All procedures were approved by the Animal Care Committee of the Université de Sherbrooke and were used in accordance with policies and directives of the Canadian Council on Animal Care (Protocol 2022–3,349). Four adult domestic short-haired cats, 2 males and 2 females, aged 317, 325, 435, and 793 days and weighing between 3.35 and 5.1 kg (4.37 ± 0.81 kg), were used in the present study. We followed ARRIVE guidelines for animal studies (du Percie Sert et al., 2020). Before and after experiments, cats were housed and fed in a dedicated room within the animal care facility of the Faculty of Medicine and Health Sciences at the Université de Sherbrooke. All cats were used in other studies to maximize their scientific output.

The general protocol of this study is illustrated in Figure 1A. We first performed MRI scans (Ingenia 3.0 T, Philips Healthcare, Best, The Netherlands). After being used for other experimental objectives, the animal was euthanized with a lethal dose (120 mg/kg) of pentobarbital via the left or right cephalic vein. We then conducted postmortem Micro-CT imaging (MiLabs, Utrecht, The Netherlands) before extracting the spinal cord from the vertebral column. The extracted tissue was then processed following dedicated protocols, with the dura matter prepared for histological analysis and the spinal cord treated to enhance imaging quality (see below). Finally, high-resolution ex-vivo Micro-CT scans (Skyscan 1,172, Bruker Micro-CT, Kontich, Belgium) were conducted on the extracted spinal cord followed by image processing, segmentation, and 3D reconstruction.

To extract the spinal cord, we placed the animal in a prone position. The skin overlying the dorsal aspect of the vertebral column was excised along the length of the spine using surgical scissors. Next, incisions were made alongside the spinous processes, laminae, and transverse processes to expose the vertebrae’s spinous processes. Using sharp prong retractors or a scalpel, we retracted the paraspinal muscles to expose the posterior elements of the spine. Using rongeurs, we performed a dorsal laminectomy to expose the entire spinal cord. Then, we carefully lifted the spinal cord from the caudal end and severed the spinal nerves just distal to the DRG. The approximately 40 cm of extracted spinal cord was cut into two equal segments and placed in 500 mL tubes containing 4% paraformaldehyde solution in a 0.1 m phosphate buffer solution (PBS) at 4 °C. After 5 days, the spinal cord was cryoprotected in PBS with 30% sucrose for 72 h at 4 °C.

Before preparing the spinal cord for high-resolution ex-vivo Micro-CT imaging, the dura mater was carefully removed circumferentially between each nerve root using microsurgical instruments. Once dissected, each section was individually placed in a cassette, immersed twice for 5 min in 70% industrial methylated spirits (IMS), and then soaked for 24 h in fresh 70% IMS. The tissues were then completely dehydrated using the following protocol: 1 h each in 70, 75, 90, and 95% IMS, followed by three immersions in 100% IMS for 1 h, 1 h 30 min, and 1 h 30 min, respectively. They were subsequently cleared in three 1-h baths of 100% xylene and finally infiltrated with paraffin wax during two incubations of 2 h 20 min each. We mounted 4 μm slices on glass slides for each spinal segment and then stained them using hematoxylin and eosin following a standardized protocol (Lee et al., 2002). High-resolution imaging of each slide was performed using a NanoZoomer microscope (Hamamatsu Photonics, Hamamatsu, Japan) for detailed visualization and quantitative morphological analysis.

MRI and Micro-CT datasets registration was performed using a combination of two open-source software packages: Spinal Cord Toolbox (SCT) (De Leener et al., 2017), and 3D Slicer (Fedorov et al., 2012). Among the MRI sequences acquired, only the T2 FatSat axial sequence was used for registration due to its high contrast for deep spinal cord tissues. An initial registration step involved spinal cord straightening to facilitate alignment across modalities. For MRI data, the sct_straighten_spinalcord program from SCT was employed, while for CT data, the Curved Planar Reformat module from 3D Slicer was used. MicroCTpm was designated as the fixed reference volume for all subsequent registration procedures. Linear alignment was then performed on MRI data and MicroCTex to align them with the fixed reference volume, utilizing the Transform module in 3D Slicer. Due to the loss of spinal cord tension following ex vivo dissection and tissue preparation, anisotropic scaling was applied to MicroCTex data to restore tissue volume to its original in vivo dimensions. Specifically, scaling factors were determined based on measurements from the intact spinal cord MRI data, resulting in an average scaling of 45% in the longitudinal (z) axis—compensating for elongation due to the release of tension during dissection—and 10% in the transverse (x and y) axes to account for volumetric reduction from dehydration during tissue processing. Finally, the Landmark Registration module in 3D Slicer was used to apply a thin-plate spline (TPS) deformation to both the MRI and MicroCTex volumes to achieve precise alignment and centering of the spinal cord within the spinal canal of the MicroCTpm scans. Although TPS is a nonlinear transformation, in this case it was applied in a controlled manner—primarily to refine alignment within individual axial planes. Multiple fiducial landmarks were placed per axial slice and adjusted collectively to ensure consistent in-plane (x/y) positioning with minimal deformation. Because the volumes had been pre-straightened, this approach introduced only minor bending necessary to ensure the spinal cord conformed accurately to the canal geometry.

