[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
Next Article in Journal
Risk Prioritizing with Weighted Failure Mode and Effects Analysis and Fuzzy Step-Wise Weight Assessment Ratio Analysis: An Application Software Service Provider Company in the Defense Industry
Previous Article in Journal
Investigation of CuTi Alloy for Applications as Non-Sparking Material
Previous Article in Special Issue
Noninvasive Deep Learning Analysis for Smith–Magenis Syndrome Classification
You seem to have javascript disabled. Please note that many of the page functionalities won't work as expected without javascript enabled.
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Applied Sciences
Volume 14
Issue 24
10.3390/app142411575
applsci-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles thumb_up
...
Endorse textsms
...
Comment
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Assessment of Regional Brain Volume Measurements with Different Brain Extraction and Bias Field Correction Methods in Neonatal MRI

by
Tânia F. Vaz
1,*,
Nima Naseh
2,
Lena Hellström-Westas
2,
Nuno Canto Moreira
3,
Nuno Matela
1 and
Hugo A. Ferreira
1
1
Instituto de Biofísica e Engenharia Biomédica, Faculdade de Ciências, Universidade de Lisboa, 1749-016 Lisbon, Portugal
2
Department of Women’s and Children’s Health, Uppsala University, 751 85 Uppsala, Sweden
3
Department of Neuroradiology, Karolinska University Hospital, 171 76 Stockholm, Sweden
*
Author to whom correspondence should be addressed.
Appl. Sci. 2024, 14(24), 11575; https://doi.org/10.3390/app142411575
Submission received: 12 November 2024 / Revised: 3 December 2024 / Accepted: 9 December 2024 / Published: 11 December 2024
(This article belongs to the Special Issue Methods, Applications and Developments in Biomedical Informatics: 2nd Edition)
Browse Figures
Figure 1
<p>Intensity inhomogeneity on TEA brain T2w MRI preterm neonate: (<b>a</b>) skull-stripped MRI without correction; (<b>b</b>) bias field map; bias-corrected MRI with (<b>c</b>) N4ITK and (<b>d</b>) SPM-BFC methods (images displayed with the Jet color map to ease the comparison between corrections).</p> "> Figure 2
<p>TEA brain T2w MRI (manually skull-stripped) of a preterm neonate in three different axial slices bias-corrected with SPM-BFC (<b>a</b>–<b>c</b>) and N4ITK (<b>d</b>–<b>f</b>), with overlaid brain tissues/structures segmented using MANTiS (color coded: CGM, red; WM, green; CSF, blue; DNGM, yellow; Hip, light blue; Amy, pink; CB, grey; BS, dark blue).</p> "> Figure 3
<p>Coefficient of joint variation (CJV) between white and grey matter intensities from MRI uncorrected and bias field-corrected with SPM-BFC and N4ITK, for each brain extraction method (Manual, BET2, SWS, HD-BET, and SynthStrip).</p> "> Figure 4
<p>Dice coefficient (DC) based on the regional brain tissue segmentation maps obtained from MRI uncorrected and bias field-corrected using SPM-BFC or N4ITK, for each brain extraction method (Manual, BET2, SWS, HD-BET, and SynthStrip).</p> "> Figure 5
<p>TEA brain T2w MRI (example bias-corrected with N4ITK) of a preterm neonate in three different axial slices (<b>a</b>) and corresponding skull-stripped images from each BE method ((<b>b</b>) Manual, (<b>c</b>) BET2, (<b>d</b>) SWS, (<b>e</b>) HD-BET, and (<b>f</b>) SynthStrip), overlaid with segmentation from MANTiS (color coded: CGM, red; WM, green; CSF, blue; DNGM, yellow; Hip, light blue; Amy, pink; CB, grey; BS, dark blue).</p> "> Figure A1
<p>Recovered bias field map overlapped with brain MRI in three different axial slices, with N4ITK bias field correction (BFC) applied before and after manual brain extraction (BE) (bias field map displayed in HSV color map to ease the comparison between corrections).</p> ">
Versions Notes

Abstract

:
Proper selection and application of preprocessing steps are crucial for obtaining accurate segmentation in brain Magnetic Resonance Imaging (MRI). The aim of this study is to evaluate the impact brain extraction (BE) and bias field correction (BFC) methods have on regional brain volume (RBV) measurements of preterm neonates’ T2w MRI at term-equivalent age (TEA). Five BE methods (Manual, BET2, SWS, HD-BET, SynthStrip) were applied together with two BFC methods (SPM-BFC and N4ITK), before segmenting the neonatal brain into eight tissue classes (cortical grey matter, white matter, cerebral spinal fluid, deep nuclear grey matter, hippocampus, amygdala, cerebellum, and brainstem) using an automated segmentation software (MANTiS). Quantitative assessments were conducted, including the coefficient of variation (CV), coefficient of joint variation (CJV), Dice coefficient (DC), and RBV. HD-BET, together with N4ITK, showed the highest performance (mean ± standard deviation) regarding CV of 0.047 ± 0.005 (white matter) and 0.070 ± 0.005 (grey matter), CJV of 0.662 ± 0.095, DC of 0.942 ± 0.063, and RBV without significant differences (except in the brainstem) from the manual segmentation. Therefore, such combination of methods is recommended for improved skull-stripping accuracy, intensity homogeneity, and reproducibility of RBV of T2w MRI at TEA.

1. Introduction

Preterm birth stands as the primary cause of neonatal mortality and is linked to short- and long-term physical, neurodevelopmental, and socioeconomic consequences, that increase with the degree of prematurity [1]. Prematurely born neonates often exhibit delayed maturation (e.g., delayed sulcation, delayed myelination, reduced brain volumes, and slower brain growth trajectories) compared to fetuses at the same postconceptional age and compared with neonates born at term when performing assessment at term-equivalent age (TEA) [2,3,4].
Magnetic resonance imaging (MRI) plays a crucial role in confirming normal anatomy and identifying perinatal injury that can be used to predict neurodevelopmental outcomes, namely, through volumetric analysis with regional brain volume (RBV) measurements [5,6,7,8,9], which poses unique challenges compared with adults [10]. The volumetric quantification of the brain in MRI allows for the comparison with the child’s development, correlating the neuroimaging phenotype with neurocognitive disorders and therapy outcomes [11]. Several studies have described associations between RBVs at TEA and early/long-term neurodevelopmental impairment (e.g., motor, sensory, cognitive, behavioral) [7,9,12,13,14,15], highlighting the importance of studying and optimizing methods for better RBV quantification.
Some image preprocessing steps are essential for subsequent brain tissue segmentation, namely, brain extraction (BE) or skull-stripping and bias field correction (BFC), also known as intensity inhomogeneity/nonuniformity correction [16,17]. Intracranial segmentation occurs prior to regional segmentation, and is of the greatest importance as it can significantly impact the estimation of volumes, especially of the tissues close to the edges of the brain contour that can become under- or overestimated depending on the BE performance when removing non-brain tissues [10,16]. Several approaches can be used for BE in the neonatal brain, ranging from conventional (deformable surface-based, mathematical morphology, intensity, template, and hybrid-based models) to deep learning methods [18,19,20]. As such, selection of a proper BE method is warranted [10]. Additionally, correction of intensity inhomogeneities in MRI is important because it can also affect quantitative image analysis. Intensity inhomogeneity in MRI is characterized by a gradual and smooth variation in intensity across the image, leading to different intensities for the same tissue depending on the region, and is dependent on the strength of the magnetic field [21]. Several causes of intensity inhomogeneity exist due to the MRI device (e.g., static magnetic field inhomogeneity, eddy currents compelled by field gradients, radiofrequency transmission/reception, and scanner calibration) or related to the human body regions to be acquired (e.g., position of the patient’s head not centered in the scanner, and magnetic susceptibility differences between tissues, such as interface between bone or air-filled sinuses and brain tissue) [21]. It can be unnoticeable to the human eye, but such intensity inhomogeneity affects the automated quantitative analysis, which is often limited by the assumption that a tissue’s voxel intensities remain similar throughout the image [21]. In neuroimaging, several methods have been proposed for BFC of brain MR images, namely, using prospective methods (e.g., based on phantoms [22,23], multicoils [24], and special sequences [25]), or using a retrospective approach depending on the real image information or some a priori knowledge about spatial and/or intensity probability distribution (e.g., filtering methods [26,27], surface fitting methods [28], segmentation-based methods [29,30,31,32,33], and histogram-based methods [34,35,36]). A popular method, considered the gold-standard, is the N4ITK [37] algorithm developed within the open-source Insight Toolkit (ITK) software [38], an extension of the N3 algorithm [34], that assumed a simple parametric model for the bias field and used an iterative and multi-scale optimization approach, not needing a segmentation or a priori knowledge from the scanner or subject. Another approach, using a non-iterative multi-scale method [39], showed superior performance in less time compared with the gold-standard. Recently, deep convolutional neural network methods (e.g., ABCnet [40], BIASNet [41], DeepN4 [42]) have been proposed for BFC in brain MR images, showing advantages over traditional algorithms because of not requiring modeling signals and bias fields or adjustment of parameters. Despite being very flexible and efficient, these methods still have limitations in their implementation and need to be trained on large datasets to achieve proper generalization for BFC of neonatal MRI data.
Therefore, the purpose of this study was to identify the impact that different preprocessing steps, namely, BE and BFC methods, have on RBV measurements from preterm neonates MRI scanned at TEA.

2. Materials and Methods

2.1. Subjects

The original dataset consisted of T2-weighted (T2w) MRI brain scans from 22 premature neonates (28.4 ± 2.1 (mean ± standard deviation) gestational weeks) without detectable brain lesions and congenital conditions, scanned at TEA at Uppsala University Hospital, a tertiary referral centre in Sweden. Details about the included subjects and MRI protocol were previously reported [10].

2.2. MRI Protocol

All MRI scans were acquired on a Siemens Avanto 1.5 Tesla scanner (Siemens Medical Systems, Erlangen, Germany) at Uppsala University Hospital using a neonatal-adapted imaging protocol. This study included 2D fast spin echo T2w images with the following acquisition parameters: axial slice number = 33; slice thickness = 3 mm; gap distance factor 20% interleaved (spacing between slices = 3.6 mm); field-of-view read/phase = 200 mm/100%; matrix size = 320 × 320; repetition time = 6520 ms; echo time = 103 ms; and flip angle = 120°.

2.3. Preprocessing Steps

Currently, there is no standard order about the preprocessing steps prior to neonatal brain MRI segmentation: some studies performed BFC after BE [43,44,45,46,47], whilst others the opposite [13,14,48,49,50], and even some studies do not mention or apply inhomogeneity correction before segmentation.
Based on that, the order of preprocessing steps was tested, assessing the brain masks obtained and the similarity level found between them. It was observed that similarity between masks was slightly better when the MR images were skull-stripped before bias field correction, than the opposite (results in Appendix A). Despite the difference being negligible, in this study, the following preprocessing order was considered: each subject MRI underwent BE with five methods (Manual, BET2, SWS, HD-BET, SynthStrip), followed by BFC with two methods (SPM-BFC and N4ITK). Additionally, all of the images were set into a common coordinate system before segmentation.

