Correlation of 2D Linear and 3D Volumetric Measurement Techniques for Subchondral Bone Sclerosis in the Equine Distal Limb Using Standing Computed Tomography Authors: Annabelle C. Trainor, Benjamin J. Ahern, Ekaterina V. Strounina, Frederik E. Pauwels, Alex C. Young Published: 2026 ABSTRACT ABSTRACTSubchondral bone sclerosis of the distal third palmar/plantar metacarpal/metatarsal condyles is a common radiographic finding in Thoroughbred racehorses. Traditionally, assessment has relied on two‐dimensional (2D) linear measurements; however, advances in imaging software now allow three‐dimensional (3D) volumetric quantification of subchondral bone. This study aimed to assess the relationship and agreement between computed tomography (CT)‐derived 2D linear and volumetric measurements of condylar sclerosis and computer‐generated 3D volumetric measurements and determine whether 2D measurements could provide a practical alternative to 3D analysis. A total of 62 Thoroughbred racehorses (96 fetlock joints; 59 forelimbs, and 37 hindlimbs) underwent standing fan‐beam CT (FBCT). Retrospective image analysis was performed using Horos for 2D measurements and Volume Graphics Studio Max for 3D volumetric quantification. Associations between 2D and 3D measurements were assessed using Pearson's correlation (r) and coefficients of determination (R 2). Agreement was evaluated using Bland–Altman analysis, whereas linear mixed‐effects models identified predictors of 3D sclerosis volume. Inter‐ and intraobserver agreements were assessed using intraclass correlation coefficients (ICC). Estimated sclerosis volume derived from 2D measurements showed a strong positive correlation with 3D sclerosis volume (r = 0.79, R 2 = 0.62, 95% CI for r = 0.73–0.85). Dorsopalmar/Plantar sclerosis was strongly correlated with condylar 3D sclerosis volume (r = 0.76, R 2 = 0.58, 95% CI for r = 0.69–0.81). Interobserver (ICC = 0.969) and intraobserver (ICC = 0.982) agreement was excellent. Although 2D measurements may provide a practical alternative when 3D software is unavailable, the methods are not directly interchangeable, particularly with greater sclerosis volume. Abbreviations two dimensional three dimensional fan‐beam computed tomography third metacarpus third metatarsus Introduction Subchondral sclerosis in the equine distal limb is a well‐recognized radiological feature associated with repetitive loading in the Thoroughbred racehorse. On histopathology, subchondral bone sclerosis is defined by thickened trabeculae, reduced marrow space, and increased mineralization, representing an adaptive response to repetitive load [1, 2, 3, 4]. Sclerosis can be represented as hyperattenuation/hyperdensity on computed tomography (CT) and hypointensity on T1‐weighted (T1W) and T2‐weighted (T2W) magnetic resonance imaging (MRI). To ensure consistency throughout this article, the term sclerosis is used to represent the changes seen at a histopathological and advanced imaging level. Historically, the assessment of subchondral bone sclerosis has predominantly been described subjectively in equine distal limbs and relied on two‐dimensional (2D) linear measurements derived from postmortem and antemortem fan‐beam computed tomography (FBCT), standing cone‐beam CT (CBCT), and MRI [5, 6, 7, 8, 9, 10]. Several manually applied 2D linear measurements have been previously described to assess and quantify sclerosis in the distal third metacarpus/metatarsus (MCIII/MTIII). These included measurement of sclerosis thickness perpendicular to the articular surface in a proximodistal direction, the assessment of dorsopalmar/plantar sclerosis on a frontal plane orientated at 35° relative to the dorsopalmar/plantar plane and perpendicular to the articular surface, and evaluation of mid‐condylar and parasagittal groove sclerosis on a 45° dorsopalmar/plantar plane [5, 7, 10, 11, 12]. Categorical sclerosis grading systems have also been described on the basis of the percentage of condylar sclerosis present [5, 13]. Recently, three‐dimensional (3D) volumetric measurements have been described and utilized in both human