The MRI, MicroCTpm and MicroCTex acquisitions were used to generate a unified anatomical 3D model of the cat’s spine. The various tissues were segmented using 3D slicer (Figure 1E). The vertebrae were segmented on the MicroCTpm directly in the reference space. The experimental protocol described in the present study is part of a broader project dedicated to three-dimensional motor mapping of the lumbar spinal cord. As a consequence, the lumbar vertebrae included in the 3D model underwent a dorsal laminectomy to allow direct access to spinal circuits during functional mapping experiments. The comprehensive motor mapping study is currently under preparation for publication. Dura mater and CSF were also segmented in the reference space but were based on coregistration of the MicroCTex and the axial MRI T2 FatSat sequence superimposed in the background. This approach ensured resolution consistency across all tissue segmentations inside the spinal canal. The spinal cord’s gray matter, white matter, and rootlets were segmented directly on the MicroCTex in its native acquisition space and transferred to the reference space using the Transform module in 3D Slicer, applying the saved transformation matrices from the image processing step.

All segmentations were then converted into 3D meshes using 3D slicer and exported in. STL file format. Due to high resolution of CT data, segmentations spanning the entire spinal cord (C3 to caudal end), including gray and white matter, CSF, and vertebrae, produced 3D meshes too large for practical applications. To address this, the following steps were implemented to simplify and convert meshes into solid models:

The computer used for model generation and testing was equipped with an Intel Core i7-10700 CPU (2.9 GHz), NVDIA RTX 3080, 128 GB of RAM, and was running Windows 10. The compatibility and operational use of the generated models were tested using two software platforms: Sim4Life (v7.2.4, ZMT Zurich MedTech AG, Zurich, Switzerland) and COMSOL Multiphysics (v5.5, COMSOL AB, Stockholm, Sweden). For both platforms, testing was conducted using only the internal geometry handling and meshing tools provided by each software.

All statistical analyses were performed using Friedman’s test to evaluate the effect of spinal segment and anatomical location on measured parameters. Specifically, we applied Friedman’s test to assess the impact of spinal segment variations on dura mater thickness, morphological DRG dimensions and volume, rootlet dimensions, rootlet proportion relative to the spinal cord, gray and white matter dimensions. Additionally, we used Friedman’s test to compare measurements across four anatomical locations (dorsal, ventral, lateral left, and lateral right). When a significant effect was detected, a post-hoc pairwise comparison with Bonferroni correction was performed to identify specific differences between anatomical locations. Data are presented as mean ± standard deviation, and a significance threshold of p < 0.05 was used for all statistical comparisons. Analyses were conducted using SPSS Statistics 20.0 (IBM Corp., Armonk, NY, United States).

To determine variations in dura mater along the spinal cord, we measured its thickness from C3 to L7 (Figure 2). Histological sections predominantly revealed a unique fibrous layer composed of wavy collagen fibers, arranged in an organized manner with dispersed fibroblasts (Figure 2A). In some cases, a second layer, potentially corresponding to the dural border cell layer, can be distinguished. These observations contrast with the dura mater of the brain, which is known to exhibit three well-defined layers (Kinaci et al., 2020). At this stage, identifying a consistent pattern in the appearance of this second layer across spinal levels remains challenging. Further investigations, including additional histological staining are currently underway to refine these observations.

We next explored dura mater thickness in four anatomical regions (dorsal, ventral, left lateral, and right lateral) from C3 to L7 across cats (Figure 2B). Dura mater thickness ranged from 36.1 to 172.3 μm in the dorsal region, 49.3–178.8 μm in the ventral region, 35.8–151.8 μm in the left lateral region, and 34.0–145.0 μm in the right lateral region. Visually, thickness fluctuated more prominently in the ventral and dorsal regions compared to the lateral sides. In the lateral regions, dura mater thickness appeared to gradually decrease caudally. We found a significant difference between the four anatomical regions (p = 0.038). On average, the thickness appeared greater on the ventral side (114.52 μm ± 26.73) compared to the dorsal side (97.46 μm ± 28.41), followed by the right lateral (92.44 μm ± 25.55) and left lateral (90.32 μm ± 25.44) sides. Post hoc analysis revealed significant differences between left and ventral sides (p = 0.006), and right and ventral sides (p = 0.028), while all other pairwise comparisons were non-significant (left and right, p = 0.584; left and dorsal, p = 0.273; right and dorsal, p = 0.584; dorsal and ventral, p = 1.00). Dura mater thickness significantly differed across spinal segments. These data indicate that dura mater thickness varies depending on anatomical side and spinal segment.

