Automatic brain segmentation using fractional signal modeling of a multiple flip angle, spoiled gradient-recalled echo acquisition

Abstract

The aim of this study was to demonstrate a new automatic brain segmentation method in magnetic resonance imaging (MRI). The signal of a spoiled gradient-recalled echo (SPGR) sequence acquired with multiple flip angles was used to map T1, and a subsequent fit of a multi-compartment model yielded parametric maps of partial volume estimates of the different compartments. The performance of the proposed method was assessed through simulations as well as in-vivo experiments in five healthy volunteers. Simulations indicated that the proposed method was capable of producing robust segmentation maps with good reliability. Mean bias was below 3% for all tissue types, and the corresponding similarity index (Dice's coefficient) was over 95% (SNR = 100). In-vivo experiments yielded realistic segmentation maps, with comparable quality to results obtained with an established segmentation method. Relative whole-brain cerebrospinal fluid, grey matter, and white matter volumes were (mean ± SE) respectively 6.8 ± 0.5, 47.3 ± 1.1, and 45.9 ± 1.3% for the proposed method, and 7.5 ± 0.6, 46.2 ± 1.2, and 46.3 ± 0.9% for the reference method. The proposed approach is promising for brain segmentation and partial volume estimation. The straightforward implementation of the method is attractive, and protocols that already rely on SPGR-based T1 mapping may employ this method without additional scans.