2.3.1. Brain Extraction Methods

The protocol for manual segmentation for BE, showing excellent (greater than 0.98) intra- and inter-rater reliability, and for the selection of automated methods was previously described and justified [10]. Five BE methods were selected for this study: manual segmentation, two conventional methods (Brain Extraction Tool (BET2) [51] and Simple Watershed Scalping (SWS) [52]), and two deep learning methods (HD Brain Extraction Tool (HD-BET) [53] and SynthStrip [54]).

2.3.2. Bias Field Correction Methods

The images were bias field corrected using two parametric methods with different optimization approaches to remove low-frequency intensity variations: a method developed and optimized by Ashburner and Friston [55] and implemented on the Statistical Parametric Mapping (SPM12) software package [56], which we denote as SPM-BFC; and the N4ITK algorithm, currently the most popular and widely used BFC method for MRI, proposed by Tustison et al. [37]. No deep learning BFC methods were used in this study due to several factors: code unavailability; trained models with adult datasets and T1-weighted MRI; or ground-truth (i.e., precise bias fields) images needed for model training before running the algorithm.
SPM-BFC is included in a unified segmentation framework, as the algorithm uses a probabilistic model that interleaves segmentation, registration, and BFC [55]. It is based on a finite Gaussian mixture, which is extended to include a smooth intensity variation and non-linear registration with tissue probability maps [48]. For running the SPM-BFC algorithm, the parameters defined on the Morphologically Adaptive Neonatal Tissue Segmentation [52] (MANTiS) toolbox for SPM12 were used [57]: amount of bias regularization—light or 0.001 (to limit biases due to smoothness and spatially varying artifacts that modulate the intensity of the image); full width at half maximum (FWHM) of Gaussian smoothness of bias—60 mm (to prevent the algorithm from trying to model out intensity variation due to different tissue types); number of Gaussians—2 (representing the intensity distribution for each tissue class); parameters for warping regularization—[0, 0.001, 0.5, 0.05, 0.2] (the objective function for registering the tissue probability maps to the image to process).
The N4ITK algorithm assumes that the value of each control point in the B-spline model is determined by N neighboring points. Instead of minimizing the squared difference of isolated points, it minimizes the sum of squared differences of the N neighboring points [37]. Since the algorithm is performed locally (instead globally as N3), it could help improving the robustness and stability, whereas the bias field estimates are based on the output of the previous iteration, improving its convergence [37]. The N4ITK algorithm was applied with the default parameters implemented in 3D Slicer [58]: B-spline grid resolution—[1, 1, 1] (initial resolution defined as a sequence of three numbers for each spatial dimension); bias field FWHM—0.00 mm; maximum number of iterations—[50, 40, 30] (performed at each level of B-spline grid resolution); convergence threshold—0.0001 (stopping criterion needed to decide when the iterative process has converged sufficiently and should terminate and move to the next stage); B-spline order—3 (cubic B-spline used in the approximation of the bias field); shrink factor—4 (defines the amount of upsampling applied to the input image prior to estimating the bias field); Wiener filter noise—0.00 mm (predetermined noise level for the filter); number of histogram bins—0 (default number of bins is typically determined automatically based on characteristics of the data).

2.4. Regional Brain Volume Measurements

After the preprocessing steps, all of the MRIs were segmented using MANTiS (version 1.1) to obtain the RBV measurements. MANTiS was selected because it was specifically developed to accurately segment infant brains, and has been validated in infant cohorts with and without brain lesions [52], being reported in several studies [13,14,15,45,46,50,59,60,61,62,63,64,65,66,67]. MANTiS is a toolbox for automated tissue segmentation of the neonatal brain from T2w MRI scans, into eight tissue classes: cortical grey matter (CGM); white matter (WM); cerebral spinal fluid (CSF); deep nuclear grey matter (DNGM); hippocampus (Hip); amygdala (Amy); cerebellum (CB); and brainstem (BS). It uses a combination of unified segmentation, a neonatal brain atlas, morphological segmentation, and filtering techniques to better adapt to neonatal anatomy [52]. The unified segmentation algorithm [55] combines registration, segmentation, and normalization into a single generative model. The tissue probability maps used are deformed during the registration process to better fit the image being segmented. MANTiS uses a neonatal brain atlas to provide a priori anatomical information to guide tissue segmentation. The atlas contains probabilistic maps of different brain tissue types in a reference space, which helps the algorithm classify the tissues in the MRI scans. A 40-week gestation average intensity template (created from 19 neonatal subjects) was used, and the corresponding tissue probability maps, with the correct sizes and shapes of the structures, were based on the probabilistic atlas created by Murgasova et al. [68]. The original template was modified to separate the subcortical grey matter into deep nuclear grey matter, hippocampus, and amygdala. However, these approaches alone may not provide adequate flexibility to accurately segment neonatal brain images, especially those exhibiting abnormalities or unusual anatomy (e.g., enlarged ventricles). So, in addition to the unified segmentation and atlas-based approaches, MANTiS also employs morphological image processing techniques (e.g., segmentation via morphological watershed transform from markers and filtering via reconstruction by dilation) to further refine the segmentation results [52]. MANTiS was implemented as a toolbox for SPM12 [56] in MATLAB® R2023a [57] and required compilation of MATLAB® code using the SPM infrastructure and C++ source code from ITK [38].

2.5. Software

The software used to perform this study were: 3D Slicer (version 5.2.2) [58]; Docker (version 4.18.0) [69]; FSL (version 6.0.6.4) [70]; MATLAB® R2023a (version 9.14) [57]; SPM12 (revision 7771) [56]; and IBM® SPSS® v29.0.0 [71].

2.6. Evaluation Strategies and Statistical Analysis

2.6.1. Evaluation Strategies

Several evaluation methods can be used in the field of intensity inhomogeneity correction, and can be categorized as qualitative and quantitative assessments [21]. This study was focused on quantitative assessment, because subjective visual examination of the BFC results (qualitative assessment) could be difficult to interpret if the inhomogeneity were mild, therefore not enabling a proper comparison of the different BFC methods. Among the three methodologies of quantitative assessment (inhomogeneity field, intensity variation, and segmentation based) [21], the latter two were selected, as there was no information regarding the true inhomogeneity field, and the quality of the nonuniformity correction would be derived indirectly. The measures based on intensity variability considered were: the coefficient of variation (CV), as the ratio of standard deviation ( σ ) and mean ( μ ) intensity for WM (Equation (1)) and GM (Equation (2)); and the coefficient of joint variation (CJV) (Equation (3)), by summing the σ of the intensities of two tissue classes (WM and GM), normalized by the difference between the μ intensities for such tissues:
C V W M = σ W M μ W M
C V G M = σ G M μ G M
C J V W M , G M = σ W M + σ G M μ W M − μ G M
Considering the mean intensity of each tissue to be relatively constant, its standard deviation should be reduced when intensity inhomogeneity is corrected, translated as a decrease in CV and CJV. As the segmentation-based measure, the Dice coefficient (DC) (Equation (4)) was considered to assess the BFC via the results from brain tissue segmentation maps (considering M as the reference map obtained from manual BE, and A the maps obtained from automated BE methods), defined as:
D C ( M , A ) = 2 M ⋂ A M + A
Therefore, the DC describes the similarity level between the segmentation results, and it ranges from 0 (no overlap) to 1 (perfect match).

2.6.2. Statistical Analysis

The assumption of normality of the distributions of measures was tested using the Shapiro–Wilk test [72]. Therefore, the Wilcoxon signed-rank test [72] was used to compare the RBVs obtained from the dataset bias field corrected with SPM-BFC and N4ITK. To test for differences between the RBVs obtained from brain MRIs extracted previously with the five methods (manual, BET2, SWS, HD-BET, and SynthStrip), Friedman’s two-way analysis of variance by ranks test was used with Bonferroni adjustment for multiple comparisons [72].
All statistical analyses were performed using a significance value of 5%, resorting to the IBM® SPSS® v29.0.0 software [71].

3. Results

3.1. Assessment of Intensity Variability and Segmentation Performance

Exemplary results obtained before and after application of BFC methods on T2w images from the dataset are shown in Figure 1. Visually, it is possible to identify variations in intensity (Figure 1c,d) after different bias corrections, but is hard to visually perceive changes after the respective brain tissue segmentations (Figure 2), as these are quite discreet.
Quantitative assessments regarding the intensity variability of brain tissues are shown in Table 1 and Figure 3, and regarding segmentation performance in Figure 4.
From Table 1 and Figure 3, it is observed that the CV (WM), CV (GM), and CJV (WM,GM) were reduced after BFC using N4ITK, but not using SPM-BFC, compared with the uncorrected images. The decrease in the CJV values after BFC using N4ITK indicates the improved separation between different region intensities, more specifically when HD-BET (0.662 ± 0.095) was used as the BE method. Additionally, the segmentation performance was assessed, and according to Figure 4, the highest similarity levels between the segmentation maps were obtained when using SWS (DC of 0.940 ± 0.060) or HD-BET (DC of 0.942 ± 0.063) as BE methods, and the lowest similarity levels when using BET2 (DC of 0.779 ± 0.144) or SynthStrip (DC of 0.713 ± 0.056), with a similar behavior for all uncorrected or bias field-corrected images.
Quantitative assessments showed differences between BFC methods regarding intensity variability (Figure 3), but those were not as evident on the segmentation performance (Figure 4). As such, the impact of BFC for each RBV was assessed.

3.2. Assessment of Regional Brain Volume Measurements After Different Preprocessing Steps

The RBVs obtained with different BFC and BE methods are shown in Table 2. The use of the SPM-BFC algorithm results in slightly higher volumes than when using the N4ITK algorithm for all brain tissues, except for CGM.

3.2.1. Comparison Between Bias Field Correction Methods

To assess if there were statistically significant differences between RBVs when employing different BFC methods (SPM-BFC and N4ITK), the Wilcoxon signed-rank test was applied to the volumes obtained with each BE method (Table 3).
Table 3 showed that the use of different BFC methods results in statistically significant changes in RBVs, for several BE methods: CGM (all); WM (all, except BET2 with p = 0.065); CSF (all); DNGM (BET2, p = 0.027); Hip (BET2, p = 0.019); Amy (Manual, p = 0.013; BET2, p = 0.035); CB (none); and BS (all).

