Longitudinal evaluation of white matter tracts post traumatic brain injury using a pediatric porcine model

Abstract

Background:

Traumatic brain injury (TBI) induced white matter damage in the brain has a lasting impact on brain functions, leading to cognitive and motor impairments.

New method:

To better predict patient deficits, develop treatment plans, and identify novel therapies, longitudinal effects of TBI on white matter tracts were evaluated using a pediatric porcine model. Diffusion-weighted MRI data from a cohort of female piglets (n = 22) consisting of three groups, including sham, mild- and severe-TBI pigs, was acquired at 4 time points (pre-TBI, 1D-, 63D-, and 119D-post TBI). A group of common white matter tracts was generated using a probabilistic independent component tractography approach. The tracts were evaluated, both laterally between two hemispheres and longitudinally among 4 time points, to measure changes in injured tracts over a 120-day period. Fractional anisotropy (FA), mean diffusivity (MD), radial diffusivity (RD), and axial diffusivity (AD) were calculated for these tracts and compared.

Results:

9 pairs of white matter tracts were generated, 7 of which corresponded to those found in adult pigs in the Porcine Neurological Imaging Space. Our results showed consistently lower FA and higher diffusivity in the ipsilateral hemisphere (with TBI injury) for all 9 pairs. Longitudinal changes in these metrics indicated a complex interplay between typical development and TBI recovery.

Comparison with existing methods:

This method captured similar trends in FA and diffusivity as found in empirical data.

Conclusion:

The probabilistic IC tractography approach effectively characterized lateral and longitudinal changes to white matter tracts using a multi-group and multi-time-point pediatric porcine TBI model.

1. Introduction

Despite significant research into the effects of traumatic brain injury (TBI), there is much that remains unanswered, particularly in the understudied pediatric population (Ignacio et al., 2024; Kundu and Singh, 2023; Shin et al., 2023; Stojanovski et al., 2019). White matter tracts are bundles of axons that connect multiple regions of the brain and transport information throughout the network. These tracts consist primarily of myelinated axons. Significant changes to these tracts can have profound consequences (Abdullah et al., 2022; Armstrong et al., 2016) and are associated with multiple disorders and injuries such as Alzheimer’s disease, Parkinson’s disease, and TBI (Festa et al., 2024; Spitz et al., 2017; Thomas et al., 2024). Thus, a more complete understanding of how these neurological diseases and injuries affect white matter tracts can aid in understanding the types of patient deficits that may occur (e.g., motor disfunction) and the development of novel treatment methods.

Multiple metrics can be associated with the overall integrity of white matter tracts. Fractional anisotropy (FA) measures the directional dependence of water molecules within tissue. Reductions in FA are typically associated with white matter integrity degradation, meaning that diseases and injuries are often coupled to an overall lowering of FA within the impacted region (Veeramuthu et al., 2015). On the other hand, metrics of diffusion usually follow the opposite trend. Three such metrics are mean diffusivity (MD), radial diffusivity (RD), and axial diffusivity (AD). MD measures the overall water diffusion within a voxel, while the other two quantify the diffusion in specific directions. AD is measured along the direction in which diffusion is strongest, which is along the length of the white matter tract. On the other hand, RD measures the diffusion orthogonal to the surface of the tract. Increases in these metrics represent heightened water diffusion, and so are indicative of white matter injury. Previous studies have shown how these metrics change over time in pigs and humans, and how these changes can be associated with neurodevelopment and cognitive maturation (Kochunonov et al., 2024; Lebel et al., 2017).

Pigs and humans have similar brain anatomy and physiology. During the first year of life, the human and pig brain undergo significant growth and development including myelination and neurogenesis (Babikian and Asarnow, 2009; Conrad et al., 2012; Dobbing and Sands, 1979; Jonsson et al., 2013; Ryan et al., 2016). Both the pig and human brain have gyrencephalic brains and are more than 60 % white matter. Insults to the human pediatric brain uniquely and drastically alter white matter development showing arrested white matter development and decreased white matter integrity years after injury (Ewing-Cobbs et al., 2008; Wilde et al., 2006). Therefore, studying the potential effects of TBI in a translational pediatric piglet model will likely be highly predictive of what will occur in human patients.

Previous work has been done in identifying white matter structures in injured subjects. In particular, one study (Squarcina et al., 2012) showed that difficulties arise when performing probabilistic tractography for injured subjects, which can contribute to fewer injured tracts being generated and an underestimation in the amount of damage. However, templates generated from healthy subjects were successfully used in injury analysis as an alternative to tractography. Asymmetry of white matter structure was also associated with TBI (Presson et al., 2015). Independent component analysis (ICA) algorithms have been previously used in the segmentation of the brain using connectivity matrices (Wu et al., 2015), where group ICA was applied to the structural connectivity matrices to generate independent spatial maps and their connectivity profiles. These spatial maps were shown to be consistent with anatomical tracts and were beneficial in distinguishing healthy controls and schizophrenia subjects. Instead of relying on group ICA, individual structural connectivity can also be used to generate tracts, as outlined previously.(Sun et al., 2024)

In this study, we investigated the longitudinal effects of TBI within piglet data using a probabilistic IC tractography method. Specifically, common pairs of tracts were generated and subsequently compared between groups varying in TBI injury severity. Changes resulting from TBI were measured using a combination of spatial correlation to uninjured brains and through the comparison of commonly used diffusion metrics between injured and uninjured hemispheres.