and veterinary research [6, 14, 15, 16, 17, 18, 19]. CBCT has been used to accurately quantify the 3D volume of experimentally induced maxillary bone defects in a human model, showing strong agreement with corresponding micro‐CT measurements [16]. These findings collectively support the use of 3D volumetric analysis as a logical reference for evaluating subchondral bone sclerosis in the distal palmar/plantar metacarpo/metatarsophalangeal condyles. These developments offer new opportunities for accurate assessment and monitoring of subchondral bone sclerosis and potential for guiding management of subchondral bone pathology. Despite the use of manual 2D linear and 3D volumetric techniques across various species, direct comparisons between these methods remain limited, particularly in the context of antemortem imaging of the equine distal limb. The extent to which manual 2D linear measurements can reliably approximate true 3D sclerosis volume has not been clearly established. This knowledge is critical to validate simpler methods that are currently used clinically where 3D volumetric software may not be available. The purpose of this study was to assess manually obtained linear proximodistal, dorsopalmar/plantar, and axial–abaxial measurements as well as a manually obtained volume measurement (proximodistal × dorsopalmar/plantar × axial–abaxial) for correlation with a computer‐generated 3D volumetric measurement of condylar sclerosis. We hypothesized that the manually obtained 2D volumetric measurement would closely correlate with the computer‐generated 3D volumetric measurement, and if a single manual 2D linear measurement were to be used, proximodistal would be most reflective of 3D sclerosis volume, as sclerosis typically extends dorsoproximal from the articular surface toward the physeal scar. Methods Case Selection and Data Recording A retrospective analysis of an existing imaging dataset was conducted. Standing CT records at REC Equine Specialists, Sydney, New South Wales, Australia were searched for Thoroughbred racehorses who had a CT of the metacarpo/metatarsophalangeal joint (MCPJ/MTPJ) between August 2021 and July 2023. Subject inclusions were determined by an equine sports medicine resident. Cases were included if they were a Thoroughbred racehorse greater than 1 year of age presenting for standing CT due to suspected MCPJ/MTPJ pathology. Horses with one or more MCPJ/MTPJ imaged (forelimb and/or hindlimb) were eligible, where more than one MCPJ/MTPJ was imaged per individual horse, each MCPJ/MTPJ contributed independently. Horses were included on suspicion of MCPJ/MTPJ pathology irrespective of lameness status. Horses were excluded if they had evidence of a fracture and/or parasagittal groove fissure in the MCPJ/MTPJ, compromised image quality due to motion artifact, beam hardening artifact, or complete visualization of the distal MCIII/MTIII condyles was not achieved. CT Image Acquisition and Preparation Horses underwent standing FBCT under standing sedation. CT images of equine MCPJ/MTPJ were acquired using a gantry‐mounted 32‐slice large‐bore (90 cm) scanner (Qalibra Canon Aquilion Large Bore; Germany). Imaging acquisition parameters were kVp: 135, mA s: 400, rotation time: 0.5 s, and detector configuration: 32 × 0.5 mm2 in helical mode with an effective slice thickness of 0.5 mm. Digital images were exported and reconstructed in digital imaging and communications in medicine (DICOM) format and imported into Volume Graphics Studio Max (VG Studio Max, version 2024.2, Volume Graphics, Heidelberg, Germany) or Horos (Horos Project, United States) to evaluate 3D volumetrics and manual 2D linear measurements, respectively. Digital images were reconstructed and viewed in a bone algorithm (window width [WW] = 1500; window level [WL] = 300) and accessible in sagittal, transverse, and dorsal planes. CT Image Analysis CT images were reviewed by an ACVSMR third‐year resident who was blinded to measurements obtained by the other observer. The resident performed and recorded all manual 2D linear measurements of sclerosis. The 3D computed volume