Unified segmentation based correction of R1 brain maps for RF transmit field inhomogeneities (UNICORT)

Abstract

Quantitative mapping of the longitudinal relaxation rate (R1=1/T1) in the human brain enables the investigation of tissue microstructure and macroscopic morphology which are becoming increasingly important for clinical and neuroimaging applications. R1 maps are now commonly estimated from two fast high-resolution 3D FLASH acquisitions with variable excitation flip angles, because this approach is fast and does not rely on special acquisition techniques. However, these R1 maps need to be corrected for bias due to RF transmit field (B1(+)) inhomogeneities, requiring additional B1(+) mapping which is usually time consuming and difficult to implement. We propose a technique that simultaneously estimates the B1(+) inhomogeneities and R1 values from the uncorrected R1 maps in the human brain without need for B1(+) mapping. It employs a probabilistic framework for unified segmentation based correction of R1 maps for B1(+) inhomogeneities (UNICORT). The framework incorporates a physically informed generative model of smooth B1(+) inhomogeneities and their multiplicative effect on R1 estimates. Extensive cross-validation with the established standard using measured B1(+) maps shows that UNICORT yields accurate B1(+) and R1 maps with a mean deviation from the standard of less than 4.3% and 5%, respectively. The results of different groups of subjects with a wide age range and different levels of atypical brain anatomy further suggest that the method is robust and generalizes well to wider populations. UNICORT is easy to apply, as it is computationally efficient and its basic framework is implemented as part of the tissue segmentation in SPM8.

Fig. 1

Dependence of UNICORT performance on the choice of the smoothness (FWHM) and regularization (κ) parameters (see also Table 1). Deviation of UNICORT corrected R1 maps (D_unicort) from the established standard using measured B1⁺ maps.

Fig. 2

Spatially normalized and averaged R1 maps of healthy volunteers with different corrections for RF transmit field inhomogeneities. (a) R1 map corrected with UNICORT (R1_UNICORT; left column), uncorrected apparent R1 map (R1_app; center column), R1 map corrected with measured B1⁺ map (R1_m; right column). In the uncorrected map, the spuriously increased R1 values in the center of the brain due to RF inhomogeneities can be well seen (red arrow). (b) Percent differences in R1 values between the B1⁺ map corrected R1 map (R1_m) and the UNICORT (R1_UNICORT; left column) or uncorrected apparent (R1_app; right column) R1 map, respectively. In the difference between R1_app and R1_m the typical asymmetric RF inhomogeneities can be appreciated (green arrow).

Fig. 3

R1 maps of single subject with enlarged ventricles with different corrections for RF transmit field inhomogeneities. For detailed description, see Fig. 2. A similar pattern and magnitude of bias reduction was observed as in the group of healthy volunteers (Fig. 2).

Fig. 4

B1⁺ map (in percent of the nominal flip angle) of single subject (same as in Fig. 3) estimated by UNICORT (B1_UNICORT; left column) and measured with the 3D EPI technique (B1_m; center column). Percent differences in B1⁺ values between B1_UNICORT and B1_m (B1_UNICORT − B1_m; right column). At the edge of the brain, in the connective tissue and muscles the measured B1⁺ maps are less accurate and cause edge artifacts.

Fig. 5

Median percent deviation of the UNICORT corrected and uncorrected R1_app maps compared to the B1⁺ map corrected R1_m maps. Descriptive statistics are provided for the group of healthy volunteers and presymptomatic Huntington disease gene mutation carriers (PSC): blue box = 25%/75% percentile, red line = median, black whisker = most extreme data value excluding outliers, red cross = outlier (probability < 0.01 under assumption of normally distributed data).

Fig. 6

Median percent deviation of the B1⁺ map estimated by UNICORT from the measured B1⁺ map for the group of healthy volunteers and presymptomatic Huntington disease gene mutation carriers (PSC). For explanation of the box plot see Fig. 5.