3.2.2. Comparison Amongst Brain Extraction Methods

Initially, the differences between the BE methods with both BFC methods were assessed (Table 4), and significant differences in RBVs were identified among the methods, except for the hippocampus (χ2(4) = 7.116, p = 0.130) with N4ITK correction, and for the hippocampus (χ2(4) = 8.160, p = 0.086) and amygdala (χ2(4) = 8.442, p = 0.077) with SPM-BFC. Post hoc tests were applied to statistically assess the differences obtained in between each BE method (Table 5).
Comparing the five BE methods regarding the RBVs obtained, statistically significant differences were observed between them when using SPM-BFC (Table 5): CGM—differences between Manual, BET2, and SWS methods with SynthStrip, and Manual and SWS compared with HD-BET; WM—differences between SWS, HD-BET, and SynthStrip methods with BET2; CSF—differences between Manual, BET2, SWS, and HD-BET methods with SynthStrip, and BET2 and SWS methods compared with HD-BET; DNGM—differences between BET2, HD-BET, and SynthStrip methods with SWS; CB—differences between Manual, BET2, HD-BET, and SynthStrip methods with SWS; BS—differences between HD-BET and SynthStrip methods with SWS, and Manual and BET2 methods compared with HD-BET. On the other hand, applying the N4ITK correction showed statistically significant differences between the five BE methods, regarding the following RBVs (Table 5): CGM—differences between Manual, BET2, and SWS methods with SynthStrip; WM—differences between Manual, SWS, HD-BET, and SynthStrip methods with BET2; CSF—differences between Manual, BET2, SWS, and HD-BET methods with SynthStrip; DNGM—differences between BET2 and SynthStrip methods with SWS; CB—differences between Manual, BET2, HD-BET, and SynthStrip methods with SWS, and BET2 method compared with HD-BET; BS—differences between HD-BET and SynthStrip methods with SWS, and Manual method compared with HD-BET. Most statistically significant results showed a large effect size (r > 0.5) [72], indicating a substantial difference between groups.
Figure 5 provides a visual perspective of the regional segmentation differences when using the five studied BE methods. It is perceptible how they differ between them in the edges, exceeding the brain area (e.g., BET2) or the opposite (e.g., SynthStrip), compromising the regional segmentation of the remaining tissues differently.
Considering the results from DC (Figure 4), the most suitable automated BE methods were SWS and HD-BET, with the following differences (Table 5) between them: (1) manual BE and SWS were significantly different regarding the CB (p < 0.001), independently of the BFC method used; (2) manual BE and HD-BET showed differences regarding the CGM (p = 0.017) and BS (p = 0.003), with SPM-BFC, and differences regarding the BS volume (p = 0.002), with N4ITK correction. SWS and HD-BET differ from each other on the volumes obtained from CGM (p = 0.014), CSF (p = 0.010), DNGM (p = 0.003), CB, and BS (p < 0.001), after applying SPM-BFC (Table 5). However, when correcting with N4ITK, they did not differ from each other on the cerebrum tissue volumes, only on CB and BS (p < 0.001) (Table 5).

4. Discussion

The present study evaluated the effect of different preprocessing steps to determine how they affect the quantification of RBVs in neonatal brain MRI, since the published studies [13,14,43,44,45,46,47,48,49,50] were typically focused on the tissue segmentation methods, with less emphasis on the preceding BE or BFC methods. The effectiveness of the preprocessing steps for skull-stripping and removal of the bias field inhomogeneities in TEA brain T2w MRI was validated quantitatively through the analysis of intensity variability (CV and CJV), segmentation performance (DC), and volumetric (RBV) results for eight brain tissues (CGM, WM, CSF, DNGM, Hip, Amy, CB, and BS).
The higher reduction in CV and CJV values in MRI bias-corrected with N4ITK indicates reduced intensity variability (separate and joint) in both WM and GM, and lower overlap between their intensity distributions, suggesting N4ITK as a more efficient method for neonatal brain MRI BFC than SPM-BFC. As such, the performance of SPM-BFC might be more sensitive to its regularization and bias field smoothing parameters, and default settings can be suboptimal, leading to inadequate correction, potentially increasing the CV and CJV. Other potential reasons why N4ITK might perform better is due to its independence from segmentation and iterative approach, which may lead to more accurate corrections. It was evident for all BE methods, but particularly for skull-stripping with HD-BET, explicitly: CV (WM)—uncorrected (5.6%), SPM-BFC (5.8%), and N4ITK (4.7%); CV (GM)—uncorrected (7.5%), SPM-BFC (7.8%), and N4ITK (7.0%); CJV (WM,GM)—uncorrected (70.0%), SPM-BFC (71.8%), and N4ITK (66.2%). Previous studies in adults and in simulations, using older versions of the BFC methods selected, showed similar patterns of variation. For instance, Gispert et al. [73] reported CV (WM)—uncorrected (6.5%), SPM2 (7.5%), and N3 (5.9%); CV (GM)—uncorrected (16.7%), SPM2 (16.0%), and N3 (16.5%); CJV (WM,GM)—uncorrected (83.9%), SPM2 (88.7%), and N3 (83.9%). Average CV for random noise and bias field amplitude levels with SPM2 (3.0%) and N3 (2.5%) were described by Manjón et al. [74]. Chen et al. [75] reported CV (WM)—SPM (6.2%) and N3 (6.0%); and CV (GM)—SPM (10.8%) and N3 (10.7%). According to Xie et al. [76], studies using synthetic data resulted in CV (GM)—SPM8 (10.4%) and N3 (9.8%); CV (WM)—SPM8 (5.2%) and N3 (4.7%); CJV (WM,GM)—SPM8 (51.5%) and N3 (51.0%), and using real data resulted in CV (GM)—SPM8 (17.5%) and N3 (8.5%); CV (WM)—SPM8 (13.5%) and N3 (9.5%); CJV (WM,GM)—SPM8 (63.5%) and N3 (54.5%).
Regarding DC, the BFC methods behaved similarly between them, but showed a higher overlap index of the regional segmentation maps when using SWS (DC of 0.940) and HD-BET (DC of 0.942) as BE methods. In imaging studies of term infants (0–24 months) using several BE methods (BET (a), iBEAT (b), and U-net (c)), Bae et al. [77] showed the variation of average DC in brain masks (2D imaging: (a) 0.949, (b) 0.926, (a) 0.965; and 3D imaging: (a) 0.962, (b) 0.945, (a) 0.975). Additionally, in a previous study [10], the effect that BE methods had on intracranial volume measurements was tested, and the automated brain masks showed high similarity with manual segmentation (DC of 0.966 with SWS, and 0.968 with HD-BET), so a similar effect on subsequent volumetric tissue segmentation was expected.
Considering brain tissue quantification, there is an influence of the bias correction method (SPM-BFC or N4ITK) used, as shown by the significant differences mainly in the CGM, WM, CSF, and BS volumes. Assessing them separately for each BE, better results are observed when using N4ITK than SPM-BFC, i.e., less statistically significant differences between the RBVs obtained when applying the different BE methods. Concerning the influence of BE methods in neonatal brain MRI, SWS and HD-BET, when bias-corrected with N4ITK, only show differences on CB and BS volumes; if segmentation of the cerebrum’s tissue volumes is intended, then both BE methods can be used interchangeably without compromising the results. Regarding the other two automated BE methods (BET2 and SynthStrip), careful interpretation should be done of the obtained volumes, as results consider the total volume of the segmented region, without considering the spatial overlap of those regions separately, so having a comparable brain tissue volume (i.e., without statistically significant differences) may not be correlated with the similarity level between the tissue segmentation [10,78,79]. Consequently, the choice of the BE method impacts subsequent brain tissue segmentation, as demonstrated by small differences in Dice resulting in significant differences in the following processes for brain analysis.
The selection of the preprocessing methods (BE and BFC) has a direct impact on the volumetric measurements, potentially affecting clinical interpretations. The accuracy of RBV data complements the neuroimaging phenotype and might have an influence on clinical decision-making (e.g., detecting subtle brain abnormalities, assessing the effectiveness of intervention, optimizing brain growth, predicting neurodevelopmental outcomes) [7,8,9]. However, the prognostic value of the results remains limited because the reference values are not yet established, and also the partial volume effect associated with the neonatal MRI resolution may compromise the assessment of the small brain structures, emphasizing the need for continued research and refinement of the image processing methods.
In this work, the limitations are also related to the small sample size and the biased selection of subjects without major brain lesions. The reasons for this are twofold: (1) the evaluation of automated BE methods typically requires manual segmentation as a reference standard, and this process is highly time-consuming, which explains why many studies in the field have relied on small sample sizes (ranging from 5 to 22 subjects) [80,81,82,83]; and (2) it is recommended to initially test the impact of image processing methods on subjects without detectable brain lesions, congenital conditions, or severe motion artifacts [52,83,84,85]. These factors combined to constrain the size of our study sample, reflecting a common challenge in this area of research. The fact that the dataset was collected with one scanner at one center ensures consistency in data acquisition and image features, which is important when assessing the impact of different preprocessing methods. However, it might limit the generalizability of the results to other scanners, namely, in MRI scanners with different magnetic field strengths (e.g., higher field strengths have several advantages, such as higher spatial resolution, contrast and signal-to-noise ratio, but with a drawback of increasing the intensity inhomogeneities [86]). In this study we did not consider the ethnicity, race, or gender as a variable; however, some studies reported neonatal brain morphological variability [87,88,89] and others even defined population-specific templates [90] to be used. Despite the brain volume differences that might exist intrinsic to those variables, we are unsure how it affects the impact of the preprocessing methods. Finally, the application of BFC methods should be carefully interpreted in cases with motion artifacts and pathological lesions, since it was reported [91] in adults with significant white matter pathology that BFC might distort the real hyperintensities, although with the overall volumetric change not being significantly affected.
As future work, inclusion of diverse populations is suggested to assess the impact of demographic variables, as well as performing a multi-scanner validation, correcting for partial volume effects, testing new methods of BE and BFC, and applying methods with varying parameters of correction (e.g., BFC varying the bias FWHM). Recent approaches using deep convolutional neural network methods such as ABCnet (evaluated on neonatal, infant, and adult brain MRI datasets) [40], BIASNet (simulated and healthy young-adults subjects) [41], and DeepN4 (diverse datasets obtained from different scanners and protocols) [42] showed some improvement in DC and CV reduction compared to the BFC traditional methods. The results from neonatal data available in the ABCnet [40] study presented a CV (WM)—uncorrected (0.40%), N3 (0.56%), N4WI (0.38%), and ABCnet Original/N4WI (0.36%/0.33%); and CV (GM)—uncorrected (0.35%), N3 (0.51%), N4WI (0.33%), and ABCnet Original/N4WI (0.32%/0.30%). Additionally, the latest proposed neonatal MRI BE model, named ANUBEX [92], showed an average DC of 0.955, slightly higher than our tested methods. These types of methods are considered the future of research, despite the current limitations of network implementation constraints due to the available neonatal MRI data, which should include validation (real ground-truth data) and variability (e.g., different brain development stages, anatomical variation and pathologies, MR protocols).

5. Conclusions

The findings from this study will help guide the selection of appropriate preprocessing techniques for accurate neonatal brain MRI analysis and volumetric measurements, which are important for understanding typical and atypical brain development in preterm infants. The performance of HD-BET (BE) and N4ITK (BFC), tested in a TEA brain T2w MRI dataset, showed better results than the other preprocessing methods regarding improvement of intensity homogeneity and reproducibility of regional brain volumes.

Author Contributions

Conceptualization, T.F.V. and H.A.F.; methodology, T.F.V.; software, T.F.V.; validation, H.A.F. and N.C.M.; formal analysis, T.F.V.; investigation, T.F.V.; resources, T.F.V., N.C.M., L.H.-W. and N.N.; data curation, T.F.V.; writing—original draft preparation, T.F.V.; writing—review and editing, T.F.V., H.A.F. and N.M.; visualization, T.F.V.; supervision, H.A.F., N.M. and N.C.M.; project administration, T.F.V., H.A.F. and N.M.; funding acquisition, T.F.V. All authors have read and agreed to the published version of the manuscript.

