, Heidelberglaan 100, Utrecht)
Medical Imaging: ImagO, PhD Programme of the Graduate School of Life Sciences
University Medical Center Utrecht – Q02.4.45, Heidelberglaan 100, 3584 CX Utrecht, The Netherlands
Tel: +31 88 755 7772, E-mail: info@imago.uu.nl, URL: http://www.imago.uu.nl
Download