Article Text

PDF
342 SEMIAUTOMATED SEGMENTATION OF THE CAUDATE NUCLEUS.
  1. A. B. Williams1,
  2. E. H. Aylward1,
  3. M. F. Beg1,
  4. A. Khan1,
  5. P. Konopa1,
  6. K. C. Stegbauer1,
  7. V. Yallapragada1
  1. 1Department of Radiology, University of Washington School of Medicine, Seattle, WA; Medical Image Analysis Lab, Simon Fraser University, Vancouver, BC.

Abstract

Background The basal ganglia (BG) modulate cerebral output by forming loops with the thalamus and cortex. The striatum is a portion of the BG, which includes the caudate and putamen, both of which exhibit significant neuronal degeneration in Huntington's disease (HD) in the decade leading up to the onset of clinical symptoms and throughout the course of the clinical disease. Thus, striatal volume should prove to be a sensitive biomarker for HD clinical trials. Traditionally, segmentation of striatal shape and volume has been performed manually on sequential MRI cross sections using various computer programs. Automated segmentation programs could prove to be a more effective way to map structures as they should, in principle, be faster, be more objective, and not require extensive training. The purpose of this study was to test SegLDDMM, an experimental computer program designed to automatically quantify the shape and volume of the caudate nuclei.

Study Design and Methods Two people ran 18 scans through the program. This involved creating computer models of the lateral ventricles and placing landmarks along the caudate ridges. Results from the two SegLDDMM users were compared against each other, as well as against the manual tracings of a third individual who had demonstrated proficiency in defining the boundaries of the caudate nucleus.

Results As a measure of validity, the interclass correlations between the manual tracer and the two SegLDDMM users were calculated; these values were 0.76 and 0.81. As a measure of reliability, the interclass correlation between the two SegLDDMM users was calculated and found to be 0.91. When files were analyzed to visually assess computer caudate estimates, several estimates (6 of the 18) were obviously wrong, with other regions of the brain (often the thalamus) inappropriately included as caudate. When these scans were excluded, the correlation was 0.97 between program users.

Conclusion Although SegLDDMM is not ready to be marketed, we believe it has potential to assist researchers in the future, although program parameters need to be redefined.

Statistics from Altmetric.com

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.