Funding

This research was supported by Fundação para a Ciência e a Tecnologia (FCT), under the IBEB Strategic Program UIDB/00645/2020 (https://doi.org/10.54499/UIDB/00645/2020), and Bolsa de Investigação para Doutoramento (2022.12483.BD).

Institutional Review Board Statement

This study was conducted in accordance with the Declaration of Helsinki, and approved by the Human Research Ethical Committee of the Medical Faculty at Uppsala University (protocol code 2014/236/1 and date of approval 24 February 2015, with one amendment 2014/236/2 dated as of 18 May 2016), and the data were included after written consent from the parents for the acquisition of images and retrospective data analysis.

Informed Consent Statement

Informed parental consent was obtained from all subjects involved in the study.

Data Availability Statement

The original dataset is unavailable due to privacy and ethical restrictions; however, the data analyzed and generated are available in the main article and appendices. Further details are available on request from the corresponding author.

Conflicts of Interest

The authors declare no conflicts of interest.

Appendix A

To define the preprocessing order to be applied in this study, the similarity level (Dice coefficient) was assessed using the segmentation results from 10 subjects (i.e., brain masks obtained using the different automatic brain extraction methods vs the “ground truth” obtained using manual segmentation), before and after BFC using the N4ITK algorithm (Table A1). Figure A1 shows that the bias field map used for correction depends on the order of preprocessing steps; nonetheless, quantitatively (Dice coefficient), the differences were not significant.
Table A1. Statistics of the Dice coefficients and differences between brain masks (automated vs. manual) segmented before and after bias field correction (BFC), and assessment of statistical significance with the Wilcoxon signed ranks test.
Table A1. Statistics of the Dice coefficients and differences between brain masks (automated vs. manual) segmented before and after bias field correction (BFC), and assessment of statistical significance with the Wilcoxon signed ranks test.
Dice Coefficient DifferenceWilcoxon Test
Brain Extraction MethodMeanStandard DeviationMean95% Confidence IntervalZp
Before BFCAfter BFCBefore BFCAfter BFC
BET20.9380.9340.0040.0040.004[0.002; 0.007]−1.6040.109
SWS0.9680.9680.0010.0030.001[−0.001; 0.002]−1.0000.317
HD-BET0.9680.9660.0020.0070.002[−0.004; 0.007]−1.3420.180
SynthStrip0.9530.953−0.0010.007−0.001[−0.002; 0.001]−1.0000.317
Applsci 14 11575 g0a1
Figure A1. Recovered bias field map overlapped with brain MRI in three different axial slices, with N4ITK bias field correction (BFC) applied before and after manual brain extraction (BE) (bias field map displayed in HSV color map to ease the comparison between corrections).
Figure A1. Recovered bias field map overlapped with brain MRI in three different axial slices, with N4ITK bias field correction (BFC) applied before and after manual brain extraction (BE) (bias field map displayed in HSV color map to ease the comparison between corrections).
Applsci 14 11575 g0a1

