Automated coregistration and statistical analyses of SPECT brain images

Journal Article · Sun May 01 04:00:00 EDT 1994 · Journal of Nuclear Medicine

OSTI ID:198040

Gong, W; Devous, M D ^[1]

---

1. Univ. of Texas Southwestern Medical Center, Dallas, TX (United States)

Statistical analyses of SPECT image data often require highly accurate image coregistration. Several image coregistration algorithms have been developed. The Pellizari algorithm (PA) uses the Powell technique to estimate transformation parameters between the {open_quotes}head{close_quotes} (model) and {open_quotes}hat{close_quotes} (images to be registered). Image normalization and good initial transformation parameters heavily affect the accuracy and speed of convergence of the PA. We have explored various normalization methods and found a simple technique that avoids most artificial edge effects and minimizes blurring of useful edges. We have tested the effects on accuracy and convergence speed of the PA caused by different initial transformation parameters. From these data, a modified PA was integrated into an automated coregistration system for SPECT brain images on the PRISM 3000S under X Windows. The system yields an accuracy of approximately 2 mm between model and registered images, and employs minimal user intervention through a simple graphic user interface. Data are automatically resliced, normalized and coregistered, with the user choosing only the slice range for inclusion and two initial transformation parameters (under computer-aided guidance). Coregistration is accomplished (converges) in approximately 8 min for a 128 x 128 x 128 set of 2 mm{sup 3} voxels. The complete process (editing, reslicing, normalization, coregistration) takes about 20 min. We have also developed automated 3-dimensional parametric images ({open_quotes}t{close_quotes}, {open_quotes}z{close_quotes}, and subtraction images) from coregistered data sets for statistical analyses. Data are compared against a coregistered normal control group (N = 50) distributed in age and gender for matching against subject samples.

• OSTI ID: 198040
• Report Number(s): CONF-940605--
• Journal Information: Journal of Nuclear Medicine, Journal Name: Journal of Nuclear Medicine Journal Issue: Suppl.5 Vol. 35; ISSN JNMEAQ; ISSN 0161-5505
• Country of Publication: United States
• Language: English

Related Subjects

55 BIOLOGY AND MEDICINE
BASIC STUDIES
ACCURACY
ALGORITHMS
BRAIN
DATA ANALYSIS
IMAGES
PATIENTS
SINGLE PHOTON EMISSION COMPUTED TOMOGRAPHY