2. Materials and methods

2.1. Subjects and induced traumatic brain injury

Subjects chosen for this study consisted of a total of 22 female piglets (swine, Sus scrofa domesticus, Yorkshire crossbreed), supplied by the University of Georgia, 8 weeks old. No pigs were excluded in the final analysis. Pigs at 8 weeks old are at a similar level of development to a human toddler around 2–3 years old while their ages at the end of the study (25 weeks) are generally more similar to a human teenager around 13–14 years old (Tohyama and Kobayashi, 2019). Pigs were subsequently divided into three groups. The first group (n = 6) consisted of sham pigs that underwent a craniectomy, but no TBI. The second (n = 8) and third groups (n = 8) received a mild (mTBI) or severe (sTBI) TBI, respectively. The surgical procedure was performed as in (Ahmed et al., 2025). Specifically, the surgical site was sterilized using Betadine solution and 70 % ethanol, and then covered with a betadine-impregnated drape (Ioban2) and sterile disposable drape. A 4 cm, left-sided incision was made at the top of the cranium to expose the underlying skull. A 24×16mm elliptical craniectomy was then performed using an air drill 8 mm below the left posterior junction of the coronal and sagittal sutures to expose the underlying dura. After skull removal, the area was flushed with sterile saline. Each pig was then secured in a controlled cortical impact (CCI) device (University of Georgia Instrument Design and Fabrication Shop, Athens, GA) (Baker et al., 2019; Kinder et al., 2019; Osier and Dixon, 2016). The blunt impactor tip (solid stainless steel, 15 mm diameter) was centered over and touching the exposed dura, and a TBI at the left motor cortex region was subsequently induced with the following parameters: 4 m/s velocity, 3 mm (mTBI) or 9 mm (sTBI) depth of depression, and 400 ms dwell time. Following the impact, consistent moderate pressure was applied for 10 min, then the site was flushed with sterile saline, and the skin was finally re-apposed with suture and surgical glue. The administration of analgesics for pain relief is outlined in the supplementary information.

A diagram showing the location and size of the TBI can be found in Fig. 1A. This study was performed in accordance with the National Institutes of Health (NIH) Guide for the Care and Use of Laboratory Animals. All procedures were reviewed and approved by the University of Georgia Institutional Animal Care and Use Committee (Animal Use Protocol: A2023 07-021-Y2-A7).

Fig. 1.

Traumatic brain injury summary and data processing/analysis flowchart: (A) Diagram representing the induced TBI size and location for each pig. (B)This flowchart outlines the steps taken to process and analyze the data. Raw images are processed and corrected before using probabilistic tractography to generate the voxelwise connectivity. Using independent component analysis, candidate tracts are generated, which are then tested to determine if they are common to the majority of subjects without TBI. Common tracts are then categorized into symmetric pairs and metrics are calculated for lateral comparison.

2.2. Data acquisition

All MRI data were acquired using a GE MR750 3.0 T magnet (GE Healthcare). Each piglet was scanned a total of four times, beginning approximately one week before (pre-TBI) the piglet underwent the craniectomy or TBI. Next, each piglet was scanned 1 day (1Day), 63 days (63Day), and 119 days (119Day) post-injury. For the first two time points (pre-TBI and 1Day), diffusion tensor imaging (DTI) images were collected using an echo planar spin echo sequence with 2 mm isotropic voxels, TR = 10,000 ms, minimum TE (optimized TE, typically less than 90 ms), flip angle = 90 °, 64 × 64 acquisition matrix (k-space), 28 slices, 128 mm FOV, two acquisition averages (NEX=2), and a total scan time of 11 min 10 s. For the final two time points (63Day and 119Day), to accommodate larger head sizes and to keep isotropic voxels, the parameters were changed to acquisition matrix 96 × 96 (k-space), 32 slices, 192 mm FOV, with other parameters remaining the same. Each scan consisted of 33 volumes, with (3 volumes of b = 0 and 30 volumes of diffusion-weighted gradients (equally distributed over a sphere, details of b vectors provided in the supplementary file) and b = 1000. Additional b = 0 images with opposite phase encoding (anterior-posterior and posterior-anterior) were collected for susceptibility artifact correction. An adult pig T1 anatomical image that was used as an initial reference space for tract analysis was previously collected by another group (Benn et al., 2020) using a Philips 3 Tesla Achieva scanner with TR= 10ms, TE = 4.8ms with an acquisition matrix of 264 × 238 and 150 slices and 0.8 mm isotropic voxels (Benn et al., 2020).