measurements were assessed by the same ACVSMR resident 6 months after 2D linear measurements were obtained. An independent postdoctoral research fellow from the Centre of Advanced Imaging at the Australian Institute for Bioengineering and Nanotechnology also reviewed the 3D‐computed volume measurements. The two observers had no restrictions on the image plane, magnification, or movement of images during examination of the 3D CT dataset. Measurements were performed at two levels (1) condylar 3D sclerosis volume, representing sclerosis volume for each individual MCIII/MTIII condyle (n = 192), and (2) total 3D sclerosis volume, representing the combined sclerosis volume of all assessed condyles within each fetlock joint (n = 96). 3D‐Computerized Volumetric Measurement 3D Image Registration and Surface Extraction CT datasets were first registered using the simple registration “Register Object” function in VG Studio Max (version 2024.2, Volume Graphics, Heidelberg, Germany) that aligns a 3D object to a common coordinate position through rotation and translation, without altering shape, to enable consistent anatomical orientation in 3D space. Registration standardizes the spatial orientation of distal MCIII/MTIII, so it lines up with another distal MCIII/MTIII, reducing positional variability. Then “Determine surface” was used to identify where the cortical bone boundary of distal MCIII/MTIII is and convert this boundary into a 3D surface that can be measured. A surface extraction threshold of greater than 370 HU and less than 2000 HU was manually implemented across all cases to define the external boundary of distal MCIII/MTIII and generate a 3D structure for analysis. A separate manually predefined threshold of 1250–1470 HU was subsequently applied to identify regions of sclerosis within the defined bone surface volume of distal MCIII/MTIII (Figure 1). The threshold interval was visually cross‐checked against the generated histogram to confirm accurate placement within the specified attenuation range (Figure 1). Within the analysis software, gray value and Hounsfield Units (HUs) are used interchangeably depending on input data format. As all datasets were reconstructed in HU, attenuation values are reported accordingly. These processes served two distinct purposes: Surface extraction established the anatomical limits of distal MCIII/MTIII, whereas the predefined threshold classified bone characteristics, namely, sclerosis, within distal MCIII/MTIII. This step was critical for accurate anatomical modeling and region‐of‐interest (ROI) segmentation. Surface extraction histogram of the palmar/plantar distal MCIII/MTIII condyle demonstrating the distribution of Hounsfield Units (HU) associated with condylar sclerosis. The x‐axis displays tissue density in HU, whereas the y‐axis tracks absolute voxel frequency (raw count) displayed on a log10 scale. The purple shaded region highlights a predefined sclerosis threshold interval (1250–1470 HU), capturing a subtle increase in voxel frequency. This represents a transitional zone of elevated attenuation of palmar/plantar condylar sclerosis occurring just before the dominant cortical bone peak (to the right of the purple region). ROI Segmentation and Thresholding A solid ROI was created around the entire distal condyle using the “ROI from volume/CAD” tool. The reconstruction was performed using a 4‐bit precision setting in VG Studio Max (version 2024.2, Volume Graphics, Heidelberg, Germany), where the CT data was processed by simplifying HUs into 16 discrete intensity levels rather than a continuous range prior to segmentation. This standardized HUs resolution reduced minor intensity variations, allowing for consistent identification and delineation of the ROI across datasets without altering the original CT acquisition data. To ensure complete coverage of the palmar/plantar condyles, the ROI was expanded using a one‐voxel dilation function, which was applied uniformly across the dataset to minimize potential under‐segmentation at cortical boundaries and standardize segmentation. To identify regions of sclerosis, a