Figure 2C provides a consolidated view of the total mean dura mater thickness across spinal segments (C3–L7), represented using a grayscale gradient, where darker and lighter shades indicate greater and thinner thickness, respectively. We found a significant difference in thickness across spinal segments. On average, the spinal dura mater thickness ranged between 83 and 123 μm, with the greatest thickness observed at C6–C7 (122.62 μm ± 5.84) and the thinnest at L3–L4 (80 μm ± 20.1). Spinal dura mater thickness is known to be thinner than in the brain, which is approximately 200 μm in cats (Kinaci et al., 2020).

Figure 2D illustrates a segmentation of the dura mater (red) at three spinal levels (C3, T3, and L7), providing a spatial representation of its position relative to the DRG. Our observations indicate that the spinal dura mater terminates just before the DRG across all spinal segments.

To further characterized dura matter characteristics, we also quantified the dural circumference on the 3D model combining MRI and Micro-CT data in one cat (Figure 2E). This measurement revealed maximal values at the cervical enlargement (C6–C8, ~28 mm) and at the lumbosacral junction (L7–S1, ~25 mm), while the thoracolumbar segments (T3–L4) exhibited narrower circumferences, averaging around 20 mm.

The DRG, located bilaterally along the spinal cord, house the cell bodies of primary afferent neurons responsible for transmitting somatosensory and nociceptive information to the central nervous system. Figure 3A illustrates the segmentation of DRG (red) at four segments (C3, C7, T7, and L6), revealing morphological variations that reflect the pattern observed in humans, where DRG are generally larger in the cervical and lumbar regions (Haberberger et al., 2019). Figure 3B quantifies these variations, illustrating changes in length (L_DRG), width (W_DRG), and thickness (T_DRG) from C3 to S1. For Cat 4, measurements at C8 and T1 could not be obtained due to tissue damage during dissection. Consequently, statistical analysis was performed on three cats from C3 to S2, demonstrating significant differences in L_DRG, W_DRG, and T_DRG across spinal segments. Notably, we observed two enlargements at C5–C8 and L5–L7. The longest lengths were recorded at C8 (2.76 ± 0.37 mm) and L7 (3.53 ± 0.55 mm) in the cervical and lumbar regions, respectively. The maximum width and thickness were observed at C7 (1.77 ± 0.07 mm; 1.37 ± 0.14 mm, respectively) and at L6 (1.56 ± 0.02 mm; 1.33 ± 0.16 mm, respectively) in the cervical and lumbar regions, respectively. In contrast, DRG in the thoracic region (T2–T13) exhibited consistently smaller dimensions, aligning with their primary role in trunk innervation.

We observed significant DRG volume (V_DRG) differences across spinal segments, following a distribution consistent with previous measurements, with the highest volumes found at lower cervical (C6–C8, ~2.5–4.8 mm³) and lumbar (L5–L7, ~2.5–4.1 mm³) segments (Figure 3C).

By leveraging high-resolution imaging, we investigated morphological characteristics, architectural organization, and proportional distribution of dorsal and ventral rootlets across spinal segments. Figure 4 presents a comprehensive anatomical reconstruction of spinal rootlet arborization along the spinal cord, emphasizing regional variations in rootlet distribution and orientation. The left panel provides a schematic representation of the spinal cord, while the right panels present individual rootlet tracings for each segment from C3 to S2, offering a comparative visualization of rootlet expansion patterns across the spinal cord for one cat. In the upper cervical region (C3–C5), rootlets extend in a rostro-caudal direction, forming a widespread, dispersed architecture, likely to maximize sensory and motor innervation coverage. Moving caudally, the lower cervical segments (C6–C8), where the brachial plexus emerges (Emamhadi et al., 2016), rootlets became progressively denser and more clustered, exhibiting a more condensed and interwoven pattern. From C7 to T10, rootlets exhibited an arciform, rostrally oriented organization, maintaining a linear, streamlined configuration along the spinal cord. From T10 to L2, rootlets began to re-expand laterally, resembling the bilateral arborization pattern observed in upper cervical segments. Notably, the ventral surface area covered by rootlets appeared larger than the dorsal surface. Finally, in the lumbosacral region (L3–S2), rootlets again adopt an arciform organization.

