Enhanced brain extraction improves the accuracy of brain atrophy estimation.

BET (Brain Extraction Tool) is a widely used computer program to automatically separate brain from non-brain structures in MR images. This procedure is used in SIENAX and SIENA, which are robust approaches to quantifying brain volume (atrophy state) and volume change (atrophy rate), respectively. Occasionally, however, BET produces imperfect results (e.g., inclusion of non-brain structures). This is usually either ignored (if inaccuracies are small) or corrected by manual adjustment, with the disadvantages of user intervention. We describe here a new, automated option in BET. This is based on the original BET, but uses standard-space masking to remove tissue around the eyes, and further morphological operations and thresholding to refine eyeball removal and eliminate additional non-brain tissues. To assess whether the new BET procedure improves brain volume measurements, this was compared with the traditional and manual editing procedures in SIENA and SIENAX. Measures of atrophy rate and state were significantly higher with the traditional procedure than with the manual editing and new procedures. In contrast, both atrophy measures were almost identical and highly correlated when the manual editing and new procedures were used. The voxels excluded with these two procedures showed close overlap, as judged by the Dice overlap coefficient. We conclude that, in SIENA and SIENAX, the proposed BET procedure shows results matching those obtained after manual editing, thus more closely approximating the "true" brain volume. Multicentre studies monitoring brain atrophy in clinical trials may receive benefit by using this unbiased, fully automated procedure.