histogram‐based segmentation approach was applied (Figure 2). Using the “split interval” and “ROI from Gray Value Interval” functions within the software, the predefined threshold of 1250–1470 HU, corresponding to increased bone density, remained in place for the isolation of sclerotic subchondral bone. This method allowed objective, density‐based extraction of sclerosis regions, minimizing user subjectivity, and improving repeatability. Segmentation workflow for 3D quantification of subchondral bone sclerosis. A solid ROI was generated around the distal condyle and expanded using dilation to ensure complete coverage. Sclerotic regions were identified using a manual threshold (1250–1470 HU) and visually cross checked against the generated histogram (bottom right image) to confirm accurate placement. ROIs were manually refined to remove any sclerosis that was present on the dorsal aspect of distal MCIII/MTIII. A minimum threshold of 1000 HU was applied to exclude artifactual or clinically irrelevant regions. Between 5 and 8 ROIs were generated automatically by the software per case. When a threshold (1250–1470 HUs) was applied, the software identified all disconnected groups of HUs that met the threshold criteria. Therefore, the ROI corresponding to the distal MCIII/MTIII must be selected by the operator on the basis of anatomical location and retained for further analysis. All remaining ROIs were excluded from further analysis. Minor adjustments were made using the “Pencil” tool to remove any dorsal condylar sclerosis if present. To allow separate analysis of the medial and lateral condyles, the distal MCIII/MTIII ROI was duplicated. Each copy was then partitioned into medial and lateral components using a consistent anatomical midline reference (intercondylar groove) in multiplanar reconstructions. This ensured that both analyses were performed on the same initial segmented ROI of distal MCIII/MTIII, with only the ROI isolated for each condyle. Regions of demineralization within the region of sclerosis were included in the volume calculations to ensure the entire condyle volume was captured. Exclusion of regions of demineralization would have required additional manual segmentation, introducing greater variability. Dorsal condylar sclerosis was excluded to maintain anatomical consistency and focus analysis on palmar/plantar pathology of MCIII/MTIII. Final 3D Volumetric Measurement Volumetric data were extracted using the “Object Properties” tool that quantified the volume in cubic millimeters (mm3) of the segmented regions of subchondral bone sclerosis. Measurements were recorded separately for the medial and lateral condyles and then combined to calculate total sclerosis volume per joint. These values formed the basis for further statistical correlation against manual 2D linear measurements. Linear Measurements CT scans of the MCPJ/MTPJ were assessed using Horos DICOM viewer software (Horos Project, United States). Three‐dimensional multiplanar reconstruction (3D MPR) was performed using a bone window level setting (WL = 300, WW = 1500) to optimize visualization of subchondral bone sclerosis. The sclerosis of each condyle was independently assessed using two differently oriented images: a 45° dorsopalmar/plantar plane and a parasagittal plane. The 45° dorsopalmar/plantar plane image was acquired first, aligned perpendicular to the articular surface of the palmar/plantar aspect of the distal MCIII/MTIII bone as previously described in the literature [5, 7], and standardized by passing through the center of the distal physeal scar (Figure 3). These images were then reviewed from dorsal to palmar until the point of maximum sclerosis of either the lateral or medial condyle was visually identified. The corresponding parasagittal image was subsequently obtained through the point of maximum sclerosis for each condyle. Proximodistal sclerosis was measured visually on the parasagittal image in a proximodistal direction, defined as the distance from the articular surface to the most proximal extent of sclerosis propagating in the direction of the