Table 1 documents the number of rootlets observed in both the dorsal and ventral regions across spinal segments in one cat, as the high-resolution images from the other three cats did not allow for sufficiently precise quantification. The data revealed variations in the number of rootlets across segments, with higher numbers at cervical and lumbar enlargements. Interestingly, we also observed a difference in the number of rootlets between dorsal and ventral regions. Specifically, at C3, the number of dorsal rootlets was higher compared to the ventral side. At C4, C5, T3, T8, T9, T13, L3, L4, L7, and S2, the dorsal and ventral rootlet counts were similar. However, for C6 to T2, T4 to T7, T10, T11, L1, L2, L5, L6, and S1, ventral rootlet counts were visibly higher, consistent with greater numbers of motor/efferent axons.

To investigate morphological variations of the rootlets, we measured the width of the dorsal (W_D) and ventral (W_V) columns, the lengths of the dorsal (L_DREZ) and ventral (L_VREZ) rootlet entry zones, and the distances between these zones on both the dorsal (L_DRBZ) and ventral (L_VRBZ) aspects across different spinal segments (Figure 5). Both column widths exhibited significant variations across spinal segments (Figure 5A). In the cervical region, W_D increased progressively, peaking at C6–C7 (~3.0 mm). Caudal to C8, W_D decreased, reaching a plateau between T3 and T12 (~1.5 mm). We then observed a gradual increase from T13 to L6, with a second peak at L6 (~2.5 mm), followed by a decline to ~0.5 mm at S2. Interestingly, W_V exhibited minimal variations between C3 and L4 (~2.5 mm) before decreasing considerably from L5 and more caudally, being nearly absent at S1–S2 (~0.1 mm).

Regarding the lengths that characterize the rootlet span as they enter the spinal cord, we observed significant segmental variations for both dorsal and ventral entry zones (Figure 5B). L_DREZ and L_VREZ decreased progressively from C3 to T2, then increased from T2 to L1/L2, followed by another decline more caudally. Across spinal segments, the ventral rootlets covered a significantly greater rostrocaudal distance than the dorsal rootlets (L_DREZ, 4.89 ± 0.57 mm; L_VREZ, 6.26 ± 0.45 mm; p = 0.041). These findings suggest that motor rootlets cover a larger portion of the spinal cord. We also observed that dorsal rootlets entered the dorsal columns in an organized linear pattern, while ventral rootlets exited the ventral columns less clearly and without a distinct demarcation (Figure 5C).

To quantify the portion of the spinal cord not covered by rootlets, we measured the inter-rootlet distance between adjacent rootlets in the dorsal (L_DRBZ) and ventral (L_VRBZ) regions (Figure 5D). These both significantly varied across spinal segments. L_DRBZ progressively decreased from C3 to C8 before increasing from T1 to T13 followed by a decline more caudally. These results indicate that, at cervical (C6–C8) and lumbar (L4–L7) enlargements, the dorsal and ventral spinal cord is almost entirely covered by rootlets, whereas at thoracic levels, there are significantly larger gaps between two adjacent segments. L_VRBZ followed a similar trend as L_DRBZ, but ventral inter-rootlet distances were significantly shorter than dorsal distances (L_DRBZ, 2.57 ± 0.43 mm; L_VRBZ, 1.19 ± 0.16 mm; p = 0.046). This reflects a denser coverage of the ventral cord by rootlets. At L7–S2, ventral rootlets exhibited significant overlap, making precise measurements difficult based solely on raw Micro-CT images.

At the caudal segments of the cervical and lumbar cord, dorsal and ventral rootlets covered a greater portion of the spinal cord compared to white matter (Figure 6A). Rootlet proportion varied significantly with spinal segments in the four anatomical regions studied (Figure 6B). The data revealed a first peak at lower cervical segments (50–60%), a smaller mainly constant proportion (25–34%) from T6–L2 before increasing sharply from L5 to S2, where rootlets covered most of the cord in all four regions. Across the four regions, rootlet proportion varied significantly from C3 to S2, ranging from 27 to 97% (Figure 6C). Collectively, these findings underscore region-specific variations in rootlet distribution, with increased coverage at spinal enlargements.

We measured the morphological characteristics of the gray and white matter in the spinal cord to illustrate how these structures vary along its length (Figure 7). In the cervical region, we observed the largest cross-sectional areas in the cervical enlargement (C5–C8), as expected (Figure 7A). The size of the gray matter (i.e., the H-shape portion) was visibly larger at more rostral cervical segments. At thoracic segments, the cord was smaller with a small gray matter, particularly in the ventral horn. Then, we observed a gradual increase in spinal cord size from L1 to L7, marking the lumbar enlargement responsible for innervating hindlimb muscles. The gray matter expanded significantly, with large ventral horns. In the sacral region, spinal cord size diminished, and the contrast between gray and white matter became less distinct.