2.3. Processing

All data was processed using the FMRIB Software Library (FSL) (Jenkinson et al., 2012). Full details are given in the supplementary material. Processing and analysis steps are briefly outlined in Fig. 1B. Corrections were made to mitigate distortions from movement and magnetic susceptibility differences before probabilistic tractography was performed. This tractography requires seed and termination masks. In this case, both of the masks were of the whole brain. For each iteration, the default number of 5000 streamlines were generated. Each streamline was limited to a curvature of ±80 ° by default. While a specific termination mask was not given, streamlines were terminated if they were about to pass through a previously traversed region. Using the Matrix2 advanced option, a connectivity matrix was constructed in which each entry is the number of streamlines passing from a specific seed voxel through another voxel. This connectivity matrix was then used as input to an ICA algorithm, which generated a predetermined number (50 in this case) of potential white matter tracts (to be verified) as in (Sun et al., 2024).

Once all the tracts were generated, the masked DTI image of each pig was registered using FSL’s FLIRT to the same adult pig T1 anatomical image (Benn et al., 2020) using the following parameters in FSL’s FLIRT (Jenkinson et al., 2002): unaligned and incorrectly oriented, 12 ° of freedom, correlation ratio cost function, and trilinear interpolation. This was necessary as the tracts were originally computed in each individual subject’s diffusion space. The transformation for this registration was then applied to the tract images themselves to transform them from diffusion space to reference space in order to fairly compare each tract.

2.4. Tract analysis

Common tracts are necessary to determine and faithfully compare changes over time as a result of injury. To determine commonality, a comparison to a reference should be made. While the sham pigs do exhibit minimal head injury due to craniectomy, their white matter tracts are assumed to remain intact. Therefore, the sham pigs were chosen as the primary group through which common tracts were found. This selection allows for tracts that are consistently generated, with consideration of possible developmental factor during the longitudinal evaluation period. Therefore, a total of 24 scans were used (i.e., the same 6 sham subjects appearing at four time points) in this process with a reference pig chosen from among the 6 pre-TBI shams. This set will be referred to as the reference group. The spatial Pearson correlation was calculated between each of the 50 components of the reference pig with each component of the other 23 data sets. The reference components were classified as common tracts if their spatial correlation was above a threshold value for a majority of the other 23 subjects. For each subject, the component assigned to each reference component was the one with highest correlation. In this study, a common tract must appear in at least 21 of the 23 (91.3 % repeatability) other scans with a minimum correlation threshold of 0.50. Given that all time points were considered, the common tracts should be those that were not significantly altered in shape and location by neurodevelopment and should be consistently found in similar healthy subjects, while still potentially exhibiting changes in FA, MD, etc. that would be consistent with neuro-development. Because ICA components can be fairly similar and might have high correlations to multiple reference tracts, components were ranked according to their spatial correlation. Tracts with the highest correlation were then iteratively removed and correlations were recalculated for the remaining tracts until each reference tract is paired with a unique tract for each subject. After all common tracts were determined, a group of average tracts, to be used as a reference in the following steps, was calculated by averaging the corresponding components in the reference group.

2.5. Pair evaluation

To describe the impact of TBI on these tracts, it is necessary to have corresponding tracts paired between the ipsilateral and contralateral hemispheres to determine if the injury may have contributed to significant lateral differences. These pairs were determined using masks for 52 regions of interest (ROI) in a gray matter atlas (26 in each hemisphere), see Table 1 of the supplementary materials (Saikali et al., 2010). If two tracts are symmetric across hemispheres, it is expected that they will intersect corresponding regions in similar ways. The spatial Pearson correlation of each average tract with each of the 52 masks was calculated and stored in a 52-component vector. Tract symmetry was determined by correlating the ipsilateral and contralateral components of specific pairs.

A supplementary method used was to compare with known pairs. A group of 27 common white matter tracts were found in a group of adult pigs in the Porcine Neurological Imaging Space (PNI50) (Benn et al., 2020). Similarly to what was done for individual tracts, the spatial correlation of each common tract was calculated in relation to the 27 known tracts. Tracts that were found to be pairs using the previous method should also have high correlation to adult pairs in the common space. Thus, both methods were leveraged to classify tract pairs.

2.6. Analysis of tract changes

To investigate differences between groups and the effect of TBI, this exploratory work uses a combination of spatial correlation and diffusion metrics to make lateral comparisons between corresponding tract pairs. Spatial correlations can be made between each subject’s common tracts and their corresponding average reference tracts constructed just from the shams. At each time point, the spatial Pearson correlation of each of the 50 components in all 22 subjects was calculated relative to the average tracts to find each subject’s components that matched best with the reference. These correlations were calculated by applying Matlab’s corrcoef function to the 3D matrices that represent the distributions of the tracts in space. Since there could be slight variations in the tracts over time, these correlations were performed relative to the average at that time point. The subject tract that corresponds to each of the average tracts, which will later be used for analysis, was determined by ranking correlations and choosing the component with the maximum correlation. These spatial correlations quantify the physical overlap of the tracts in individuals with the average reference tracts. Thus, significant changes in these values can indicate changes in the physical location or shape relative to the reference tracts. The spatial correlations to each of the average reference tracts were analyzed using a 3-way ANOVA to account for injury severity, time point, and hemisphere. This was followed by post-hoc tests for significance implemented using Matlab’s multcompare function. These tests were used to study the lateral differences between hemispheres as well as longitudinal effects from the first time point (pre-TBI) to the last one (119Day post TBI).

