Curvelets: A patient specific prior for multi-modal image registration
Speaker: Moti Freiman , Hebrew University of Jerusalem, Israel
Date: June 26 2009
Time: 2:00PM to 3:00PM
Location: 32-D508
Host: Polina Golland, CSAIL
Contact: Polina Golland, x8005, polina@csail.mit.edu
We present a new non-uniform patient specific sampling method for the
estimation of mutual information in multi-modal image rigid
registration. The method uses the 3D Fast Discrete Curvelet Transform
to identify the most informative regions in the image on which the
mutual information is computed. It has better properties compared to
other non-uniform sampling schemes for mutual information computation.
Extensive evaluation on 20 validated clinical brain CT images to
Proton Density (PD) and T1 and T2-weighted MRI images from the public
RIRE database show the effectiveness of our method. For 10 clinical
targets for each patient, our method yields a mean target registration
error of 0.73mm (std=0.3mm) and maximal registration error of 1.6mm
with respect to the ground truth. The accuracy improvement is 0.3mm
on average and 1mm for the patient worst case average with respect to
existing sampling methods. Our method has the lowest mean
registration errors recorded to date for CT-MR image registration in
the RIRE website for methods tested on more than than five datasets.
At the end of my lecture I'll present a new application of fMRI
analysis for the detection and classification of several liver
diseases such us colorectal metastases and fibrosis.
See other events that are part of Biomedical Imaging and Analysis 2008/2009
See other events happening in June 2009