To quantify rostro-caudal variations, we measured several parameters, including white matter height (D1) and width (D2), ventral horn lateral (D3) and medial (D9) distances from the midline, ventral (D4) and dorsal (D7) depths from the dorsal surface, and dorsal horn lateral (D5) and medial (D8) distances from the midline, as well as dorsal depth (D6) (Figure 7B). All measured parameters significantly varied across spinal segments. D1 reached its highest values in the cervical enlargement (C5–C8, ~4.2–4.6 mm) and at L6 (~3.6–4.1 mm), with a reduction in thoracic and upper lumbar segments (~2.8 mm at T3–L2). Similarly, D2, peaked at C5-C8 (~5.9–6.4 mm) and L6 (~5.5 mm), while thoracolumbar levels (T4–L2) exhibited lower values (~3.6 mm). D3 through D9 exhibited two peaks at cervical and lumbar enlargements, except for D6, which showed a decrease in these regions. To complement these linear measurements with cross-sectional data, we also evaluated the areas of gray matter (A1) and white matter (A2). A1 had peak values at C7–C8 (~4.9–5.9 mm²) and L6–L7 (~5–6.1 mm²) with marked reductions at thoracic and upper lumbar segments (~1.0 mm² at T3–L2). A2 followed the same distribution, with the largest areas observed at lower cervical (C6–C7, ~15.5–18.4 mm²) and lumbar (L6–L7, ~10–12.7 mm²) segments.

Measurement C1, representing white matter circumference at each spinal segment, peaked at C6–C8 (~16.5–17.3 mm) and L6–L7 (~14 mm), while significantly lower values were observed across thoracic and upper lumbar segments (~10.5 mm at T4–L2).

We characterized morphological properties of the (CSF) by measuring both its volume (Figure 8A) and thickness (Figure 8B) on the 3D model by combining MRI and MicroCTex data. The CSF volume varied along the spinal axis, ranging from 100 to 260 mm³ depending on the segment, with a total spinal CSF volume of approximately 4.40 mL (4401.89 mm³) between C3 and S2. CSF thickness, measured at four anatomical positions (dorsal, ventral, left and right lateral), fluctuated with spinal level and ranged between 60 and 160 μm.

To provide the scientific community with a comprehensive resource for advanced morphological analysis and computational modeling, we developed a high-resolution 3D model of the spinal cord and its surrounding structures, integrating vertebrae, dura mater, CSF, rootlets, dorsal root ganglia, white matter, and gray matter from multimodal imaging datasets, including MRI, high-resolution Micro-CT (MicroCTpm and MicroCTex), from C3 to S3 in one cat. This female cat was 466 days old, weighed 3.35 kg, and measured 378.13 mm from the base of the skull to the base of the tail.

To evaluate the operational use of our model in computational frameworks, we tested its integration with two commonly used simulation platforms: COMSOL (v5.5, COMSOL AB, Stockholm, Sweden) and Sim4Life (v7.2.4, ZMT Zurich MedTech AG, Zurich, Switzerland). Figures 9, 10 presents the different layers of the model within COMSOL and Sim4Life, respectively. The 3D model includes the vertebrae (A), dura mater (B), CSF (C), white matter (D), and gray matter (E). Rootlets were excluded from the model due to COMSOL’s meshing limitations. Specifically, the automatic geometry unification process within the software led to major self-intersecting edges and faces, rendering the meshing of rootlets unfeasible. In contrast, the structured meshing approach used by Sim4Life successfully processed the entire model, including the rootlets (Figure 10F).

Table 2 summarizes the computational demands for each software, highlighting the difference in the number of elements, physical memory usage, and time to open. While COMSOL required over 3.5 GB of memory and nearly 15 min of processing to mesh the model without rootlets, Sim4Life processed the full model in under a minute using less than 1 GB of memory. We initially attempted meshing in Sim4Life using its unstructured mesher (comparable to COMSOL’s meshing approach), but this method proved computationally prohibitive, requiring excessive memory and runtime. Consequently, we adopted Sim4Life’s structured meshing, which produced a high-quality mesh efficiently. These findings demonstrate that our 3D spinal cord model is computationally viable, particularly in Sim4Life, and can serve as a foundation for advanced simulations.