References

  1. Ohuma, E.O.; Moller, A.-B.; Bradley, E.; Chakwera, S.; Hussain-Alkhateeb, L.; Lewin, A.; Okwaraji, Y.B.; Mahanani, W.R.; Johansson, E.W.; Lavin, T.; et al. National, Regional, and Global Estimates of Preterm Birth in 2020, with Trends from 2010: A Systematic Analysis. Lancet 2023, 402, 1261–1271. [Google Scholar] [CrossRef] [PubMed]
  2. Barkovich, M.J.; Barkovich, J. Normal Development of the Fetal, Neonatal, and Infant Brain, Skull, and Spine. In Pediatric Neuroimaging; Wolters Kluwer Health: Philadelphia, PA, USA, 2019; pp. 47–119. ISBN 978-1-4963-3722-1. [Google Scholar]
  3. Schwartz, E.S.; Barkovich, A.J. Brain and Spine Injuries in Infancy and Childhood. In Pediatric Neuroimaging; Wolters Kluwer Health: Philadelphia, PA, USA, 2019; pp. 371–533. ISBN 978-1-4963-3722-1. [Google Scholar]
  4. Weindling, M. Clinical Aspects of Brain Injury in the Preterm Infant. In The Newborn Brain; Lagercrantz, H., Hanson, M.A., Ment, L.R., Peebles, D.M., Eds.; Cambridge University Press: Cambridge, UK, 2010; pp. 301–320. ISBN 978-0-511-71184-8. [Google Scholar]
  5. Woodward, L.J.; Anderson, P.J.; Austin, N.C.; Howard, K.; Inder, T.E. Neonatal MRI to Predict Neurodevelopmental Outcomes in Preterm Infants. N. Engl. J. Med. 2006, 355, 685–694. [Google Scholar] [CrossRef] [PubMed]
  6. Rutherford, M.A. Imaging the Neonatal Brain. In The Newborn Brain; Lagercrantz, H., Hanson, M.A., Ment, L.R., Peebles, D.M., Eds.; Cambridge University Press: Cambridge, UK, 2010; pp. 199–210. ISBN 978-0-511-71184-8. [Google Scholar]
  7. Gui, L.; Loukas, S.; Lazeyras, F.; Hüppi, P.S.; Meskaldji, D.E.; Borradori Tolsa, C. Longitudinal Study of Neonatal Brain Tissue Volumes in Preterm Infants and Their Ability to Predict Neurodevelopmental Outcome. NeuroImage 2019, 185, 728–741. [Google Scholar] [CrossRef] [PubMed]
  8. Dimitrova, R.; Arulkumaran, S.; Carney, O.; Chew, A.; Falconer, S.; Ciarrusta, J.; Wolfers, T.; Batalle, D.; Cordero-Grande, L.; Price, A.N.; et al. Phenotyping the Preterm Brain: Characterizing Individual Deviations From Normative Volumetric Development in Two Large Infant Cohorts. Cereb. Cortex 2021, 31, 3665–3677. [Google Scholar] [CrossRef]
  9. Romberg, J.; Wilke, M.; Allgaier, C.; Nägele, T.; Engel, C.; Poets, C.F.; Franz, A. MRI-Based Brain Volumes of Preterm Infants at Term: A Systematic Review and Meta-Analysis. Arch. Dis. Child.-Fetal Neonatal Ed. 2022, 107, 520–526. [Google Scholar] [CrossRef]
  10. Vaz, T.F.; Canto Moreira, N.; Hellström-Westas, L.; Naseh, N.; Matela, N.; Ferreira, H.A. Brain Extraction Methods in Neonatal Brain MRI and Their Effects on Intracranial Volumes. Appl. Sci. 2024, 14, 1339. [Google Scholar] [CrossRef]
  11. Volpe, J.J. Specialized Studies in the Neurological Evaluation. In Neurology of the Newborn; Saunders: Philadelphia, PA, USA, 2008; pp. 154–191. ISBN 978-1-4160-3995-2. [Google Scholar]
  12. Keunen, K.; Išgum, I.; van Kooij, B.J.M.; Anbeek, P.; van Haastert, I.C.; Koopman-Esseboom, C.; Fieret-van Stam, P.C.; Nievelstein, R.A.J.; Viergever, M.A.; de Vries, L.S.; et al. Brain Volumes at Term-Equivalent Age in Preterm Infants: Imaging Biomarkers for Neurodevelopmental Outcome through Early School Age. J. Pediatr. 2016, 172, 88–95. [Google Scholar] [CrossRef]
  13. GilchristKelly, C.P.; Thompson, D.K.; Alexander, B.; Kelly, C.E.; Treyvaud, K.; Matthews, L.G.; Pascoe, L.; Zannino, D.; Yates, R.; Adamson, C.; et al. Growth of Prefrontal and Limbic Brain Regions and Anxiety Disorders in Children Born Very Preterm. Psychol. Med. 2023, 53, 759–770. [Google Scholar] [CrossRef]
  14. Collins, S.E.; Thompson, D.K.; Kelly, C.E.; Gilchrist, C.P.; Matthews, L.G.; Pascoe, L.; Lee, K.J.; Inder, T.E.; Doyle, L.W.; Cheong, J.L.Y.; et al. Development of Regional Brain Gray Matter Volume across the First 13 Years of Life Is Associated with Childhood Math Computation Ability for Children Born Very Preterm and Full Term. Brain Cogn. 2022, 160, 105875. [Google Scholar] [CrossRef]
  15. Strahle, J.M.; Triplett, R.L.; Alexopoulos, D.; Smyser, T.A.; Rogers, C.E.; Limbrick, D.D.; Smyser, C.D. Impaired Hippocampal Development and Outcomes in Very Preterm Infants with Perinatal Brain Injury. NeuroImage Clin. 2019, 22, 101787. [Google Scholar] [CrossRef]
  16. Li, G.; Wang, L.; Yap, P.-T.; Wang, F.; Wu, Z.; Meng, Y.; Dong, P.; Kim, J.; Shi, F.; Rekik, I.; et al. Computational Neuroanatomy of Baby Brains: A Review. NeuroImage 2019, 185, 906–925. [Google Scholar] [CrossRef] [PubMed]
  17. Devi, C.N.; Chandrasekharan, A.; Sundararaman, V.K.; Alex, Z.C. Neonatal Brain MRI Segmentation: A Review. Comput. Biol. Med. 2015, 64, 163–178. [Google Scholar] [CrossRef]
  18. Kalavathi, P.; Prasath, V.B.S. Methods on Skull Stripping of MRI Head Scan Images—A Review. J. Digit. Imaging 2016, 29, 365–379. [Google Scholar] [CrossRef]
  19. Fatima, A.; Shahid, A.R.; Raza, B.; Madni, T.M.; Janjua, U.I. State-of-the-Art Traditional to the Machine- and Deep-Learning-Based Skull Stripping Techniques, Models, and Algorithms. J. Digit. Imaging 2020, 33, 1443–1464. [Google Scholar] [CrossRef]
  20. Makropoulos, A.; Counsell, S.J.; Rueckert, D. A Review on Automatic Fetal and Neonatal Brain MRI Segmentation. NeuroImage 2018, 170, 231–248. [Google Scholar] [CrossRef] [PubMed]
  21. Vovk, U.; Pernus, F.; Likar, B. A Review of Methods for Correction of Intensity Inhomogeneity in MRI. IEEE Trans. Med. Imaging 2007, 26, 405–421. [Google Scholar] [CrossRef] [PubMed]
  22. Condon, B.R.; Patterson, J.; Wyper, D.; Jenkins, A.; Hadley, D.M. Image Non-Uniformity in Magnetic Resonance Imaging: Its Magnitude and Methods for Its Correction. Br. J. Radiol. 1987, 60, 83–87. [Google Scholar] [CrossRef] [PubMed]
  23. Simmons, A.; Tofts, P.S.; Barker, G.J.; Arridge, S.R. Sources of Intensity Nonuniformity in Spin Echo Images at 1.5 T. Magn. Reson. Med. 1994, 32, 121–128. [Google Scholar] [CrossRef]
  24. Lai, S.-H.; Fang, M. A Dual Image Approach for Bias Field Correction in Magnetic Resonance Imaging. Magn. Reson. Imaging 2003, 21, 121–125. [Google Scholar] [CrossRef]
  25. Mihara, H.; Iriguchi, N.; Ueno, S. A Method of RF Inhomogeneity Correction in MR Imaging. Magn. Reson. Mater. Phys. Biol. Med. 1998, 7, 115–120. [Google Scholar] [CrossRef]
  26. Brinkmann, B.H.; Manduca, A.; Robb, R.A. Optimized Homomorphic Unsharp Masking for MR Grayscale Inhomogeneity Correction. IEEE Trans. Med. Imaging 1998, 17, 161–171. [Google Scholar] [CrossRef] [PubMed]
  27. Zhou, L.Q.; Zhu, Y.M.; Bergot, C.; Laval-Jeantet, A.-M.; Bousson, V.; Laredo, J.-D.; Laval-Jeantet, M. A Method of Radio-Frequency Inhomogeneity Correction for Brain Tissue Segmentation in MRI. Comput. Med. Imaging Graph. 2001, 25, 379–389. [Google Scholar] [CrossRef] [PubMed]
  28. Vemuri, P.; Kholmovski, E.G.; Parker, D.L.; Chapman, B.E. Coil Sensitivity Estimation for Optimal SNR Reconstruction and Intensity Inhomogeneity Correction in Phased Array MR Imaging. Inf. Process Med. Imaging 2005, 19, 603–614. [Google Scholar] [CrossRef]
  29. Pham, D.L.; Prince, J.L. Adaptive Fuzzy Segmentation of Magnetic Resonance Images. IEEE Trans. Med. Imaging 1999, 18, 737–752. [Google Scholar] [CrossRef]
  30. Ashburner, J.; Friston, K.J. Voxel-Based Morphometry—The Methods. NeuroImage 2000, 11, 805–821. [Google Scholar] [CrossRef]
  31. Zhang, Y.; Brady, M.; Smith, S. Segmentation of Brain MR Images through a Hidden Markov Random Field Model and the Expectation-Maximization Algorithm. IEEE Trans. Med. Imaging 2001, 20, 45–57. [Google Scholar] [CrossRef]
  32. Zhou, Y.; Bai, J. Atlas-Based Fuzzy Connectedness Segmentation and Intensity Nonuniformity Correction Applied to Brain MRI. IEEE Trans. Biomed. Eng. 2007, 54, 122–129. [Google Scholar] [CrossRef]
  33. Zhou, S.; Wang, J.; Zhang, S.; Liang, Y.; Gong, Y. Active Contour Model Based on Local and Global Intensity Information for Medical Image Segmentation. Neurocomputing 2016, 186, 107–118. [Google Scholar] [CrossRef]
  34. Sled, J.G.; Zijdenbos, A.P.; Evans, A.C. A Nonparametric Method for Automatic Correction of Intensity Nonuniformity in MRI Data. IEEE Trans. Med. Imaging 1998, 17, 87–97. [Google Scholar] [CrossRef]
  35. Styner, M.; Brechbühler, C.; Székely, G.; Gerig, G. Parametric Estimate of Intensity Inhomogeneities Applied to MRI. IEEE Trans. Med. Imaging 2000, 19, 153–165. [Google Scholar] [CrossRef]
  36. Likar, B.; Viergever, M.A.; Pernuš, F. Retrospective Correction of MR Intensity Inhomogeneity by Information Minimization. In Proceedings of the Medical Image Computing and Computer-Assisted Intervention—MICCAI 2000, Pittsburgh, PA, USA, 11–14 October 2000; Delp, S.L., DiGoia, A.M., Jaramaz, B., Eds.; Springer: Berlin/Heidelberg, Germany, 2000; pp. 375–384. [Google Scholar]
  37. Tustison, N.J.; Avants, B.B.; Cook, P.A.; Zheng, Y.; Egan, A.; Yushkevich, P.A.; Gee, J.C. N4ITK: Improved N3 Bias Correction. IEEE Trans. Med. Imaging 2010, 29, 1310–1320. [Google Scholar] [CrossRef] [PubMed]
  38. Ibáñez, L.; Schroeder, W.; Ng, L.; Cates, J. Insight Software Consortium. The ITK Software Guide, 2nd ed.; Kitware Inc.: Clifton Park, NY, USA, 2005; ISBN 1-930934-15-7. [Google Scholar]
  39. George, M.M.; Kalaivani, S.; Sudhakar, M.S. A Non-Iterative Multi-Scale Approach for Intensity Inhomogeneity Correction in MRI. Magn. Reson. Imaging 2017, 42, 43–59. [Google Scholar] [CrossRef] [PubMed]