Individual tracts were masked using FSL and then FA, MD, RD, and AD were calculated (see supplementary materials for details). Masks associated with each tract were applied to the generated whole-brain FA, MD, RD, and AD images to extract those parameters within the tract volume. The value assigned to each tract was the mean value of the metric for all voxels within the tract and so could be associated with overall changes in the tract but did not measure local changes to specific regions within the tract itself. Data for diffusion metrics for each tract were also analyzed using a 3-way ANOVA to account for injury severity, time point, and hemisphere. Each analysis was followed by post-hoc tests for significance, which were used to compare both lateral and longitudinal effects of TBI.

3. Results

3.1. Generated tracts

Using a correlation threshold of 0.50 in at least 21 of the other 23 sham scans in the reference group, a total of 24 tracts were selected as common tracts. Calculating the correlation between spatial distribution vectors, there were 9 unique pairs out of the 24 tracts with correlations of above 0.89. Seven of these pairs additionally had relatively high correlation (>0.5) to the known 27 adult tracts (Benn et al., 2020), while the other 2 pairs were not found to be associated with any of the tracts. The 9 tract pairs are named in Table 1. The matrix (size of 18 × 52) storing the spatial correlations of the 18 tracts with the 52 gray matter ROIs is seen in Fig. 1S–A of the supplementary material. Specifically, the ipsilateral and contralateral tract values are stored in the upper left (rows 1–9) and lower right (rows 10–18) quadrants, respectively. The two smaller 9 × 26 matrices represent the spatial distributions of the tracts within their respective hemispheres. The similarity of the two matrices emphasizes the corresponding similarity of the pairs of tracts. A representative comparison between the left and right superior thalamic radiation is shown in Fig. 1S–B, followed by the comparisons for four additional tracts (posterior thalamic radiation, corticospinal tract, optic radiation, and the unidentified tract 6 which connects the putamen and primary visual cortex)in Fig. 1S–C to F. Comparisons for the other 4 pairs of tracts not discussed in the main text are shown in Fig. 2S (supplementary material). Each of the 26 points in the scatter plot represents the correlation of the tracts to one of the 26 gray matter ROIs (Table 1, supplementary material), in their respective hemispheres, associated with the gray matter atlas. If these tracts were completely symmetric, the points in the scatter plot would be perfectly correlated and lie on the diagonal line, y = x. Fig. 1B to F show that the correlation is distinctively high with a Pearson correlation of at least 0.93 for all but the optic radiation. The least square fit is plotted to compare to a perfect correlation. These lines both align well with the data and thus show the high degree to which the tracts are similar.

Table 1.

Names of each pair of common tracts. The first and second number of the middle column are associated with the ipsilateral and contralateral sides, respectively.

Pair Number	Tract numbers	Anatomy Name
1	1,10	Corticospinal Tract (CST)
2	2,11	Superior Thalamic Radiation (STR)
3	3,12	Anterior Thalamic Radiation (ATR)
4	4,13	Optic Radiation (OR)
5	5,14	Uncinate Fasciculus (UF)
6	6,15	Not listed in PNI50. Connects the primary visual cortex to the putamen by passing near the lateral ventricle.
7	7,16	Not listed in PNI50. Circular structure wrapping around the ventral anterior thalamic nucleus and the mediodorsal thalamic nucleus.
8	8,17	Posterior Thalamic Radiation (PTR)
9	9,18	Fornix (FX)

The averaged pre-TBI tracts are visualized in Fig. 2A in reference to the anatomical PNI50 T1 image (Benn et al., 2020). The correlations between the pre-TBI tracts of all subjects to the average reference tracts at the first time point (pre-TBI) were found to be very high. Besides, a good symmetry between tracts on the ipsi- and contra-lateral sides of the brain is also seen in Fig. 2B. Every tract had a median correlation of at least 0.6. As would be expected from the method of analysis, high correlation was associated with each sham group at all four time points (Fig. 3S, supplementary material). The high level of correlation (average of median correlation for each tract=0.826) of these tracts shows the stability in the method for generating these tracts in healthy subjects.

Fig. 2.

Uninjured white matter tracts: (A) Visualization of tracts within the PNI50 T1 space. Tracts are numbered in the same manner as in Table 1. (1): corticospinal tract, (2): superior thalamic radiation, (3): anterior thalamic radiation, (4): optic radiation, (5): uncinate fasciculus, (6): Unidentified tract connecting the putamen and primary visual cortex, (7): Unidentified tract wrapping around the ventral anterior nucleus and mediodorsal nucleus, (8): posterior thalamic radiation, (9): fornix. Each pair is shown in the coronal (left) and sagittal planes (right) and exhibits bilateral symmetry. Red and blue color for ipsilateral and contralateral tracts, respectively.; (B) Plot of the spatial Pearson correlation between the tracts from the 22 individual subjects before TBI with the computed average sham tracts at that time point for each of the 9 pairs. All tracts showed high correlation to the average at this stage, with tract #5 performing the poorest in each group. Tracts are ordered as in Table 1. Red and blue color for ipsilateral and contralateral tracts, respectively.

