New path-based operators for the detection of thin tubular objects in 3D

The Faculty of Informatics is pleased to announce a seminar given by Hugues Talbot

DATE: Thursday, March 27th, 2014
PLACE: USI Lugano Campus, room SI-004, Informatics building (Via G. Buffi 13)
TIME: 14.00

ABSTRACT:
The filtering and segmentation of thin tubular objects in 3D has many applications, for instance in medical imaging (blood vessel segmentation in CT or MRI) or materials science (analysis of fibrous materials).
However, these objects are sensitive to noise and difficult to identify, and most classical restoration or segmentation techniques are not well adapted to thin objects because of their geometry. As a simple example, classical median filtering or convolution is not possible because no single window can fit everywhere in the object of interest. Also most segmentation method failed because these object may not have an identifiable interior or indeed not even visible contours.
In this talk, we will present a multi-scale approach based on recently developed path operators. These non-local operators strike a useful, practical balance between spatially-limited operators, which may fail to identify the isotropic nature of thin tubular objects, and global operators that may impose too much regularity and may also be computationally intensive. Path based operators can also be useful to model the connectivity of tubular structure, which is a crucial feature of various anatomical networks. They can also be robust to noise and can deal with a high degree of tortuosity.
We illustrate these operators on MRI images of brain vessels, on CT images of the coronaries, and we validate them on synthetic data as well as on a database of images from a recent heart vessel segmentation MICCAI challenge.

BIO:
Associate professor at ESIEE since November 2004, in the informatics department A2SI (Algorithmique et Architecture des systems informatiques)

HOST: Prof. Jürgen Schmidhuber