physeal scar of distal MCIII/MTIII (Figure 3). Dorsopalmar/plantar sclerosis was measured on the same parasagittal image, in a dorsopalmar/plantar direction (Figure 3). Axial–abaxial sclerosis was measured on the 45° dorsopalmar/plantar frontal plane image in a mediolateral direction (Figure 3). Individual measurements were recorded for medial and lateral condyles when biaxial condylar sclerosis was present. Manual 2D Linear measurements of sclerosis. The proximodistal extent of sclerosis is illustrated in (A) (parasagittal plane), the dorsopalmar/plantar extent in (B) (parasagittal plane), and the axial–abaxial extent in (C) (frontal plane). A simplified volumetric estimate of sclerosis was calculated by multiplying the individual linear dimensions of proximodistal, dorsopalmar/plantar, and axial–abaxial, expressed in cubic millimeters (mm3). Observer Repeatability A random subset of 40 condyles from the total study population of 192 condyles was selected using a computer‐generated random sequence in R statistical software (version 4.4.3) for repeatability analysis of condylar 3D sclerosis volume, total 3D sclerosis volume, and 2D linear measurements. Interobserver repeatability of 3D volumetric measurements was assessed independently by two observers (ACVSMR resident and a postdoctoral research fellow), each blinded to the other observers’ measurements. Intraobserver repeatability was assessed by repeat measurements of all individual 2D linear variables (dorsopalmar/plantar, proximodistal, and axial–abaxial) and volume by the same observer (ACVSMR resident) after a washout period of 12 months, with the observer blinded to the original measurements. Statistical Analysis Statistical analysis was performed by a biostatistician using R software (version 4.4.3, released 2025‐02‐28) and conducted in RStudio (version 2024.12.1.563). Continuous variables were summarized using descriptive statistics depending on distribution. Associations between 2D linear measurements and 3D sclerosis volumes were assessed using Pearson's correlation coefficients with corresponding 95% confidence intervals (CI). Linear regression analysis was subsequently performed to quantify the relationship between variables and determine the proportion of variance explained (R 2). Bland–Altman analysis was performed to assess agreement between 2D and 3D volumetric measurements, as correlation analysis alone does not evaluate systematic bias or interchangeability of two different measurement techniques. The method was used to quantify proportional bias and limits of agreement between the two different techniques, enabling assessment of clinical agreement. Given the hierarchical structure of the data, with condyles within horses, linear mixed effects models with horses included as a random intercept were used to account for clustering and non‐independence of observations. Standardized fixed‐effects coefficients from these models were used to compare the relative contribution of each 2D measurement to 3D sclerosis volume. To compare the strength of correlations between non‐independent variables measured within the same sample and against a shared outcome variable, Steiger's test for dependent correlations was applied. Scatterplots were visually inspected to assess linearity and identify potential outliers prior to analysis. Intra‐ and interobserver agreement for 2D linear and 3D volumetric measurements, respectively, was assessed using two‐way mixed‐effects intraclass correlation coefficients (ICC) for absolute agreement with corresponding 95% CI. ICC values were interpreted as poor (<0.50), moderate (0.50–0.75), good (0.75–0.90), or excellent (>0.90). Bland–Altman analysis was additionally performed to evaluate systematic bias and limit of agreement between measurements. All statistical tests were two‐sided, and a p value of <0.05 was considered statistically significant. Results A total of 62 Thoroughbred racehorses and 96 fetlocks (59 forelimb and 37 hindlimb) underwent standing FBCT. Descriptive statistics for 3D and 2D sclerosis volume and 2D linear sclerosis measurements are