In the present study, we found that the thickness of the spinal dura mater varied more prominently in the dorsal and ventral regions compared to the lateral sides, ranging from 83 to 123 μm across spinal segments (Figure 2C). The dura mater is composed of a dense connective tissue layer primarily composed of collagen fibers and fibroblasts, which confer its strength and mechanical resilience (Kinaci et al., 2020; Yu et al., 2013). Thickness varied more prominently in the dorsal and ventral regions compared to the lateral sides, though no significant differences were found between dorsal and ventral measurements. In humans, the thickness of the spinal dura mater also varies across spinal segments (Cavelier et al., 2022a; Hong et al., 2011; Kwon et al., 2018). For instance, it measures approximately 260 μm in the cervical region, 224 μm in the thoracic region, and 159 μm in the lumbar region (Cavelier et al., 2022a). However, other studies have reported slightly higher values at cervical and thoracic levels (Hong et al., 2011; Kwon et al., 2018). Discrepancies between studies likely arise from different methodological and analytical techniques. In landrace cross pigs, the dura mater also varies depending on the spinal segments (Cavelier et al., 2022b), as well as between the dorsal and ventral sides, although conclusions differ among studies (Cavelier et al., 2022b; Chin et al., 2024). At the thoracic level in pigs, thickness ranges from 90 μm to 134 μm, depending on the study. Functionally, greater thickness may indicate increased mechanical protection against trauma, particularly at cervical enlargements (Nagel et al., 2018). Furthermore, from a biophysical standpoint, a thicker dura also contributes to greater electrical insulation (Sin and Coburn, 1983). As such, this parameter is particularly important to consider when conducting computational simulations or designing neuromodulation protocols targeting the spinal cord (Holsheimer, 1998).

Between the dura mater and the spinal cord flows the CSF, which serves essential physiological roles by cushioning the spinal cord, delivering nutrients, and clearing metabolic waste (Czarniak et al., 2023). Using our 3D model reconstructed from MRI and MicroCTex datasets, we were able to estimate the total spinal CSF volume and its morphological characteristics at each spinal segment in one animal. The thickness of the CSF layer varied across spinal segments, ranging on average from 61 to 171 μm. We found a total spinal CSF volume of approximately 4 mL in the cat. In humans, the spinal CSF volume ranges between 81 and 110 mL, depending on the methodology and population studied (Edsbagge et al., 2011; Strbačko et al., 2024). Regional variations have also been described, with the cervical region containing ~19 mL of CSF, the thoracic region ~38 mL, and the lumbosacral region ~25 mL (Edsbagge et al., 2011). The CSF can play an important role in the effectiveness of spinal stimulation protocols. A thicker CSF layer increases the distance between the electrode and the spinal cord, thereby reducing stimulation efficiency, raising activation thresholds, and increasing current spread (Holsheimer et al., 1995; Huang et al., 2014). Moreover, it is well established that spinal CSF volume and geometry vary with body position (e.g., supine, prone, lateral). Studies have shown that spinal cord and CSF dynamics are posture-dependent, with ventral or dorsal shifts of the cord occurring based on gravity and spinal curvature (Magnaes, 1989; Strbačko et al., 2024). It is also important to note that during epidural electrode implantation, the device inevitably compresses the dura mater toward the spinal cord, thereby reducing CSF thickness underneath the electrode. These shifts alter the electrode-to-cord distance, thus modifying activation thresholds and stimulation selectivity (Holsheimer, 1998; Holsheimer and Wesselink, 1997).

On either side of the spinal cord, the DRG house cell bodies of primary afferent neurons responsible for transmitting peripheral sensory inputs to the central nervous system. In cats, there are 36 pairs compared to 31 in humans (Haberberger et al., 2019). In this study, we found significant variations in DRG size along the spinal cord in cats, with prominent enlargements at cervical (C5–C8) and lumbar (L5–L7) levels (Figure 3B). The longest DRG were measured at C8 (2.76 ± 0.37 mm) and L7 (3.53 ± 0.55 mm), while maximum width and thickness were found at C7 (1.77 ± 0.07 mm; 1.37 ± 0.14 mm) and L6 (1.56 ± 0.02 mm; 1.33 ± 0.16 mm), respectively. Conversely, DRG in the thoracic segments were consistently smaller, reflecting their primary role in innervating the trunk rather than the limbs. In humans, DRG morphology also varies across spinal segment (Haberberger et al., 2019), but much of the existing research has focused on the lumbar region. For example, in humans, both the width and length of DRG gradually increase from rostral to caudal lumbar segments, ranging from 4.36 mm (length) and 5.39 mm (width) at L1 to 5.82 mm and 10.83 mm at L5, respectively. DRG have emerged as promising targets for neuromodulation to evoke motor responses after spinal cord injury in cats (Urbin et al., 2019) and humans (Soloukey et al., 2020, 2021). In humans, stimulation of the L4 DRG induces robust knee extension, enabling assisted weight-bearing and demonstrating selective activation of quadriceps muscles (Soloukey et al., 2021). This level of specificity highlights the potential utility of targeting DRG in sensorimotor recovery strategies.