3.2. Lateral comparison

Varying levels of correlation were found for the sTBI group tracts in the later time points. For this group, visualization of one tract (superior thalamic radiation) across each time point can be seen in Fig. 3A. Of the generated tracts, 5 tracts, including 2 majorly injured tracts (superior thalamic radiation and posterior thalamic radiation) and 3 minorly affected tracts (corticospinal tract, optic radiation, and tract #6 - an unclassified tract), were analyzed. Among them, two important pairs of tracts, the superior thalamic radiation (STR) and posterior thalamic radiation (PTR), showed lower correlations (p < 0.001) on the ipsilateral hemisphere (right side of the brain in Fig. 2A) for the sTBI group. The correlations between these 2 injured tracts with the average tracts generated by the shams at pre-TBI,1Day, 63Day, and 119Day are shown in Fig. 3B and C, respectively. The correlation of both the STR and PTR for the sTBI group was significantly lower ipsilaterally by 1Day, which continued to persist through the 119Day time point.

Fig. 3.

Longitudinal visualization of an injured tract and spatial correlations with uninjured tracts over time: (A) Visualization of the superior thalamic radiation in the sTBI group over the four time points. The approximate TBI location is marked with a red circle. The brain image is that of the T1 PNI50 reference and thus the TBI itself cannot be seen; (B) Correlations for the STR at Pre, 1Day, 63Day, and 119Day; (C) Correlations for the PTR at Pre, 1Day, 63Day, and 119Day; (D) Correlations for the CST at Pre, 1Day, 63Day, and 119Day; (E) Correlations for the OR at Pre, 1Day, 63Day, and 119Day; (F) Correlations for the unidentified pair #6 at Pre, 1Day, 63Day, and 119Day; Spatial correlations were calculated between the average reference tracts at each time point with the corresponding tracts in the individuals. Correlations values were compared using a 3-way ANOVA followed by post-hoc testing for significance. Both lateral and longitudinal significant differences are shown. The two major pairs (STR and PTR) show significant lateral differences in the sTBI group at all post-TBI times, while the CST only shows this difference at 1Day. Ipsilateral tracts are significantly less similar to the sham tracts than their contralateral counterparts. (1 star * = p < 0.05, 2 stars ** = p < 0.01, 3 stars *** = p < 0.001) Red and blue color for ipsilateral and contralateral tracts, respectively.

Another tract showing a significantly lower ipsilateral correlation was the CST. At 1Day, the ipsilateral value was significantly lower (p < 0.05) than its contralateral counterpart. However, this significant difference was not detected in the final two time points (Fig. 3D). The optic radiation (OR) and tract #6, one of the unclassified pairs, did not display any significant lateral differences in correlation values for any group at any time point, which suggests very minor spatial changes (Fig. 3E and F). Plots showing the correlations for all other pairs of tracts are included in Fig. 4S (supplementary material).

Representative plots for lateral differences in four diffusion metrics of the two majorly injured tracts, STR and PTR are presented in Fig. 4, with results for the STR in the left column and for the PTR in the right column. These two tracts appear to have been severely impacted by the TBI. The STR showed significantly lower ipsilateral FA compared to contralateral FA for all three groups and time points (Fig. 4A). In particular, the sTBI group showed a significant difference at all post-TBI time points. However, a similar trend was seen at the pre-TBI time point as well as in both the sham group and mTBI group at 1Day. The significant lateral difference disappeared for both these groups by 63Day, but it reappeared for the mTBI group at 119Day. A corresponding trend of significantly higher ipsilateral MD, RD, and AD was also present. In this case, the same trend was seen in each of the three groups, but the sTBI group values reached a higher range than the other two groups (Fig. 4C,E,G). Similarly, the PTR had a significantly lower ipsilateral FA for all groups at all time points, which can be seen in Fig. 4B. However, there was not such a constant significant difference for the MD, RD, and AD (Fig. 4D,F,H). These metrics had significantly higher ipsilateral values for the sTBI group at both 63Day and 119Day along with higher ipsilateral values for the mTBI group at these two time points for all but the AD at 63Day. Significantly higher ipsilateral values were also detected for the sham group at 119Day, but these differences were at a lower level of significance (p < 0.05).

Fig. 4.

Lateral evaluations of majorly impacted tracts: (A) Fractional anisotropy of the STR; (B) Fractional anisotropy of the PTR; (C) Mean diffusivity of the STR; (D) Mean diffusivity of the PTR; (E) Radial diffusivity of the STR); (F) Radial diffusivity of the PTR; (G) Axial diffusivity of the STR; (H) Axial diffusivity of the PTR; Distributions for each group and time point were compared laterally using a 3-way ANOVA followed by post-hoc testing. (1 star * = p < 0.05, 2 stars ** = p < 0.01, 3 stars *** = p < 0.001). Red and blue color for ipsilateral and contralateral tracts, respectively.

Three of the minorly injured tracts also showed significant lateral differences, but their trends were not as clear. Their results are given in Fig. 5. These tracts tended to exhibit more minor injuries than the previous two pairs. FA values of the CST were significantly lower for the ipsilateral tracts as compared to contralateral tracts for the sTBI group at 1Day and 63Day. However, this difference was not detected at 119Day (Fig. 5A). Neither of the other two groups, sham and mTBI, showed any significant lateral differences in FA at any time point. Furthermore, no significant lateral differences were found for MD, RD, or AD in the CST (Fig. 5D,G,J).

