Evaluation of a template-based B1 field correction approach for 3T MRI brain images

Purpose: To evaluate a template-based approach to correct B1 field inhomogeneity in brain MRI using B1 maps from other subjects to reduce the acquisition time. Introduction: Accurate estimation of T1 maps is increasingly important for some clinical applications. Low noise, high resolution, fast and accurate T1 maps from MRI images of the brain can be performed using a dual flip angle method [1]. However, B1 inhomogeneity limits the ability of the scanner to deliver the prescribed flip angle and hence introduces errors into the T1 map [2]. Obtaining a B1 map at the time of imaging can correct for this error at the expense of increasing imaging time. In this work, a template-based approach using previously acquired B1 maps is used to obviate the need for acquiring B1 maps in each subject. Methods: T1 maps were generated using a dual flip angle method (5° and 15°) after rigid registration of two 3D T1W fast field echo (FFE) images, and B1 maps were acquired using a dual repetition time (50ms and 250ms) strategy with similar sequences [3] on a 3T Philips Scanner (Philips Healthcare, Best, NL) in five volunteers. T1 maps were computed without B1 correction and also with B1 correction obtained at the same scan session to create a “reference” (R) T1 map. Additionally, each B1 map was transformed by means of an affine registration algorithm to match the geometry of the other subjects. Afterwards, each T1 map was also corrected using the other non-subject-specific B1 maps. The quality of each correction was characterized by the percentage of voxels having a relative difference less than 10% with respect to the reference T1 map. Intensity histograms were generated for each map and white matter (WM) and gray matter (GM) peaks were computed from a 3-Gaussian fitting. The T1 value of those peaks was compared in three series: reference T1 map, best non-patient-specific T1 map and the average values for the other corrections using a ANOVA single factor test to determine whether or not those series had different mean values. Results: Histograms of T1 values when B1 correction was not used were unimodal, i.e. the two peaks corresponding to WM and GM tissues were not distinguishable. The T1 histograms when B1 correction was applied had two distinguishable peaks (Table 1) with relaxation times in agreement with previously reported data [4]: WM from 1000 to 1100 ms and various GM tissues between 1200 and 1700 ms. Fig.1 shows the T1 histograms for Subject #5 without B1 correction, corrected using the B1 map from the same subject, and corrected using the B1 map from Subject #1, which is the non-subject-specific correction that showed the best performance for that particular subject. In that case corrected maps exhibit similar distributions and 76% of their voxels have intensities that differ in less than 10% with respect to the other map (Table 2). That percentage drops to 28% when comparing the reference map to the non-corrected map. For all subjects, every non-subject-specific correction significantly improved the quality of the T1 map (Table 2). Figures 2 and 3 show the intensity value of WM and GM peaks, respectively, for the reference (R), best corrected (B) and mean T1 values of the other corrected T1 maps (M). Table 3 shows that R and B series do not differ. Although p-values computed when comparing R and M series are not sufficiently small to prove that those series are different, that is due to the small size of the sample and significance could be achieved by considering as many as twice the number of subjects. Subject WM GM WM GM # 1 824 1271 1028 1430 # 2 899 1323 1156 1662 # 3 94