  40. Chen, L.; Wu, Z.; Hu, D.; Wang, F.; Smith, J.K.; Lin, W.; Wang, L.; Shen, D.; Li, G.; UNC/UMN Baby Connectome Project Consortium. ABCnet: Adversarial Bias Correction Network for Infant Brain MR Images. Med. Image Anal. 2021, 72, 102133. [Google Scholar] [CrossRef] [PubMed]
  41. Xu, Y.; Wang, Y.; Hu, S.; Du, Y. Deep Convolutional Neural Networks for Bias Field Correction of Brain Magnetic Resonance Images. J. Supercomput. 2022, 78, 17943–17968. [Google Scholar] [CrossRef]
  42. Kanakaraj, P.; Yao, T.; Cai, L.Y.; Lee, H.H.; Newlin, N.R.; Kim, M.E.; Gao, C.; Pechman, K.R.; Archer, D.; Hohman, T.; et al. DeepN4: Learning N4ITK Bias Field Correction for T1-Weighted Images. Neuroinform 2024, 22, 193–205. [Google Scholar] [CrossRef]
  43. Makropoulos, A.; Gousias, I.S.; Ledig, C.; Aljabar, P.; Serag, A.; Hajnal, J.V.; Edwards, A.D.; Counsell, S.J.; Rueckert, D. Automatic Whole Brain MRI Segmentation of the Developing Neonatal Brain. IEEE Trans. Med. Imaging 2014, 33, 1818–1831. [Google Scholar] [CrossRef]
  44. Cardoso, M.J.; Melbourne, A.; Kendall, G.S.; Modat, M.; Robertson, N.J.; Marlow, N.; Ourselin, S. AdaPT: An Adaptive Preterm Segmentation Algorithm for Neonatal Brain MRI. NeuroImage 2013, 65, 97–108. [Google Scholar] [CrossRef]
  45. Mongerson, C.R.L.; Wilcox, S.L.; Goins, S.M.; Pier, D.B.; Zurakowski, D.; Jennings, R.W.; Bajic, D. Infant Brain Structural MRI Analysis in the Context of Thoracic Non-Cardiac Surgery and Critical Care. Front. Pediatr. 2019, 7, 315. [Google Scholar] [CrossRef]
  46. Rudisill, S.S.; Wang, J.T.; Jaimes, C.; Mongerson, C.R.L.; Hansen, A.R.; Jennings, R.W.; Bajic, D. Neurologic Injury and Brain Growth in the Setting of Long-Gap Esophageal Atresia Perioperative Critical Care: A Pilot Study. Brain Sci. 2019, 9, 383. [Google Scholar] [CrossRef]
  47. Adamson, C.L.; Alexander, B.; Kelly, C.E.; Ball, G.; Beare, R.; Cheong, J.L.Y.; Spittle, A.J.; Doyle, L.W.; Anderson, P.J.; Seal, M.L.; et al. Updates to the Melbourne Children’s Regional Infant Brain Software Package (M-CRIB-S). Neuroinformatics 2024, 22, 207–223. [Google Scholar] [CrossRef]
  48. Gui, L.; Lisowski, R.; Faundez, T.; Hüppi, P.S.; Lazeyras, F.; Kocher, M. Morphology-Driven Automatic Segmentation of MR Images of the Neonatal Brain. Med. Image Anal. 2012, 16, 1565–1579. [Google Scholar] [CrossRef] [PubMed]
  49. Moeskops, P.; Viergever, M.A.; Mendrik, A.M.; de Vries, L.S.; Benders, M.J.N.L.; Išgum, I. Automatic Segmentation of MR Brain Images With a Convolutional Neural Network. IEEE Trans. Med. Imaging 2016, 35, 1252–1261. [Google Scholar] [CrossRef] [PubMed]
  50. Thompson, D.K.; Kelly, C.E.; Chen, J.; Beare, R.; Alexander, B.; Seal, M.L.; Lee, K.; Matthews, L.G.; Anderson, P.J.; Doyle, L.W.; et al. Early Life Predictors of Brain Development at Term-Equivalent Age in Infants Born across the Gestational Age Spectrum. NeuroImage 2019, 185, 813–824. [Google Scholar] [CrossRef]
  51. Jenkinson, M.; Pechaud, M.; Smith, S. BET2: MR-Based Estimation of Brain, Skull and Scalp Surfaces. In Proceedings of the Eleventh Annual Meeting of the Organization for Human Brain Mapping, Toronto, ON, Canada, 12–16 June 2005. [Google Scholar]
  52. Beare, R.J.; Chen, J.; Kelly, C.E.; Alexopoulos, D.; Smyser, C.D.; Rogers, C.E.; Loh, W.Y.; Matthews, L.G.; Cheong, J.L.Y.; Spittle, A.J.; et al. Neonatal Brain Tissue Classification with Morphological Adaptation and Unified Segmentation. Front. Neuroinform. 2016, 10, 12. [Google Scholar] [CrossRef]
  53. Isensee, F.; Schell, M.; Pflueger, I.; Brugnara, G.; Bonekamp, D.; Neuberger, U.; Wick, A.; Schlemmer, H.-P.; Heiland, S.; Wick, W.; et al. Automated Brain Extraction of Multisequence MRI Using Artificial Neural Networks. Hum. Brain Mapp. 2019, 40, 4952–4964. [Google Scholar] [CrossRef]
  54. Hoopes, A.; Mora, J.S.; Dalca, A.V.; Fischl, B.; Hoffmann, M. SynthStrip: Skull-Stripping for Any Brain Image. NeuroImage 2022, 260, 119474. [Google Scholar] [CrossRef]
  55. Ashburner, J.; Friston, K.J. Unified Segmentation. NeuroImage 2005, 26, 839–851. [Google Scholar] [CrossRef] [PubMed]
  56. Penny, W.D.; Friston, K.J.; Ashburner, J.T.; Kiebel, S.J.; Nichols, T.E. Statistical Parametric Mapping: The Analysis of Functional Brain Images, 1st ed.; Academic Press: Cambridge, MA, USA, 2006; ISBN 978-0-08-046650-7. [Google Scholar]
  57. MATLAB, Version 9.14.0 (R2023a); The MathWorks Inc.: Natick, MA, USA, 2023.
  58. Fedorov, A.; Beichel, R.; Kalpathy-Cramer, J.; Finet, J.; Fillion-Robin, J.-C.; Pujol, S.; Bauer, C.; Jennings, D.; Fennessy, F.; Sonka, M.; et al. 3D Slicer as an Image Computing Platform for the Quantitative Imaging Network. Magn. Reson. Imaging 2012, 30, 1323–1341. [Google Scholar] [CrossRef]
  59. Alexander, B.; Kelly, C.E.; Adamson, C.; Beare, R.; Zannino, D.; Chen, J.; Murray, A.L.; Loh, W.Y.; Matthews, L.G.; Warfield, S.K.; et al. Changes in Neonatal Regional Brain Volume Associated with Preterm Birth and Perinatal Factors. NeuroImage 2019, 185, 654–663. [Google Scholar] [CrossRef]
  60. Thompson, D.K.; Kelly, C.E.; Chen, J.; Beare, R.; Alexander, B.; Seal, M.L.; Lee, K.J.; Matthews, L.G.; Anderson, P.J.; Doyle, L.W.; et al. Characterisation of Brain Volume and Microstructure at Term-Equivalent Age in Infants Born across the Gestational Age Spectrum. NeuroImage Clin. 2019, 21, 101630. [Google Scholar] [CrossRef]
  61. Alexander, B.; Loh, W.Y.; Matthews, L.G.; Murray, A.L.; Adamson, C.; Beare, R.; Chen, J.; Kelly, C.E.; Anderson, P.J.; Doyle, L.W.; et al. Desikan-Killiany-Tourville Atlas Compatible Version of M-CRIB Neonatal Parcellated Whole Brain Atlas: The M-CRIB 2.0. Front. Neurosci. 2019, 13, 34. [Google Scholar] [CrossRef] [PubMed]
  62. Kelly, C.E.; Thompson, D.K.; Spittle, A.J.; Chen, J.; Seal, M.L.; Anderson, P.J.; Doyle, L.W.; Cheong, J.L.Y. Regional Brain Volumes, Microstructure and Neurodevelopment in Moderate-Late Preterm Children. Arch. Dis. Child. Fetal Neonatal Ed. 2020, 105, 593–599. [Google Scholar] [CrossRef] [PubMed]
  63. Treyvaud, K.; Thompson, D.; Kelly, C.; Loh, W.; Inder, T.; Cheong, J.; Doyle, L.; Anderson, P. Early Parenting Is Associated with the Developing Brains of Children Born Very Preterm. Clin. Neuropsychol. 2021, 35, 885–903. [Google Scholar] [CrossRef] [PubMed]
  64. Monson, B.B.; Anderson, P.J.; Matthews, L.G.; Neil, J.J.; Kapur, K.; Cheong, J.L.Y.; Doyle, L.W.; Thompson, D.K.; Inder, T.E. Examination of the Pattern of Growth of Cerebral Tissue Volumes From Hospital Discharge to Early Childhood in Very Preterm Infants. JAMA Pediatr. 2016, 170, 772–779. [Google Scholar] [CrossRef]
  65. Granger, C.; Spittle, A.J.; Walsh, J.; Pyman, J.; Anderson, P.J.; Thompson, D.K.; Lee, K.J.; Coleman, L.; Dagia, C.; Doyle, L.W.; et al. Histologic Chorioamnionitis in Preterm Infants: Correlation with Brain Magnetic Resonance Imaging at Term Equivalent Age. BMC Pediatr. 2018, 18, 63. [Google Scholar] [CrossRef] [PubMed]
  66. Matthews, L.G.; Smyser, C.D.; Cherkerzian, S.; Alexopoulos, D.; Kenley, J.; Tuuli, M.G.; Michael Nelson, D.; Inder, T.E. Maternal Pomegranate Juice Intake and Brain Structure and Function in Infants with Intrauterine Growth Restriction: A Randomized Controlled Pilot Study. PLoS ONE 2019, 14, e0219596. [Google Scholar] [CrossRef]
  67. Vanderhasselt, T.; Naeyaert, M.; Watté, N.; Allemeersch, G.-J.; Raeymaeckers, S.; Dudink, J.; de Mey, J.; Raeymaekers, H. Synthetic MRI of Preterm Infants at Term-Equivalent Age: Evaluation of Diagnostic Image Quality and Automated Brain Volume Segmentation. AJNR Am. J. Neuroradiol. 2020, 41, 882–888. [Google Scholar] [CrossRef]
  68. Kuklisova-Murgasova, M.; Aljabar, P.; Srinivasan, L.; Counsell, S.J.; Doria, V.; Serag, A.; Gousias, I.S.; Boardman, J.P.; Rutherford, M.A.; Edwards, A.D.; et al. A Dynamic 4D Probabilistic Atlas of the Developing Brain. NeuroImage 2011, 54, 2750–2763. [Google Scholar] [CrossRef]
  69. Merkel, D. Docker: Lightweight Linux Containers for Consistent Development and Deployment. Linux J. 2014, 239, 2. [Google Scholar]
  70. Jenkinson, M.; Beckmann, C.F.; Behrens, T.E.J.; Woolrich, M.W.; Smith, S.M. FSL. NeuroImage 2012, 62, 782–790. [Google Scholar] [CrossRef]
  71. IBM Corp. Released 2022. IBM SPSS Statistics for Windows, Version 29.0.0; IBM Corp: Armonk, NY, USA, 2022.
  72. Portney, L.G. Foundations of Clinical Research: Applications to Evidence-Based Practice, 4th ed.; F.A. Davis: Philadelphia, PA, USA, 2020; ISBN 978-0-8036-6113-4. [Google Scholar]
  73. Gispert, J.D.; Reig, S.; Pascau, J.; Vaquero, J.J.; García-Barreno, P.; Desco, M. Method for Bias Field Correction of Brain T1-weighted Magnetic Resonance Images Minimizing Segmentation Error. Hum. Brain Mapp. 2004, 22, 133–144. [Google Scholar] [CrossRef] [PubMed]
  74. Manjón, J.V.; Lull, J.J.; Carbonell-Caballero, J.; García-Martí, G.; Martí-Bonmatí, L.; Robles, M. A Nonparametric MRI Inhomogeneity Correction Method. Med. Image Anal. 2007, 11, 336–345. [Google Scholar] [CrossRef]
  75. Chen, Y.; Zhao, B.; Zhang, J.; Zheng, Y. Automatic Segmentation for Brain MR Images via a Convex Optimized Segmentation and Bias Field Correction Coupled Model. Magn. Reson. Imaging 2014, 32, 941–955. [Google Scholar] [CrossRef]
  76. Xie, M.; Gao, J.; Zhu, C.; Zhou, Y. A Modified Method for MRF Segmentation and Bias Correction of MR Image with Intensity Inhomogeneity. Med. Biol. Eng. Comput. 2015, 53, 23–35. [Google Scholar] [CrossRef]
