An improved FSL-FIRST pipeline for subcortical gray matter segmentation to study abnormal brain anatomy using quantitative susceptibility mapping (QSM)

Abstract

Accurate and robust segmentation of subcortical gray matter (SGM) nuclei is required in many neuroimaging applications. FMRIB's Integrated Registration and Segmentation Tool (FIRST) is one of the most popular software tools for automated subcortical segmentation based on T₁-weighted (T1w) images. In this work, we demonstrate that FIRST tends to produce inaccurate SGM segmentation results in the case of abnormal brain anatomy, such as present in atrophied brains, due to a poor spatial match of the subcortical structures with the training data in the MNI space as well as due to insufficient contrast of SGM structures on T1w images. Consequently, such deviations from the average brain anatomy may introduce analysis bias in clinical studies, which may not always be obvious and potentially remain unidentified. To improve the segmentation of subcortical nuclei, we propose to use FIRST in combination with a special Hybrid image Contrast (HC) and Non-Linear (nl) registration module (HC-nlFIRST), where the hybrid image contrast is derived from T1w images and magnetic susceptibility maps to create subcortical contrast that is similar to that in the Montreal Neurological Institute (MNI) template. In our approach, a nonlinear registration replaces FIRST's default linear registration, yielding a more accurate alignment of the input data to the MNI template. We evaluated our method on 82 subjects with particularly abnormal brain anatomy, selected from a database of >2000 clinical cases. Qualitative and quantitative analyses revealed that HC-nlFIRST provides improved segmentation compared to the default FIRST method.

Keywords: Brain; FIRST; MRI; Quantitative susceptibility mapping; Segmentation; Subcortical gray matter.

Fig. 1

Representative examples of T1w images of eight subjects who were included in the study showing particularly abnormal brain anatomy (atrophy).

Fig. 2

Linear registration to the MNI space of a T1w image of one subject with severe atrophy. (a) shows an axial view of the original T1w image, (b) the linearly registered T1w image, and (c) the corresponding MNI template. While the position misalignment between the original T1w image (a) and the MNI template (c) can be resolved by applying a translation during linear registration, the mismatch of the ventricles and the caudate nuclei, however, remain between the registered image (b) and the MNI template (c).

Fig. 3

Flow chart of the proposed HC-nlFIRST segmentation framework. T1bc is the abbreviation of bias field corrected T1w image, T1norm stands for the intensity normalized, bias field corrected T1w image, GRE abbreviates gradient echo, and QSM denotes the quantitative susceptibility map.

Fig. 4

Three-dimensional rendering of the subcortical reference ROIs in MNI space from four different views of (a) left, (b) right, (c) top, and (d) bottom. Abbreviations in coordinate system: S – superior, R – right, A – anterior, P – posterior. Different colors are used to indicate the subcortical GM structures: pink – putamen, cyan – caudate, orange – accumbens, yellow – hippocampus, blue – globus pallidus, and green – thalamus.

Fig. 5

Exemplary slices of subject #2 at the level of the basal ganglia. (a) T1w image, (b) T1n image, (c) quantitative susceptibility map, and (d) the HC image, which was obtained by combining the T1n image and the susceptibility map (w₁=1.50 and w₂=−111.19 ppm⁻¹). Improved SGM contrast with respect to the surrounding white matter is discernible on the HC image (d) compared to the other T1w images (a,b). The arrows point to the globus pallidus (red) and thalamus (yellow) whose contrast relative to the internal capsule was substantially enhanced in the HC image (d). Bias field correction successfully suppressed low frequency, non-uniform intensity noticeable in the center of the T1w image, but did not affect the SGM contrast. See Fig. S2 (in Supplementary Material) for a different subject of the non-training dataset.

Fig. 6

Spatial registration to the MNI template. (a) MNI template and manual reference ROIs overlaid on the MNI template in (d). The registered HC images using FLIRT and nonlinear ANTs are presented in (b) and (c), and the subcortical reference labels warped in MNI space using FLIRT and nonlinear ANTs are shown in (e) and (f), respectively. The mean and standard deviations of the Dice coefficients and volume overlaps of the warped SGM labels with respect to those of the MNI template are displayed in the bottom row in (g) and (h), respectively.

Fig. 7

Visual assessment of subcortical segmentation with the different segmentation methods averaged over all 82 subjects. The error bars indicate the standard deviation. A significant difference in the rating scores between HC-nlFIRST and one of the other segmentation methods is denoted by asterisks (*** stands for p<0.001, * stands for p<0.05).

Fig. 8

Comparison between manual and automated subcortical segmentation (pink – putamen, cyan – caudate, orange – accumbens, yellow – hippocampus, blue – globus pallidus, green – thalamus) of (a) an MS patient with severe brain atrophy (subject #1) and (b) a healthy subject (subject #6). The axial and coronal views of the hybrid contrast images, shown in the left column, are superposed from left to right with the boundaries of the segmentation results due to manual delineation, FIRST, T1n-nlFIRST, HC-nlFIRST and atlas-based segmentations, respectively. The arrows indicate inaccurately segmented regions compared to the manual reference. Segmentation results overlaid on the susceptibility map are displayed in Fig. S4 (see Supplementary Material).

Fig. 9

Quantitative analysis of the training dataset (n=8) for the different segmentation approaches. The mean and standard deviations of Dice coefficients and FNRs are shown in the top and bottom rows, respectively. For the outliers, the standard deviations of the Dice coefficients of the default FIRST method are 0.266 for the right putamen, 0.257 for the right caudate, 0.323 and 0.305 for the left and right hippocampus. The standard deviations of FNRs of atlas-based method are 0.230 and 0.389 for the left and right accumbens, respectively. (Puta – putamen; Caud – caudate; Accu – accumbens; Hipp – hippocampus; GP – globus pallidus; Thal – thalamus)

Fig. 10

Boxplots of the calculated figure-of-merit (F) after bootstrapping for each segmentation method and SGM structure (horizontal axis: ‘1’ – FIRST ‘2’ – T1n-nlFIRST ‘3’ – HC-nlFIRST ‘4’ – atlas-based method). The lower figure-of-merit indicates that the segmentation method is more accurate. The title of each plot shows the respective SGM structure, where the prefixes L and R stand for the left and right hemisphere, respectively. (Puta – putamen; Caud – caudate; Accu – accumbens; Hipp – hippocampus; GP – globus pallidus; Thal – thalamus)

Fig. 11

Averaged Dice coefficients across all SGM structures plotted for all individual subjects. (blue squares – FIRST, orange crosses – T1n-nlFIRST, black dots – HC-nlFIRST). Substantial wrong or aborted segmentations are indicated by the black arrows. The dashed red line indicates an empirically selected threshold to identify failed and successful segmentations.