Everyone interested is welcome to attend!
No prior booking necessary.
Royal Statistical Society
Statistical Computing Section Meeting
Wednesday 24 May, 2.00 - 5.00pm (tea at 3.20pm)at
The Royal Statistical Society, 12 Errol St, London EC1
(nearest UG stations, Barbican, Moorgate & Old Street)
Software applications for image analysis
----------------------------------------------
Speakers
CHRIS GLASBEY (BioSS, Edinburgh)
GEIR TORHEIM (Norwegian University of
Science and Technology)
Programme
---------
2.00pm AGM of the Statistical Computing Session followed by:
Chris GLASBEY "Statistical analysis of sheep
CT images"
(talk, demonstration & discussion)
3.20pm Tea
3.45pm Geir TORHEIM "Analysis of Contrast-Enhanced Dynamic
MR Lung Images"
(talk, demonstration & discussion)
Synopses
CHRIS GLASBEY (BioSS, Edinburgh)
Statistical analysis of sheep CT images
Synopsis:This talk will describe joint work with Caroline Robinson of the
University of Edinburgh.
X-ray CT (computed tomography) is a non-invasive imaging technique.
Originally developed for medical diagnosis, its potential has recently
been recognised in the totally different area of sheep breeding. This
has generated challenging new statistical and image analysis problems.
The protocol established in the SAC-BioSS CT Unit in Edinburgh, is to
obtain a conventional X-ray image and three anatomically-located CT
cross-sectional images for each sheep. From these data, we estimate an
animal's fat and muscle tissue volumes, using three stages:
1. We locate boundaries to isolate the sheep's internal organs in the
CT images, by formulating Bayesian models for which the MAP estimators
are obtained using a shortest-path algorithm;
2. We use efficient, robust methods to estimate fat and muscle
proportions in each CT image, taking account of mixed pixel effects;
3. We infer whole-body tissue proportions, by pooling cross-sectional
information in the CT images with projective information in the
conventional X-ray images.
Methods have been implemented in an interactive PC package (STAR: Sheep
Tomogram Analysis Routines) written in VisualBasic.
For further details, see papers at http://www.bioss.sari.ac.uk/~chris
GEIR TORHEIM (Norwegian University of Science & Technology)
Analysis of Contrast-Enhanced Dynamic MR Lung Images
Synopsis: This is joint work with Giovanni
Sebastiani(Istituto per le Applicazioni del Calcolo,
C.N.R., Rome) and Fred Godtliebsen (University of Troms,
Norway)
Dynamic contrast enhanced Magnetic Resonance Imaging (MRI)
has been shown to give clinically useful information in
many situations, like perfusion imaging of the breast,
lung, kidneys, and the heart. Some organs, most notably
the lungs, are extremely difficult to image and process due
to low signal-to-noise ratio, and movement during the data
acquisition.
We propose a method to correct for lung motion based on a physical
model. This involves edge detection,followed by a linear
deformation. Simulated annealing was applied in the edge
detection step.
Two noise reduction methods are also presented: one based on Bayesian
statistics, and a novel method which is similar to the
Nadaraya-Watson estimator. The Bayesian noise reduction
method employs a Markov Chain Monte Carlo algorithm.
Parameter estimation is performed automatically by an
iterative approach, making the method almost completely
automatic and user-independent. The new noise filter works
by smoothing data in a neighborhood only if the data are
"similar" through time, thus preserving edges much better
than the Nadaraya-Watson estimator.
All methods were implemented into Dynalize(TM), a software package for
performing analysis of dynamic image series.
----------------------
Suzanne Evans
[log in to unmask]
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|