CVPR 2001 Short Course

Medical Image Analysis: Deformable Models and Registration

Instructor: Hervé Delingette (INRIA, Sophia Antipolis, France)

Duration: 3-4 hours

Abstract:

After a brief introduction on the increasing importance of digital image processing for diagnosis and therapy, and an overview of emerging applications involving image guided therapy, augmented reality and medical robotics, we will present two rather central topics: deformable models and registration.

Deformable models can be developed for segmentation, tracking, and simulation. Various approaches will be reviewed and applications will include automatic contouring of organs, tracking of cardiac motion, and simulation of soft tissue behavior in surgery simulation.

Registration methods can be used to compare images taken at different times, and to fuse images of different modalities. We will present both feature based and intensity-based methods. Applications will include automatic detection of evolving pathologies, augmented reality for surgery, and labeling with an atlas.

Outline of the course:

1) Introduction:

- The different image modalities

- Volumetric images vs. video (2D) images

2) Image Segmentation:

- The problem of Image Segmentation

- Iconic-based image segmentation:

- Thresholding + mathematical morphology

- Classification algorithms: the EM algorithm.

Application of brain segmentation from MR images.

- Model-based image segmentation:

- Principle of deformable model segmentation

- A menagerie of deformable models

- Example of discrete meshes: Simplex meshes

- From registration to local deformation

- Boundary and region-based external forces

- Various examples and applications: liver, heart, ...

3) Image Registration:

- Geometric (feature-based) Registration

- Extraction of geometric features in volumetric images

(LVV, crest-lines, extremal points)

- Main Algorithms:

* Tree search

* Geometric hashing

* Iterative Closest Point

- Iconic (intensity-based) Registration

- Choice of a similarity measures:

* Sum of Square Differences (SSD)

* Normalized correlation coefficient

* Correlation Ratio

- Measure of deformation smoothness

- Minimization of a dual energy

(intensity similarity and deformation smoothness)

- Applications

- Multiple sclerosis evolution

- Augmented reality in the Operating Room

- Multimodal image fusion

- Atlas superposition

Back to Main