Fully automatic segmentation of the brain from T1-weighted MRI usingBridge Burneralgorithm

    loading  Checking for direct PDF access through Ovid

Abstract

Purpose

To validate Bridge Burner, a new brain segmentation algorithm based on thresholding, connectivity, surface detection, and a new operator of constrained growing.

Materials and Methods

T1-weighted MR images were selected at random from three previous neuroimaging studies to represent a spectrum of system manufacturers, pulse sequences, subject ages, genders, and neurological conditions. The ground truth consisted of brain masks generated manually by a consensus of expert observers. All cases were segmented using a common set of parameters.

Results

Bridge Burner segmentation errors were 3.4% ± 1.3% (volume mismatch) and 0.34 ± 0.17 mm (surface mismatch). The disagreement among experts was 3.8% ± 2.0% (volume mismatch) and 0.48 ± 0.49 mm (surface mismatch). The error obtained using the brain extraction tool (BET), a widely used brain segmentation program, was 8.3% ± 9.1%. Bridge Burner brain masks are visually similar to the masks generated by human experts. Areas affected by signal intensity nonuniformity artifacts were occasionally undersegmented, and meninges and large sinuses were often falsely classified as the brain tissue. Segmentation of one MRI dataset takes seven seconds.

Conclusion

The new fully automatic algorithm appears to provide accurate brain segmentation from high-resolution T1-weighted MR images. J. Magn. Reson. Imaging 2008;27:1235–1241. © 2008 Wiley-Liss, Inc.

Related Topics

    loading  Loading Related Articles