  77. Bae, I.; Chae, J.-H.; Han, Y. A Brain Extraction Algorithm for Infant T2 Weighted Magnetic Resonance Images Based on Fuzzy C-Means Thresholding. Sci. Rep. 2021, 11, 23347. [Google Scholar] [CrossRef]
  78. Taha, A.A.; Hanbury, A. Metrics for Evaluating 3D Medical Image Segmentation: Analysis, Selection, and Tool. BMC Med. Imaging 2015, 15, 29. [Google Scholar] [CrossRef]
  79. Weiss, D.A.; Saluja, R.; Xie, L.; Gee, J.C.; Sugrue, L.P.; Pradhan, A.; Nick Bryan, R.; Rauschecker, A.M.; Rudie, J.D. Automated Multiclass Tissue Segmentation of Clinical Brain MRIs with Lesions. NeuroImage Clin. 2021, 31, 102769. [Google Scholar] [CrossRef] [PubMed]
  80. Mahapatra, D. Skull Stripping of Neonatal Brain MRI: Using Prior Shape Information with Graph Cuts. J. Digit. Imaging 2012, 25, 802–814. [Google Scholar] [CrossRef] [PubMed]
  81. Péporté, M.; Ilea Ghita, D.E.; Twomey, E.; Whelan, P.F. A Hybrid Approach to Brain Extraction from Premature Infant MRI. In Image Analysis. SCIA 2011; Lecture Notes in Computer Science; Heyden, A., Kahl, F., Eds.; Springer: Berlin/Heidelberg, Germany, 2011; Volume 6688, pp. 719–730. [Google Scholar]
  82. Wang, L.; Wu, Z.; Chen, L.; Sun, Y.; Lin, W.; Li, G. iBEAT V2.0: A Multisite-Applicable, Deep Learning-Based Pipeline for Infant Cerebral Cortical Surface Reconstruction. Nat. Protoc. 2023, 18, 1488–1509. [Google Scholar] [CrossRef]
  83. Serag, A.; Blesa, M.; Moore, E.J.; Pataky, R.; Sparrow, S.A.; Wilkinson, A.G.; Macnaught, G.; Semple, S.I.; Boardman, J.P. Accurate Learning with Few Atlases (ALFA): An Algorithm for MRI Neonatal Brain Extraction and Comparison with 11 Publicly Available Methods. Sci. Rep. 2016, 6, 23470. [Google Scholar] [CrossRef]
  84. Shi, F.; Wang, L.; Dai, Y.; Gilmore, J.H.; Lin, W.; Shen, D. LABEL: Pediatric Brain Extraction Using Learning-Based Meta-Algorithm. NeuroImage 2012, 62, 1975–1986. [Google Scholar] [CrossRef] [PubMed]
  85. Gousias, I.S.; Edwards, A.D.; Rutherford, M.A.; Counsell, S.J.; Hajnal, J.V.; Rueckert, D.; Hammers, A. Magnetic Resonance Imaging of the Newborn Brain: Manual Segmentation of Labelled Atlases in Term-Born and Preterm Infants. NeuroImage 2012, 62, 1499–1509. [Google Scholar] [CrossRef] [PubMed]
  86. Moser, E.; Stahlberg, F.; Ladd, M.E.; Trattnig, S. 7-T MR--from Research to Clinical Applications? NMR Biomed. 2012, 25, 695–716. [Google Scholar] [CrossRef] [PubMed]
  87. Bai, J.; Abdul-Rahman, M.F.; Rifkin-Graboi, A.; Chong, Y.-S.; Kwek, K.; Saw, S.-M.; Godfrey, K.M.; Gluckman, P.D.; Fortier, M.V.; Meaney, M.J.; et al. Population Differences in Brain Morphology and Microstructure among Chinese, Malay, and Indian Neonates. PLoS ONE 2012, 7, e47816. [Google Scholar] [CrossRef]
  88. Benavides, A.; Metzger, A.; Tereschenko, S.; Conrad, A.; Bell, E.F.; Spencer, J.; Ross-Sheehy, S.; Georgieff, M.; Magnotta, V.; Nopoulos, P. Sex Specific Alterations in Preterm Brain. Pediatr. Res. 2019, 85, 55–62. [Google Scholar] [CrossRef]
  89. Khan, Y.T.; Tsompanidis, A.; Radecki, M.A.; Dorfschmidt, L.; Adhya, D.; Ayeung, B.; Bamford, R.; Biron-Shental, T.; Burton, G.; Cowell, W.; et al. Sex Differences in Human Brain Structure at Birth. Biol. Sex Differ. 2024, 15, 81. [Google Scholar] [CrossRef]
  90. Geng, X.; Chan, P.H.; Lam, H.S.; Chu, W.C.; Wong, P.C. Brain Templates for Chinese Babies from Newborn to Three Months of Age. NeuroImage 2024, 289, 120536. [Google Scholar] [CrossRef]
  91. Valdés Hernández, M.d.C.; González-Castro, V.; Ghandour, D.T.; Wang, X.; Doubal, F.; Muñoz Maniega, S.; Armitage, P.A.; Wardlaw, J.M. On the Computational Assessment of White Matter Hyperintensity Progression: Difficulties in Method Selection and Bias Field Correction Performance on Images with Significant White Matter Pathology. Neuroradiology 2016, 58, 475–485. [Google Scholar] [CrossRef]
  92. Chen, J.V.; Li, Y.; Tang, F.; Chaudhari, G.; Lew, C.; Lee, A.; Rauschecker, A.M.; Haskell-Mendoza, A.P.; Wu, Y.W.; Calabrese, E. Automated Neonatal nnU-Net Brain MRI Extractor Trained on a Large Multi-Institutional Dataset. Sci. Rep. 2024, 14, 4583. [Google Scholar] [CrossRef]
Applsci 14 11575 g001
Figure 1. Intensity inhomogeneity on TEA brain T2w MRI preterm neonate: (a) skull-stripped MRI without correction; (b) bias field map; bias-corrected MRI with (c) N4ITK and (d) SPM-BFC methods (images displayed with the Jet color map to ease the comparison between corrections).
Figure 1. Intensity inhomogeneity on TEA brain T2w MRI preterm neonate: (a) skull-stripped MRI without correction; (b) bias field map; bias-corrected MRI with (c) N4ITK and (d) SPM-BFC methods (images displayed with the Jet color map to ease the comparison between corrections).
Applsci 14 11575 g001
Applsci 14 11575 g002
Figure 2. TEA brain T2w MRI (manually skull-stripped) of a preterm neonate in three different axial slices bias-corrected with SPM-BFC (a–c) and N4ITK (d–f), with overlaid brain tissues/structures segmented using MANTiS (color coded: CGM, red; WM, green; CSF, blue; DNGM, yellow; Hip, light blue; Amy, pink; CB, grey; BS, dark blue).
Figure 2. TEA brain T2w MRI (manually skull-stripped) of a preterm neonate in three different axial slices bias-corrected with SPM-BFC (a–c) and N4ITK (d–f), with overlaid brain tissues/structures segmented using MANTiS (color coded: CGM, red; WM, green; CSF, blue; DNGM, yellow; Hip, light blue; Amy, pink; CB, grey; BS, dark blue).
Applsci 14 11575 g002
Applsci 14 11575 g003
Figure 3. Coefficient of joint variation (CJV) between white and grey matter intensities from MRI uncorrected and bias field-corrected with SPM-BFC and N4ITK, for each brain extraction method (Manual, BET2, SWS, HD-BET, and SynthStrip).
Figure 3. Coefficient of joint variation (CJV) between white and grey matter intensities from MRI uncorrected and bias field-corrected with SPM-BFC and N4ITK, for each brain extraction method (Manual, BET2, SWS, HD-BET, and SynthStrip).
Applsci 14 11575 g003
Applsci 14 11575 g004
Figure 4. Dice coefficient (DC) based on the regional brain tissue segmentation maps obtained from MRI uncorrected and bias field-corrected using SPM-BFC or N4ITK, for each brain extraction method (Manual, BET2, SWS, HD-BET, and SynthStrip).
Figure 4. Dice coefficient (DC) based on the regional brain tissue segmentation maps obtained from MRI uncorrected and bias field-corrected using SPM-BFC or N4ITK, for each brain extraction method (Manual, BET2, SWS, HD-BET, and SynthStrip).
Applsci 14 11575 g004
Applsci 14 11575 g005
Figure 5. TEA brain T2w MRI (example bias-corrected with N4ITK) of a preterm neonate in three different axial slices (a) and corresponding skull-stripped images from each BE method ((b) Manual, (c) BET2, (d) SWS, (e) HD-BET, and (f) SynthStrip), overlaid with segmentation from MANTiS (color coded: CGM, red; WM, green; CSF, blue; DNGM, yellow; Hip, light blue; Amy, pink; CB, grey; BS, dark blue).
Figure 5. TEA brain T2w MRI (example bias-corrected with N4ITK) of a preterm neonate in three different axial slices (a) and corresponding skull-stripped images from each BE method ((b) Manual, (c) BET2, (d) SWS, (e) HD-BET, and (f) SynthStrip), overlaid with segmentation from MANTiS (color coded: CGM, red; WM, green; CSF, blue; DNGM, yellow; Hip, light blue; Amy, pink; CB, grey; BS, dark blue).
Applsci 14 11575 g005
Table 1. Coefficient of variation (CV) of WM and GM and coefficient of joint variation (CJV), for uncorrected and bias field-corrected images after applying different brain extraction methods.
Table 1. Coefficient of variation (CV) of WM and GM and coefficient of joint variation (CJV), for uncorrected and bias field-corrected images after applying different brain extraction methods.
Bias Field Correction Method
MeasureBrain Extraction MethodUncorrectedSPM-BFCN4ITK
CV (WM) Manual0.056 ± 0.0090.057 ± 0.0110.048 ± 0.006
BET20.056 ± 0.0090.058 ± 0.0110.048 ± 0.005
SWS0.055 ± 0.0090.057 ± 0.0110.047 ± 0.006
HD-BET0.056 ± 0.0090.058 ± 0.0110.047 ± 0.005
SynthStrip0.056 ± 0.0090.057 ± 0.0110.048 ± 0.005
CV (GM) Manual0.075 ± 0.0040.078 ± 0.0030.072 ± 0.004
BET20.075 ± 0.0030.078 ± 0.0020.073 ± 0.003
SWS0.076 ± 0.0040.078 ± 0.0020.073 ± 0.003
HD-BET0.075 ± 0.0040.078 ± 0.0020.070 ± 0.005
SynthStrip0.075 ± 0.0040.081 ± 0.0040.072 ± 0.003
CJV (WM,GM) Manual0.698 ± 0.1130.718 ± 0.1110.675 ± 0.078
BET20.700 ± 0.1090.717 ± 0.1080.670 ± 0.072
SWS0.699 ± 0.1110.720 ± 0.1090.676 ± 0.071
HD-BET0.700 ± 0.1150.718 ± 0.1120.662 ± 0.095
SynthStrip0.699 ± 0.1120.726 ± 0.1080.671 ± 0.069
Table 2. Descriptive statistics (mean ± standard deviation and median) of brain tissue volumes (mL) obtained after bias field correction with SPM-BFC and N4ITK, using different brain extraction methods (Manual, BET2, SWS, HD-BET, and SynthStrip).
Table 2. Descriptive statistics (mean ± standard deviation and median) of brain tissue volumes (mL) obtained after bias field correction with SPM-BFC and N4ITK, using different brain extraction methods (Manual, BET2, SWS, HD-BET, and SynthStrip).
Brain TissueBias Field Correction MethodBrain Extraction Method
ManualBET2SWSHD-BETSynthStrip
Cortical Grey
Matter		SPM-BFC175.5 ± 18.6
176.6		176.4 ± 18.6
176.9		175.5 ± 18.3
176.6		177.0 ± 18.3
178.0		180.2 ± 18.9
180.1
N4ITK178.0 ± 19.3
179.1		178.1 ± 18.8
179.3		177.5 ± 18.8
176.3		178.8 ± 18.7
178.5		182.1 ± 19.3
181.4
White
Matter		SPM-BFC131.7 ± 17.3
132.8		132.7 ± 16.5
134.5		131.0 ± 17.1
131.1		131.3 ± 16.6