Fig. 5.

Lateral evaluations of three less impacted tracts: (A) Fractional anisotropy of the CST; (B) Fractional anisotropy of the OR; (C) Fractional anisotropy of one of the unidentified tracts (pair #6); (D) Mean diffusivity of the CST; (E) Mean diffusivity of the OR; (F) Mean diffusivity of one of the unidentified tracts (pair #6); (G) Radial diffusivity of the CST; (H) Radial diffusivity of the OR; (I) Radial diffusivity of one of the unidentified tracts (pair #6); (J) Axial diffusivity of the CST; (K) Axial diffusivity of the OR; (L) Axial diffusivity of one of the unidentified tracts (pair #6); Distributions for each group and time point were compared laterally using a 3-way ANOVA followed by post-hoc testing. (1 star * = p < 0.05, 2 stars ** = p < 0.01, 3 stars *** = p < 0.001). Red and blue color for ipsilateral and contralateral tracts, respectively.

The OR tract showed significantly lower ipsilateral FA for the sTBI group at both 63Day and 119Day as well as for the mTBI group at 119Day (Fig. 5B). Meanwhile, this tract did not show any significant lateral differences in terms of MD, RD, or AD (Fig. 5E,H,K), except for a lower ipsilateral AD for the sham group at 1Day. While the ipsilateral values for these metrics were somewhat elevated for the sTBI group, they were not significantly higher than the contralateral values.

Lastly, tract #6, one of the unclassified tracts, showed minor trends. This tract connects the putamen with the primary visual cortex while passing adjacent to the lateral ventricle. The ipsilateral FA was significantly lower than contralateral FA at both 63Day and 119Day for the sTBI group (Fig. 5C) while the MD, RD, and AD did not show any significant corresponding differences for the sTBI group. However, these three measures of diffusivity did show significant differences at the pre-TBI and 1Day time points. At these times, ipsilateral values were significantly lower than their contralateral counterparts, with differences existing for the sham and mTBI groups at 1Day while not being detected in the sTBI group at that time, despite the same trend (Fig. 5F,I, L).

Plots showing the diffusion metrics for the other four tracts are shown in Fig. 5S (supplementary material) along with a brief discussion of their results.

3.3. Longitudinal comparisons

Both the severely and minorly impacted sets of tracts exhibited longitudinal changes as well. The STR tract showed significantly lower spatial correlation to the average reference tracts for the ipsilateral sTBI group at all post-TBI time points as compared to the values pre-TBI (Fig. 3B). This same trend was also seen in the PTR tract (Fig. 3C). Of the minorly impacted tracts, only the tract #6 showed one significant longitudinal change in spatial correlation to the reference tracts (Fig. 3F). Unlike the previous two tracts mentioned above, tract #6 had a significantly lower correlation for both the ipsilateral and contralateral hemispheres for the sTBI group at 63Day compared to pre-TBI.

Longitudinal differences were also exhibited for the two majorly injured tracts by the four diffusion metrics. Many longitudinal differences were seen for the STR in terms of FA (Fig. 6A). There was an initial significant decrease in the ipsilateral value for the mTBI group at 1Day compared to pre-TBI. This was followed by a significant increase to a stable value by the 63Day time point. Contralateral values increased over the length of the study. The MD, RD, and AD for the ipsilateral sTBI group all increased significantly by 63Day and remained elevated at 119Day (Fig. 6C,E,G). Unlike the sTBI group, both the sham and mTBI groups had decreased contralateral values for the three diffusion metrics at 119Day compared to pre-TBI. Similar trends were exhibited by the PTR. The ipsilateral FA generally increased significantly for both the sham and mTBI groups, but this trend was not seen for the sTBI group (Fig. 6B). Similarly, the contralateral FA also significantly increased for the sham and mTBI groups. Again, the ipsilateral MD, RD, and AD each increased significantly for the sTBI group from 1Day to 63Day and remained elevated (Fig. 6D,F,H). The ipsilateral AD for the mTBI group also increased significantly between the 1–63Day time points, but the MD and RD did not show the same significant trend. The contralateral MD and AD decreased significantly for the sham group between the pre-TBI and 119Day time points, but the other contralateral values remained relatively stable across the time points other than a significant increase for RD between 1Day and 63Day.

Fig. 6.

Longitudinal evaluations of majorly impacted tracts: (A) Fractional anisotropy of the STR; (B) Fractional anisotropy of the PTR; (C) Mean diffusivity of the STR; (D) Mean diffusivity of the PTR; (E) Radial diffusivity of the STR); (F) Radial diffusivity of the PTR; (G) Axial diffusivity of the STR; (H) Axial diffusivity of the PTR; Distributions for each group and time point were compared laterally using a 3-way ANOVA followed by post-hoc testing. (1 star * = p < 0.05, 2 stars ** = p < 0.01, 3 stars *** = p < 0.001). Red and blue color for ipsilateral and contralateral tracts, respectively. Significant differences with solid black lines indicate a difference compared to the pre-TBI value while differences with dotted blue lines indicate a difference compared to the 1Day value.

