The T2w image was registered to the T1w image using FSL’s FLIRT (Jenkinson et al., 2002 (link)) with 6 parameters (rigid body) and the mutual information cost function. This registration precisely aligned all brain regions except for small portions of ventral orbitofrontal cortex, overlying the sphenoid sinus, and inferior temporal cortex, overlying the mastoid air cells. In these areas, the gradient echo T1w and spin echo T2w data were affected differently by magnetic susceptibility-induced signal dephasing and signal loss (see artifactual results). The T2w image was resampled using the spline interpolation algorithm of FSL’s applywarp tool. Spline interpolation minimizes the white matter and CSF contamination of grey matter voxels that would result from the volumetric blurring inherent in trilinear interpolation. Spline interpolation yielded similar results when applied only to the T2w image or when applied separately to both the T1w and T2w images so that they were resampled the same number of times.
Division of the T1w image by the aligned T2w image mathematically cancels the signal intensity bias related to the sensitivity profile of the radio frequency receiver coils, which is the same in both images. Taking the ratio also increases the contrast related to myelin content. A simple approximation (Eq. 1 ) explains both effects: if myelin contrast is represented by x in the T1w image and 1/x in the T2w image and the receive bias field is represented by b in both images, the T1w/T2w ratio image equals x2, i.e. enhanced myelin contrast, with no bias field contribution. Because the noise in the T1w and T2w images is uncorrelated, there is increased myelin contrast relative to the noise (i.e. increased CNR).
Alternative bias field correction methods such as FSL’s FAST (Zhang et al., 2002 ) and MINC’s nu_correct (Sled et al., 1998 ) are not sufficiently accurate for the myelin mapping technique presented here. As demonstrated below, myelin mapping relies on detection of subtle differences in grey matter intensity that are obscured by either incomplete correction of the bias field or by errors in the bias field that can occur around the exterior of the brain. These errors take the form of local inhomogeneities between superficial cortex on the gyral crowns and deeper cortex in the fundi of sulci, and they result from the steep image intensity gradient between brain tissue and extra-cerebral tissues. These errors become more apparent when one runs a bias field correction utility multiple times in an attempt to completely remove the bias field. Intensity variations due to transmit field biases are minimal when using body transmit coils, as used here with the Siemens 3T Trios, because such coils produce very uniform transmit fields over the head. Further, some of the residual biases from the transmit field may also be reduced when dividing the images, since, while the transmit profiles between the two sequences are different; they are correlated. Indeed, there was no discernible global signal bias in our T1w/T2w ratio images, as the low frequency variations in grey and white matter were anti-correlated. We would expect them to be correlated if a bias field were present, as they are in the raw T1w and T2w images. These assumptions will not apply at higher resonant frequencies (i.e. at higher field strengths like 7T) where local transmit coils are used and where the transmit field biases are much stronger (Van de Moortele et al., 2009 (link)). In this case, it will be necessary to use sequences for the ratio that have very similar transmit profiles.
In volume slices of T1w and T2w images, interesting local signal inhomogeneities are evident in the grey matter, particularly in regions such as the central sulcus (Fig. 1A,B ). These inhomogeneities are enhanced in the T1w/T2w ratio images (Fig. 1C ). When a color palette is used instead of grey scale, the differences become even more apparent (Fig. 1D ). The boundaries drawn on the colorized T1w/T2w image in Figure 1D represent putative transitions between cortical areas (see Results). Indeed, a direct comparison between myelin stained histology and T1 contrast in the central sulcus reported a similar border between areas 4 and 3a that was aligned in both methodologies (Geyer et al., 2011 (link)).
Division of the T1w image by the aligned T2w image mathematically cancels the signal intensity bias related to the sensitivity profile of the radio frequency receiver coils, which is the same in both images. Taking the ratio also increases the contrast related to myelin content. A simple approximation (
Alternative bias field correction methods such as FSL’s FAST (Zhang et al., 2002 ) and MINC’s nu_correct (Sled et al., 1998 ) are not sufficiently accurate for the myelin mapping technique presented here. As demonstrated below, myelin mapping relies on detection of subtle differences in grey matter intensity that are obscured by either incomplete correction of the bias field or by errors in the bias field that can occur around the exterior of the brain. These errors take the form of local inhomogeneities between superficial cortex on the gyral crowns and deeper cortex in the fundi of sulci, and they result from the steep image intensity gradient between brain tissue and extra-cerebral tissues. These errors become more apparent when one runs a bias field correction utility multiple times in an attempt to completely remove the bias field. Intensity variations due to transmit field biases are minimal when using body transmit coils, as used here with the Siemens 3T Trios, because such coils produce very uniform transmit fields over the head. Further, some of the residual biases from the transmit field may also be reduced when dividing the images, since, while the transmit profiles between the two sequences are different; they are correlated. Indeed, there was no discernible global signal bias in our T1w/T2w ratio images, as the low frequency variations in grey and white matter were anti-correlated. We would expect them to be correlated if a bias field were present, as they are in the raw T1w and T2w images. These assumptions will not apply at higher resonant frequencies (i.e. at higher field strengths like 7T) where local transmit coils are used and where the transmit field biases are much stronger (Van de Moortele et al., 2009 (link)). In this case, it will be necessary to use sequences for the ratio that have very similar transmit profiles.
In volume slices of T1w and T2w images, interesting local signal inhomogeneities are evident in the grey matter, particularly in regions such as the central sulcus (