Premium
Automatic segmentation of white matter hyperintensities in the elderly using FLAIR images at 3T
Author(s) -
Gibson Erin,
Gao Fuqiang,
Black Sandra E.,
Lobaugh Nancy J.
Publication year - 2010
Publication title -
journal of magnetic resonance imaging
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.563
H-Index - 160
eISSN - 1522-2586
pISSN - 1053-1807
DOI - 10.1002/jmri.22004
Subject(s) - fluid attenuated inversion recovery , hyperintensity , voxel , thresholding , segmentation , pattern recognition (psychology) , computer science , artificial intelligence , magnetic resonance imaging , medicine , radiology , image (mathematics)
Purpose To determine the precision and accuracy of an automated method for segmenting white matter hyperintensities (WMH) on fast fluid‐attenuated inversion‐recovery (FLAIR) images in elderly brains at 3T. Materials and Methods FLAIR images from 18 individuals (60–82 years, 9 females) with WMH burdens ranging from 1–80 cm 3 were used. The protocol included the removal of clearly hyperintense voxels; two‐class fuzzy C‐means clustering (FCM); and thresholding to segment probable WMH. Two false‐positive minimization (FPM) methods using white matter templates were tested. Precision was assessed by adding synthetic hyperintense voxels to brain slices. Accuracy was validated by comparing automatic and manual segmentations. Whole‐brain, voxel‐wise metrics of similarity, under‐ and overestimation were used to evaluate both precision and accuracy. Results Precision was high, as the lowest accuracy in the synthetic datasets was 93%. Both FPM strategies successfully improved overall accuracy. Whole‐brain accuracy for the FCM segmentation alone ranged from 45%–81%, which improved to 75%–85% using the FPM strategies. Conclusion The method was accurate across the range of WMH burden typically seen in the elderly. Accuracy levels achieved or exceeded those of other approaches using multispectral and/or more sophisticated pattern recognition methods. J. Magn. Reson. Imaging 2010;31:1311–1322. © 2010 Wiley‐Liss, Inc.