Optimizing fractional intensity threshold for FSL-Brain Extraction Tool (BET) and comparing with freeSurfer on 3D T1W MR images

Yudthaphon Vichianin, Adisorn Kareesaw, Orasa Chawalparit, Masafumi Ohki

Research output: Contribution to journalArticle

Abstract

Objective: To find out the optimal Brain Extraction Tool (BET) parameter (fractional intensity threshold) for measuring the brain volume compared with the standard manual method in our institute and to compare with those of automated FreeSurfer software Methods: This retrospective study was performed in 10 healthy adult subjects with data of 3D-T1W on 3T MR machine. The manual gold standard brain volume measurements were done by two independent readers. The automated segmentations using BET with varied parameters and FreeSurfer software were also performed. Then, the two automated methods were compared with the manual tracing to make the optimal parameter by seeking for the highest intraclass correlation coefficient (ICC) using SPSS software. Results: The fractional intensity threshold for whole brain volume measurement of 0.1- 0.6 showed high ICC with the manual gold standard (ranging from 0.639 to 0.748). The best value was 0.1, showing highest ICC of 0.748 (p < 0.006) with confidence interval of 95% equal to (0.242; 0.932). There were no optimal parameters for right and left hippocampus volume measurement by BET due to very low ICC between BET and the reader (ICC ranging from 0.017 to 0.139 and from 0.012 to 0.110, for the right and left hippocampus volume, respectively). The ICC values of the automated FreeSurfer method with the manual tracing were also very low (0.063, 0.068, and 0.063 for right, left and bilateral hippocampi respectively). Conclusion: The optimal BET parameter (fractional intensity threshold) for automated brain volume measurement in our institute is 0.1 similar to the suggested value by prior study with high agreement (ICC=0.748) to the manual method.

Original languageEnglish
Pages (from-to)391-396
Number of pages6
JournalSiriraj Medical Journal
Volume70
Issue number5
DOIs
Publication statusPublished - Sep 1 2018

    Fingerprint

All Science Journal Classification (ASJC) codes

  • Medicine(all)

Cite this