132.6		131.1 ± 16.2
132.9
N4ITK130.2 ± 16.7
130.2		132.1 ± 16.6
132.5		129.4 ± 16.5
129.1		130.0 ± 16.1
130.1		130.3 ± 15.9
132.6
Cerebral Spinal FluidSPM-BFC67.6 ± 21.3
68.3		64.9 ± 20.4
63.1		64.9 ± 22.3
66.9		68.4 ± 20.7
68.8		58.4 ± 16.8
54.9
N4ITK66.3 ± 20.4
64.4		64.0 ± 20.2
61.4		64.2 ± 21.7
65.8		67.7 ± 20.0
68.4		57.2 ± 16.1
53.0
Deep Nuclear
Grey Matter		SPM-BFC26.0 ± 2.5
26.6		25.8 ± 2.4
26.6		26.2 ± 2.5
26.9		25.9 ± 2.5
26.5		25.8 ± 2.4
26.5
N4ITK25.9 ± 2.4
26.4		25.7 ± 2.4
26.1		26.1 ± 2.5
26.5		25.8 ± 2.4
26.2		25.7 ± 2.4
26.2
HippocampusSPM-BFC3.0 ± 0.5
3.0		3.0 ± 0.4
3.0		3.0 ± 0.5
3.0		3.0 ± 0.5
3.0		2.9 ± 0.5
2.9
N4ITK2.9 ± 0.5
2.9		2.9 ± 0.4
3.0		3.0 ± 0.4
3.0		3.0 ± 0.4
3.0		2.9 ± 0.4
2.9
AmygdalaSPM-BFC1.4 ± 0.2
1.4		1.4 ± 0.2
1.4		1.4 ± 0.3
1.4		1.4 ± 0.2
1.4		1.4 ± 0.2
1.3
N4ITK1.3 ± 0.2
1.3		1.3 ± 0.2
1.3		1.4 ± 0.2
1.4		1.4 ± 0.2
1.4		1.3 ± 0.2
1.3
CerebellumSPM-BFC27.0 ± 3.5
26.9		27.1 ± 3.4
27.0		25.8 ± 3.5
25.2		26.8 ± 3.3
26.5		26.8 ± 3.4
26.5
N4ITK26.9 ± 3.4
26.2		27.1 ± 3.3
26.8		25.8 ± 3.5
25.4		26.8 ± 3.2
26.4		26.8 ± 3.3
26.5
BrainstemSPM-BFC6.5 ± 0.8
6.3		6.4 ± 0.7
6.3		6.5 ± 0.7
6.5		6.3 ± 0.7
6.1		6.4 ± 0.7
6.2
N4ITK6.4 ± 0.7
6.3		6.3 ± 0.7
6.2		6.4 ± 0.7
6.3		6.2 ± 0.7
6.1		6.3 ± 0.7
6.1
Table 3. Wilcoxon signed-rank test statistics (Z and p-value (p)) for evaluating the differences regarding regional segmentation volumes when using two BFC methods (SPM-BFC and N4ITK) for each of the five BE methods (Manual, BET2, SWS, HD-BET, and SynthStrip).
Table 3. Wilcoxon signed-rank test statistics (Z and p-value (p)) for evaluating the differences regarding regional segmentation volumes when using two BFC methods (SPM-BFC and N4ITK) for each of the five BE methods (Manual, BET2, SWS, HD-BET, and SynthStrip).
Brain TissueBrain Extraction MethodZp
Cortical Grey MatterManual−3.703<0.001 *
BET2−3.329<0.001 *
SWS−3.524<0.001 *
HD-BET−3.0530.002 *
SynthStrip−3.0520.002 *
White MatterManual−3.739<0.001 *
BET2−1.8440.065
SWS−3.634<0.001 *
HD-BET−3.2870.001 *
SynthStrip−2.3480.019 *
Cerebral Spinal FluidManual−3.268<0.001 *
BET2−2.9750.003 *
SWS−2.0180.044 *
HD-BET−2.4390.015 *
SynthStrip−2.8680.004 *
Deep Nuclear Grey MatterManual−1.5750.115
BET2−2.2080.027 *
SWS−1.9470.052
HD-BET−1.5960.110
SynthStrip−1.6560.098
HippocampusManual−1.3530.176
BET2−2.3450.019 *
SWS−0.0460.963
HD-BET−0.4100.682
SynthStrip−0.2060.837
AmygdalaManual−2.4890.013 *
BET2−2.1110.035 *
SWS−1.8450.065
HD-BET−1.5080.132
SynthStrip−1.7320.083
CerebellumManual−0.6760.499
BET2−0.5980.598
SWS−1.3140.189
HD-BET−0.2190.827
SynthStrip−0.5640.573
BrainstemManual−2.2340.025 *
BET2−2.8410.005 *
SWS−2.5590.010 *
HD-BET−3.2070.001 *
SynthStrip−3.317<0.001 *
* Statistically significant differences.
Table 4. Friedman’s two-way analysis of variance by ranks test statistics (Chi-squared (χ2), degrees of freedom (df = 4), p-value (p) and Kendall’s W) for evaluating the general differences regarding regional segmentation using five BE methods (Manual, BET2, SWS, HD-BET, and SynthStrip) with SPM-BFC and N4ITK correction.
Table 4. Friedman’s two-way analysis of variance by ranks test statistics (Chi-squared (χ2), degrees of freedom (df = 4), p-value (p) and Kendall’s W) for evaluating the general differences regarding regional segmentation using five BE methods (Manual, BET2, SWS, HD-BET, and SynthStrip) with SPM-BFC and N4ITK correction.
Brain TissueBias Field Correction Methodχ2pKendall’s W
Cortical Grey MatterSPM-BFC49.565<0.001 *0.563
N4ITK30.032<0.001 *0.341
White MatterSPM-BFC24.176<0.001 *0.275
N4ITK24.836<0.001 *0.282
Cerebral Spinal FluidSPM-BFC50.804<0.001 *0.577
N4ITK40.545<0.001 *0.461
Deep Nuclear Grey MatterSPM-BFC31.852<0.001 *0.362
N4ITK30.378<0.001 *0.345
HippocampusSPM-BFC8.1600.0860.093
N4ITK7.1160.1300.081
AmygdalaSPM-BFC8.4420.0770.096
N4ITK11.0670.026 *0.126
CerebellumSPM-BFC55.486<0.001 *0.631
N4ITK54.740<0.001 *0.622
BrainstemSPM-BFC35.186<0.001 *0.400
N4ITK35.675<0.001 *0.405
* Statistically significant differences.
Table 5. Multiple pairwise comparisons (Bonferroni correction) between BE methods (Manual, BET2, SWS, HD-BET, and SynthStrip) for each RBV and BFC, reporting the p-value (p) and the post hoc effect size (r).
Table 5. Multiple pairwise comparisons (Bonferroni correction) between BE methods (Manual, BET2, SWS, HD-BET, and SynthStrip) for each RBV and BFC, reporting the p-value (p) and the post hoc effect size (r).
Cortical Grey Matter
SPM-BFCN4ITK
BET2SWSHD-BETSynthStripBET2SWSHD-BETSynthStrip
Manualp = 1.000
r = 0.230		p = 1.000
r = 0.007		p = 0.017 *
r = 0.474		p < 0.001 *
r = 0.884		p = 1.000
r = 0.101		p = 1.000
r = 0.072		p = 0.952
r = 0.252		p < 0.001 *
r = 0.654
BET2-p = 1.000
r = 0.237		p = 1.000
r = 0.244		p < 0.001 *
r = 0.654		-p = 1.000
r = 0.137		p = 1.000
r = 0.187		p < 0.001 *
r = 0.589
SWS- - p = 0.014 *
r = 0.482 		p < 0.001 *
r = 0.891		- - p = 0.319
r = 0.323 		p < 0.001 *
r = 0.726
HD-BET---p = 0.066
r = 0.410		---p = 0.076
r = 0.403
White Matter
SPM-BFCN4ITK
BET2SWSHD-BETSynthStripBET2SWSHD-BETSynthStrip
Manualp = 0.283
r = 0.331		p = 0.404
r = 0.309		p = 1.000
r = 0.216		p = 1.000
r = 0.237		p = 0.031 *
r = 0.446		p = 1.000
r = 0.244		p = 1.000
r = 0.144		p = 1.000
r = 0.057
BET2-p < 0.001 *
r = 0.640		p = 0.003 *
r = 0.546		p = 0.002 *
r = 0.568		-p < 0.001 *
r = 0.690		p < 0.001 *
r = 0.589		p = 0.008 *
r = 0.503
SWS- - p = 1.000
r = 0.093		p = 1.000
r = 0.072		- - p = 1.000
r = 0.101		p = 1.000
r = 0.187
HD-BET---p = 1.000
r = 0.022		---p = 1.000
r = 0.086
Cerebral Spinal Fluid
SPM-BFCN4ITK
BET2SWSHD-BETSynthStripBET2SWSHD-BETSynthStrip
Manualp = 0.250
r = 0.338		p = 0.132
r = 0.374		p = 1.000
r = 0.122		p < 0.001 *
r = 0.848		p = 0.701
r = 0.273		p = 0.453
r = 0.302		p = 1.000
r = 0.115		p < 0.001 *
r = 0.762
BET2-p = 1.000
r = 0.036		p = 0.023 *
r = 0.460		p = 0.007 *
r = 0.510		-p = 1.000
r = 0.029		p = 1.000
r = 0.388		p = 0.012 *
r = 0.489
SWS- - p = 0.010 *
r = 0.496 		p = 0.017 *
r = 0.474		- - p = 0.057
r = 0.417 		p = 0.023 *
r = 0.460
HD-BET---p < 0.001 *
r = 0.970		---p < 0.001 *
r = 0.877
Deep Nuclear Grey Matter
SPM-BFCN4ITK
BET2SWSHD-BETSynthStripBET2SWSHD-BETSynthStrip
Manualp = 0.100
r = 0.158		p = 0.066
r = 0.410		p = 1.000 r = 0.137p = 0.150
r = 0.366		p = 0.100
r = 0.388		p = 0.506
r = 0.295		p = 1.000 r = 0.122p = 0.171
r = 0.359
BET2-p = 0.002 *
r = 0.568		p = 1.000
r = 0.022		p = 1.000
r = 0.208		-p < 0.001 *
r = 0.683		p = 0.777
r = 0.266		p = 1.000
r = 0.029
SWS- - p = 0.003 *
r = 0.546		p < 0.001 *
r = 0.776		- - p = 0.057
r = 0.417		p < 0.001 *
r = 0.654
HD-BET---p = 1.000
r = 0.230		---p = 1.000
r = 0.237
Amygdala
SPM-BFCN4ITK
BET2SWSHD-BETSynthStripBET2SWSHD-BETSynthStrip
Manual----p = 1.000
r = 0.036		p = 1.000
r = 0.108		p = 1.000
r = 0.144		p = 1.000
r = 0.108
BET2-----p = 1.000
r = 0.144		p = 1.000
r = 0.180		p = 1.000
r = 0.072
SWS----- - p = 1.000
r = 0.036		p = 1.000
r = 0.216
HD-BET-------p = 0.952
r = 0.252
Cerebellum
SPM-BFCN4ITK
BET2SWSHD-BETSynthStripBET2SWSHD-BETSynthStrip
Manualp = 1.000
r = 0.108		p < 0.001 *
r = 0.891		p = 0.861
r = 0.259		p = 0.565
r = 0.287		p = 1.000
r = 0.223		p < 0.001 *
r = 0.805		p = 1.000
r = 0.216		p = 1.000
r = 0.137
BET2-p < 0.001 *
r = 0.999		p = 0.150
r = 0.366		p = 0.087
r = 0.395		-p < 0.001 *
r = 1.028		p = 0.036 *
r = 0.438		p = 0.171
r = 0.359
SWS- - p < 0.001 *
r = 0.632		p < 0.001 *
r = 0.604		- - p < 0.001 *
r = 0.589		p < 0.001 *
r = 0.668
HD-BET---p = 1.000
r = 0.029		---p = 1.000
r = 0.079
Brainstem
SPM-BFCN4ITK
BET2SWSHD-BETSynthStripBET2SWSHD-BETSynthStrip
Manualp = 1.000
r = 0.093		p = 1.000
r = 0.244		p = 0.003 *
r = 0.539		p = 0.506
r = 0.295		p = 1.000
r = 0.151		p = 1.000
r = 0.208		p = 0.002 *
r = 0.568		p = 1.000
r = 0.316
BET2-p = 0.250
r = 0.338		p = 0.031 *
r = 0.446		p = 1.000
r = 0.201		-p = 0.171
r = 0.359		p = 0.057
r = 0.417		p = 1.000
r = 0.165
SWS- - p < 0.001 *
r = 0.783		p = 0.003 *
r = 0.539		- - p < 0.001 *
r = 0.776		p = 0.005 *
r = 0.525
HD-BET---p = 1.000
r = 0.244		---p = 0.952
r = 0.252
* Statistically significant differences.
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Vaz, T.F.; Naseh, N.; Hellström-Westas, L.; Moreira, N.C.; Matela, N.; Ferreira, H.A. Assessment of Regional Brain Volume Measurements with Different Brain Extraction and Bias Field Correction Methods in Neonatal MRI. Appl. Sci. 2024, 14, 11575. https://doi.org/10.3390/app142411575

AMA Style

Vaz TF, Naseh N, Hellström-Westas L, Moreira NC, Matela N, Ferreira HA. Assessment of Regional Brain Volume Measurements with Different Brain Extraction and Bias Field Correction Methods in Neonatal MRI. Applied Sciences. 2024; 14(24):11575. https://doi.org/10.3390/app142411575

Chicago/Turabian Style

Vaz, Tânia F., Nima Naseh, Lena Hellström-Westas, Nuno Canto Moreira, Nuno Matela, and Hugo A. Ferreira. 2024. "Assessment of Regional Brain Volume Measurements with Different Brain Extraction and Bias Field Correction Methods in Neonatal MRI" Applied Sciences 14, no. 24: 11575. https://doi.org/10.3390/app142411575

APA Style

Vaz, T. F., Naseh, N., Hellström-Westas, L., Moreira, N. C., Matela, N., & Ferreira, H. A. (2024). Assessment of Regional Brain Volume Measurements with Different Brain Extraction and Bias Field Correction Methods in Neonatal MRI. Applied Sciences, 14(24), 11575. https://doi.org/10.3390/app142411575

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop