Registration of Medical Images

Registration is a recently emerged task in medical image processing used to match two independently acquired images. To register images, the geometrical relationship between them is to be determined. Matching all the geometric data available for a patient provides better diagnostic capability, better understanding of data, and improves surgical and therapy planning and evaluation.

Here you can find a short summary of methods we implemented at József Attila University Szeged, Hungary.

Not always necessary for the registration itself but for processing and inspecting the results, a tool for displaying the medical image data is essential. We developed a Windows-based application which provides the following functions:

User guide and zipped PC executables to download can be found here.

Point-based methods provide easy and robust solution to several problems. Selected corresponding pairs of points can be extrinsic (external marker points from e.g. stereotactic frames) or instrinsic (e.g. anatomical points). Disadvantage is that a good visualization tool is necessary for selecting the points and this process needs anatomical knowledge.

We implemented the following methods:

• Rigid-body method using SVD (proposed by Arun et al.)
• Rigid-body method using Levemberg-Marquardt minimization technique
• Affine method (proposed by Tanács, Palágyi, Kuba)
• TPS method (proposed by Goshtasby, Bookstein)

Automatic methods require no serious user interactivity. Nowadays they are quite fast when the problem is linear. Visual inspection is generally necessary because of possible misregistrations.

Implemented methods:

• Cross-correlation based rigid-body (Palágyi and Udupa)
  For unimodality problems only.
• SSVD based affine and projective (Szelisky and Coughlan)
  For unimodality problems only.
• Mutual information based rigid-body (Tanács, Palágyi, Kuba)
  We participated the registration evaluation project at Vanderbilt University, TN, USA. Detailed information on this project can be found here.
• Mutual information based non-linear (Tanács, Palágyi, Kuba)
  2D only, under development. Animated GIFs (cca 350 kbytes).