From the DRG, somatosensory signals are relayed into the spinal cord via dorsal rootlets, while ventral rootlets emerge directly from the spinal cord to carry motor outputs. In the present study, we observed that in upper cervical (C3–C4) and thoracolumbar (T10–L2) regions, rootlets were bilaterally distributed and displayed a widespread pattern (Figure 4), likely to enhance sensory and motor innervations. In contrast, lower cervical (C5–C8) and lumbosacral (L3–S2) segments showed dense, clustered, arciform rootlet arrangements, suggesting specialized connectivity for limb control. We also found that rootlet number varied across spinal segments, peaking at cervical and lumbar enlargements. Interestingly, cats had more ventral than dorsal rootlets, particularly in these enlargements, suggesting a potentially higher motor fiber density. Moreover, our measurements of dorsal and ventral column widths across segments showed patterns closely matching those in humans, with peaks in cervical and lumbar enlargements. Similarly, the dorsal and ventral rootlet entry zones showed a progressive decrease from C3 to T2, followed by an increase to L1–L2 and a decline in more caudal segments, again mirroring human findings. As in humans (Mendez et al., 2021), we observed that dorsal rootlets entered the dorsal columns at a very defined line, while ventral rootlets exited the ventral columns without a distinct demarcation.

Lastly, we quantified inter-rootlet distances (L_DRBZ and L_VRBZ), finding dense coverage and minimal spacing at cervical and lumbar enlargements, especially ventrally, where inter-rootlet spacing was significantly shorter than dorsally. In humans, rootlets are difficult to distinguish using MRI, and most anatomical data come from cadaveric dissections, often with variability (Mendez et al., 2021; Zhou et al., 2010). A recent study by Mendez et al. (2021) provided a detailed description of dorsal and ventral roots from C3 to L5 using surgical microscopy in human. They showed that the number of rootlets varied across spinal segments and between dorsal and ventral roots. The greatest number of rootlets was found at the cervical and lumbar enlargements (~8.4 and ~7.9 for dorsal; ~7.2 and ~5.6 for ventral), while thoracic levels had fewer. They measured the distance between the left and right rootlets and found that the width of the dorsal columns (W_D) was greatest at cervical segments (6.84 mm) and smallest at thoracic ones (5.05 mm), showing a secondary increase in the lumbar region. Conversely, the ventral column width (W_V) was largest in the cervical region and decreased caudally, reaching its minimum in the sacral segments. The lengths of the dorsal and ventral rootlet entry zones (L_DREZ and L_VREZ), measured as the distance between the most rostral and caudal rootlets at their entry point into the spinal cord, remained relatively stable from C3 to C7, then slightly decreased before increasing again around T4, followed by a progressive decline at L3 and L5.

From a functional perspective, the marked increase in rootlet proportion at the cervical and lumbar levels is particularly significant, as these spinal enlargements are responsible for the complex innervation of the limbs. The higher density of rootlets in these regions likely represents a key adaptation to enhance the efficiency of sensory and motor signal transmission. While epidural stimulation is often described as targeting the spinal cord, our findings suggest that, especially at cervical and lumbar levels, the extensive rootlet coverage over the white matter makes direct activation of the spinal cord unlikely. To illustrate this, we quantified the proportion of rootlets relative to the white matter across spinal segments and observed substantial regional differences, with 50–60% in lower cervical segments and reaching up to 97% from L5 to S2. These anatomical realities must be considered when designing electrical spinal cord stimulation protocols, particularly when aiming for specific outcomes, such as limb activation, bladder function, or postural control. For example, while certain electrode configurations may be effective in mid-thoracic regions, they may require greater spatial precision in the sacral region due to the denser rootlet architecture and compressed geometry of the ventral columns. Furthermore, rootlet orientation should also be considered, as it may critically influence stimulation efficacy.

In the present study, we found gray matter area in the transverse plane peaking at ~5.5 mm² in the cervical (C6–C8) and ~6 mm² in the lumbar (L6–L7) regions, and white matter peaking at ~17 mm² and ~12 mm², respectively. This agrees with histological data from Veshchitskii et al. (2022) in cats. We measured the position of the gray matter relative to the white matter (e.g., distances D3–D9), as in Toossi et al. (2021), but extended this analysis from C3 to S2 (Figure 7B). In humans, the white matter reaches its maximum area (~55 mm²) at the cervical enlargement (C4–C5), decreases through the thoracic region (~25–30 mm²), and increases again in the lumbar segments (~35–40 mm² at L3–L4), while the gray matter follows a similar but smaller profile (~15–20 mm² at enlargements, <10 mm² in thoracic segments; Henmar et al., 2020).