Several significant longitudinal changes were also seen in the minorly impacted set of tracts. The CST had significantly lower ipsilateral FA at 63Day compared to pre-TBI for the sTBI group (Fig. 7A). Another significant change consistent in MD, RD, and AD is the contralateral decrease from pre-TBI to 119Day of the mTBI group (Fig. 7D,G,J). The corresponding AD value also decreased significantly by 63Day, but this difference did not appear in the MD or RD. The OR had a significant contralateral increase in FA for the mTBI group between pre-TBI and 119Day (Fig. 7B) along with a corresponding significant decrease in MD, RD, and AD for the same group (Fig. 7E,H,K). Likewise, the sham group also had significantly decreased contralateral MD, RD, and AD over the experimental period, but the increases in contralateral FA were not significant. The sTBI group had a significant contralateral decrease between 1Day and 119Day for MD, RD, and AD, but no other significant longitudinal differences were found for this group in the OR tract. Finally, the tract pair #6 also displayed several significant longitudinal trends. This tract had significantly decreased ipsilateral FA by 63Day in the sTBI group along with significantly increasing contralateral FA by 63Day in the mTBI group (Fig. 7C). This tract also had corresponding trends in terms of MD, RD, and AD (Fig. 7F,I, L). The ipsilateral RD was increased by 63Day in the sTBI group along with significant decreases in contralateral MD, RD, and AD by 119Day for the mTBI group. Plots showing the longitudinal trends for the remaining tracts can be found in Fig. 6S (supplementary material).

Fig. 7.

Longitudinal evaluations of three minorly impacted tracts: (A) Fractional anisotropy of the CST; (B) Fractional anisotropy of the OR; (C) Fractional anisotropy of one of the unidentified tracts (pair #6); (D) Mean diffusivity of the CST; (E) Mean diffusivity of the OR; (F) Mean diffusivity of one of the unidentified tracts (pair #6); (G) Radial diffusivity of the CST; (H) Radial diffusivity of the OR; (I) Radial diffusivity of one of the unidentified tracts (pair #6); (J) Axial diffusivity of the CST; (K) Axial diffusivity of the OR; (L) Axial diffusivity of one of the unidentified tracts (pair #6); Distributions for each group and time point were compared laterally using a 3-way ANOVA followed by post-hoc testing. (1 star * = p < 0.05, 2 stars ** = p < 0.01, 3 stars *** = p < 0.001). Red and blue color for ipsilateral and contralateral tracts, respectively. Significant differences with solid black lines indicate a difference compared to the pre-TBI value while differences with dotted blue lines indicate a difference compared to the 1Day value.

4. Discussion

The probabilistic IC tractography method was able to consistently generate pairs of white matter tracts within all groups. Although not for every tract, this method was able to successfully capture the significant differences associated with the injured groups.

High correlations between average and individual subject’s tracts were found in the pre-TBI time point. This suggests that each tract should be consistently found in healthy subjects and are good candidates to analyze injuries caused by TBI. Reductions in correlation indicate possible changes to the tract, and potentially are a good indicator for TBI evaluation. Similar reductions in similarity for TBI white matter were likewise found in (Presson et al., 2015). We were able to find reduced correlations in several tracts for the mTBI and sTBI groups,which shows the utility of this method in detecting changes to white matter.

Among the tracts, the CST showed particular consistency between its results with reduced correlation after injury and corresponding changes to the three diffusion metrics. The CST connects the sensorimotor cortex to the spinal cord and is associated with voluntary movement (Benn et al., 2020). Lower correlations for this tract, specifically in the ipsilateral hemisphere immediately following injury, suggest alterations caused by its proximity to the injury location. Additionally, lower FA values associated with this tract provided further evidence of tract degradation. Although this tract showed significant differences in FA at multiple post-TBI time points, the significance was no longer detected by 119Day. The change in significance level indicated increasing lateral symmetry in tract integrity and thus suggested changes possibly caused by the healing process. Changes in correlation further supported this conclusion. The mean subject correlation to the average CST was at a minimum at the 1Day time point, but then differences were no longer significant at later points.

The STR tract also showed significant differences between hemispheres. However, the results did not seem to be as conclusive. A lower correlation for the TBI case could be a result from injury, but DTI metrics additionally showed differences in the sham subjects. The reason for such differences are not clear, but the trend of the data remains the same. This unusual behavior could have arisen from the proximity to the CST, which may have lead to a number of voxels being included for both tracts. The spatial correlations for this tract did not show significant lateral differences in sham subjects, suggesting a symmetry of their overall shapes. Alternatively, the significant differences in the diffusion metrics for the sham subjects suggested that the internal structure of the tracts were not symmetric. While the significantly lower ipsilateral spatial correlation was detected at each post-TBI time point, the longitudinal comparison to pre-TBI suggested that the tracts may be undergoing complex recovery and development with each hemisphere experiencing dynamic diffusion metrics. The tracts generally experienced upward trends in FA, particularly within the contralateral hemisphere, along with slight decreases in diffusivity. The combination of spatial and diffusion differences suggested that changes in shape occurred with additional effects on the internal structure of the tracts. Additionally, the TBI may have had a very minor effect on the OR and tract pair #6 as they did not display strong significant differences after the TBI.

