Abstract
Non-rigid image registration is widely used in the analysis of brain images to the extent it is provided as a standard tool in common packages such as SPM. However the performance of algorithms in specific applications remains hard to measure. In this paper a detailed comparison of the performance of an affine, B-Spline control-point and viscous fluid registration algorithm in inter-subject brain registration is presented. The comparison makes use of highly detailed expert manual labellings of a range of structures distributed in scale and in location in the brain. The overall performance is B-Spline, fluid, affine (best first) with all algorithms struggling to improve the match of smaller structures. We discuss caveats, evaluation strategies for registration and implications for future registration-based neuroimaging studies.
| Original language | English |
|---|---|
| Pages (from-to) | 679-686 |
| Number of pages | 8 |
| Journal | Lecture Notes in Computer Science |
| Volume | 3216 |
| Issue number | PART 1 |
| DOIs | |
| Publication status | Published or Issued - 2004 |
| Externally published | Yes |
| Event | Medical Image Computing and Computer-Assisted Intervention, MICCAI 2004 - 7th International Conference, Proceedings - Saint-Malo, France Duration: 26 Sept 2004 → 29 Sept 2004 |
ASJC Scopus subject areas
- Theoretical Computer Science
- General Computer Science