A new class of electrodes known as circumferential spinal electrodes is emerging as a promising approach to facilitate locomotor recovery through spinal cord stimulation. These devices can be implanted either directly around the spinal cord, as in subdural applications (Woodington et al., 2024), or within the epidural space surrounding the dura mater. In this context, our anatomical data provide valuable guidance for electrode design. Using high-resolution MicroCTex imaging in four cats, we found that spinal cord circumference is greatest at cervical and lumbar levels, which correspond to the spinal enlargements, while it is notably reduced in the thoracic region. These measurements should be interpreted cautiously due to potential sample shrinkage. For electrodes placed circumferentially around the dura mater, we calculated the dural circumference using the 3D model, meaning that shrinkage is not applicable, although the model is not yet fully anatomically accurate (see Limitations section). We observed the same rostrocaudal pattern, with larger circumferences at cervical and lumbar levels and narrower dimensions in the thoracic segments. Dural circumference values ranged from 17.89 to 28 mm across segments (Figure 2E).

In this study, we provided a high-resolution, multimodal 3D reconstruction of the feline spinal cord, encompassing key structures such as the vertebrae, dura mater, cerebrospinal fluid, rootlets, dorsal root ganglia, and both white and gray matter. This model integrates data from MRI, post-mortem Micro-CT, and ex-vivo high-resolution Micro-CT, co-registered in a common reference space to ensure anatomical consistency. Such detailed anatomical fidelity not only allows for a more comprehensive understanding of spinal architecture but also bridges the gap between experimental and computational research.

Importantly, the compatibility of our model with finite element software such as Sim4Life and COMSOL demonstrates its direct applicability for testing neuromodulation strategies. While COMSOL had limitations in handling the full model with rootlets due to meshing constraints, Sim4Life successfully imported and voxelized the entire model, underscoring its suitability for studies involving electrical field distribution, thermal effects, or bioelectronic interface development.

Moreover, this open-access 3D model can serve as a foundational tool for optimizing electrode placement, designing spinal implants, or developing computational models for injury, recovery, or neuromodulation. By offering detailed morphometric data in a manipulable digital format, it paves the way for more standardized and reproducible research across laboratories working on spinal cord physiology and repair.

This study presents limitations that must be acknowledged. First, excision of the spinal cord induces a major release of its natural rostro-caudal tension, likely due to the loss of mechanical continuity with the brainstem and anchoring of the caudal nerve roots. As shown in Table 3, realignment of the ex-vivo spinal cord with the intact MRI geometry required a 35–50% linear stretch along the z-axis, indicating substantial shortening after removal from the vertebral canal. In contrast, consistent 10% corrections along the X- and Y-axes across samples most likely reflect isotropic tissue shrinkage introduced during the staining and clearing steps, particularly those involving ethanol and xylene.

The spinal cord naturally exhibits anatomical curvatures, including cervical and lumbar lordosis. However, to facilitate consistent morphometric analysis, all measurements from high-resolution images were performed on spinal cord volumes that had undergone a straightening transformation. The final 3D model integrating data from MRI, MicroCTpm and MicroCTex was likewise constructed using these geometrically corrected, straightened volumes.

The animal’s positioning during image acquisition is another factor that may have influenced the morphological measurements. In this study, MRI scans were obtained with the cat in a supine position, while MicroCTpm imaging was performed with the animal in a prone position. These two orientations can significantly affect anatomical relationships, particularly the distribution and thickness of the CSF, due to gravitational shifts of the spinal cord within the dural sac.

Although the imaging data used in this study were of high resolution, the process of anatomical segmentation could be subjective, particularly in regions with lower contrast or complex morphology. To minimize inter-operator variability and ensure consistency, the segmentation of all structures including vertebrae, dorsal root ganglia, rootlets, white matter, and gray matter was performed by the same experimenter (Harnie). This approach enhances coherence across datasets but does not fully eliminate the inherent subjectivity associated with manual segmentation.

Anatomical variability of the spinal cord related to age and sex is also expected in cats, as reported in other mammalian species, and may influence overall spinal cord dimensions as well as segmental morphology. However, the limited sample size and the relatively homogeneous characteristics of the animals included in this study did not allow for a systematic assessment of age- or sex-related effects. Future studies including larger and more diverse cohorts will be required to specifically investigate these sources of variability.

The resulting 3D spinal cord model was derived from these anatomical segmentations rather than directly from the documented morphometric measurements. While this may lead to minor differences between measured values and the reconstructed geometry, all geometric components were derived entirely from physiologically grounded imaging data of the same specimen. The model was therefore designed to provide a stable and anatomically coherent framework for spatial reference rather than to replicate exact local dimensions.