Similar studies in pediatric subjects show reductions in FA shortly after severe TBI, with reductions appearing within 12 months for moderate TBI (Bartnik-Olson et al., 2021). However, RD was not significantly increased for the severe TBI group shortly following the injury. Comparable results are also seen in (Genc et al., 2017) and (McDonald et al., 2018). These studies suggest that white matter changes can be good indicators of patient outcomes, and thus shows that this method of comparison could be useful in these predictions.

While hemispherical differences existed within TBI groups, there were also several differences in the shams and at the pre-TBI time point. Earlier studies have noted a lateralization of certain tracts (Dai et al., 2016; Ford et al., 2023; Kumpulainen et al., 2023; Raja et al., 2022; Takao et al., 2011). Differences in the FA of healthy subjects are fairly common, and usually are consistent over the lifespan (Takao et al., 2011). A human study has shown that the CST displays a slight rightward asymmetry. The PTR/OR show the opposite trend with a leftward symmetry (Kumpulainen et al., 2023). The trends in asymmetry of these tracts flipped by the time the subjects were 5 years old. A leftward asymmetry of the CST was noted in a study consisting of a wide range of ages (Takao et al., 2011). Our results showed a similar trend to the human subjects later in life, which can suggest differences in how myelinization occurs in humans and pigs.

Unlike the other tracts, the ATR was shown to have significantly lower correlations for all post-TBI time points in the sham group without receiving any apparent injury. This difference may stem from a multitude of different factors. First, this tract may simply be one that is less commonly found using the ICA algorithm and is thus possibly penalized when compared to the averages. Although no injury was done to the brain, the craniectomy may influence the development of this particular tract. The ATR connects the thalamus to the prefrontal cortex and is involved in relaying information for executive functioning (Mamah et al., 2010). As such, it may be more sensitive to minute changes in the brain, leading to the differences we detected.

In general, the probabilistic IC tractography applied in this study was able to consistently generate common white matter tracts in both sham and TBI piglets, with ipsilateral tracts having lower correlation when significant differences existed between the hemispheres. Even with significant differences, a high level of overall correlation was achieved, which showed the consistency and stability of the probabilistic IC tractography in tract generation. For the CST, changing levels of correlation indicated possible healing effects, which was supported by the return of FA to values more consistent between hemispheres

Lastly, one factor that greatly limits these results is the small size of the data set. A larger number of sham animals could potentially lead to the generation of tracts that are even more consistent throughout healthy subjects. Further, the number of independent components per subject limits the number of potential tracts that can be generated. Increasing this number can potentially lead to the generation of new tracts that were not consistently found with the current process. Also, these results were generated using only female subjects, which can limit the scope. Subsequently collected male data can increase our sample size while also allowing for sex to be a variable of analysis. An additional limitation of the method of tract generation was the exclusion of the corpus callosum. While this is an important white matter tract, it was not included in adult pig white matter tracts in the Porcine Neurological Imaging Space (Benn et al., 2020). Although the corpus callosum is not necessarily near the injury site, it may still undergo alterations due to the TBI. The inclusion of this tract in the future could give a more complete analysis of changes made to interhemispheric communication as a result of injury. Another avenue to pursue is to alter the method of probabilistic tractography. In particular, streamlines generated from surfaces might generate different tracts with different consistency than can be achieved using volume masks. Also, interhemispheric tracts can be explicitly generated through an adjustment of the tractography process as none of the common tracts that were found were interhemispheric. Alternatively, another potential research area where the structural changes we can find might be useful is in the integration of structural and functional connectivity (Ahmed et al., 2025). Additional work is needed to more fully develop this tract generation and to test its use as a measure of neuroplasticity. However, DTI and its associated metrics acted as effective biomarkers for TBI, and so are strong candidates for use in future studies.

5. Conclusion

In this study, we were able to successfully generate 9 pairs of white matter tracts by applying the probabilistic IC tractography method to a multi-group, multi-time-point TBI dataset acquired from pediatric porcine brains. Of the 9 pairs that were found, 7 of these were shown to be highly correlated to those found in the PNI50 space. Both lateral comparison between ipsi- and contra-lateral hemisphere and longitudinal comparison among 4 time points, including pre-TBI, 1D-, 63D-, and 119D-post TBI, demonstrated a complex recovery process post TBI. Significant changes were seen in these tracts post-TBI, which were consistent with previous investigations of the effects of TBI on white matter, including lower FA and higher MD and RD values. Out of the 9 pairs of tracts, we found two tracts exhibited severe effects from the TBI, while three additional tracts were minorly impacted. As such, this method of tract generation can be applied as another tool in future studies of TBI and, more specifically, of its longitudinal impact on white matter. The interplay of development and TBI is complex and further research will be needed in the future to decouple these advanced mechanisms.

Appendix A. Supporting information

Supplementary data associated with this article can be found in the online version at doi:10.1016/j.jneumeth.2026.110684.

Data availability

Data will be made available on request.