summarized in Table 1. The data are presented as mean ± standard deviation, median, range, stratified by limb (LF, RF, LH, RH), and condyle (medial, lateral). Mean condylar 3D sclerosis volume ranged from 2269 to 3154 mm3, whereas mean 2D condylar sclerosis volume ranged from 2175 to 3148 mm3. Medial condyles generally demonstrated greater mean 3D and 2D sclerosis volumes than lateral condyles in the forelimbs. In the hindlimbs, the left hindlimb also demonstrated greater mean 3D and 2D sclerosis volume in the medial condyle; however, the lateral condyle showed greater mean 3D and 2D sclerosis volume in the right hindlimb. The 2D linear measurements showed a mean dorsopalmar/plantar sclerosis range from 9.4 to 12.2 mm, mean proximodistal sclerosis from 14.4 to 17.8 mm, and mean axial–abaxial sclerosis measurements from 12.2 to 14.4 mm. TABLE 1Descriptive statistics of three‐dimensional (3D) sclerosis volume (mm3) and two‐dimensional (2D) linear sclerosis measurements (mm), stratified by limb (LF, RF, LH, RH) and condyle (medial, lateral).VariableLF medial (n = 33)LF lateral (n = 33)RF medial (n = 26)RF lateral (n = 26)LH medial (n = 15)LH lateral (n = 15)RH medial (n = 22)RH lateral (n = 22)3D volume (mm3)Mean 3154 ± 1703Mean 2442 ± 983Mean 2648 ± 1376Mean 2351 ± 1087Mean 2868 ± 1330Mean 2622 ± 1026Mean 2269 ± 1147Mean 2559 ± 1389Median 2963 (956–7596)Median 2229 842–4474)Median 2778 (518–5848)Median 2208 (488–5092)Median 2527 (1413–5685)Median 2467 (1266–5153)Median 1860 (1081–5561)Median 2244 (862–5945)2D volume (mm3)Mean 3148 ± 2369Mean 2198 ± 1576Mean 2742 ± 1892Mean 2285 ± 1586Mean 3229 ± 2348Mean 2698 ± 1386Mean 2175 ± 2157Mean 2436 ± 1611Median 2991 (237–9302)Median 1896 (205–6638)Median 2309 (209–8333)Median 1525 (116–6091)Median 2354 (920–8754)Median 2593 (912–5342)Median 1174 (203–7964)Median 1927 (591–6109)DP (mm)Mean 11.3 ± 5.43Mean 8.69 ± 3.45Mean 10.8 ± 4.51Mean 9.84 ± 4.35Mean 11.5 ± 5.15Mean 12.2 ± 3.81Mean 9.4 ± 5.34Mean 11.4 ± 4.88Median 12.1 (3.35–24.7)Median 8.01 (3.24–16.8)Median 9.98 (3.09–22.4)Median 8.43 (3.02–18.8)Median 11.9 (4.77–18.6)Median 12.7 (5.44–18.2)Median 7.1 (2.21–21)Median 11.5 (4.27–19.8)PD (mm)Mean 16.7 ± 4.76Mean 16.7 ± 4.15Mean 17 ± 4.29Mean 17 ± 4.23Mean 17.8 ± 3.02Mean 17.3 ± 4.12Mean 14.4 ± 4.01Mean 16 ± 3.4Median 17 (5.45–26.2)Median 16.3 (4.32–25)Median 16.6 (6.1–26.2)Median 17.1 (6.74–23.6)Median 18.0 (12.6–22.6)Median 19.2 (10.1–23.3)Median 13.7 (9.23–25.2)Median 15.9 (9.69–21.4)AA (mm)Mean 14.4 ± 2.32Mean 12.6 ± 3.13Mean 13.4 ± 2.63Mean 12.2 ± 2.39Mean 14 ± 3.23Mean 12.3 ± 1.91Mean 13.2 ± 2.29Mean 12.3 ± 1.70Median 14.5 (8.69–20.94)Median 12.5 (3.32–19.19)Median 13.5 (4.79–19.58)Median 12.4 (5.69–17.17)Median 14.4 (6.89–21.29)Median 12.5 (8.21–15.93)Median 12.8 (9.94–19.69)Median 12.1 (9.04–16.95) Note: Data are presented as mean ± standard deviation, median (range).Abbreviations: AA, axial–abaxial; DP, dorsopalmar/plantar; LF, left forelimb; LH, left hindlimb; PD, proximodistal; RF, right forelimb; RH, right hindlimb.John Wiley & Sons, Ltd. Descriptive statistics of three‐dimensional (3D) sclerosis volume (mm3) and two‐dimensional (2D) linear sclerosis measurements (mm), stratified by limb (LF, RF, LH, RH) and condyle (medial, lateral). Note: Data are presented as mean ± standard deviation, median (range). Abbreviations: AA, axial–abaxial; DP, dorsopalmar/plantar; LF, left forelimb; LH, left hindlimb; PD, proximodistal; RF, right forelimb; RH, right hindlimb. A strong positive correlation was identified between 2D linear and condylar 3D sclerosis volume (r = 0.79, R 2 = 0.62, 95% CI for r = 0.73–0.85, p < 0.001) (Figure 4). A moderate correlation was observed between total 3D sclerosis volume and 2D linear volumes (r = 0.60, R 2 = 0.36, 95% CI for r = 0.46–0.71, p < 0.001). Scatterplot illustrating the relationship between total 3D volume of condylar sclerosis and total 2D linear sclerosis volume. Each point represents a single condyle. The blue line indicates the linear regression fit, with the coefficient of determination (R 2) annotated to show the strength of correlation. 2D, two dimensional; 3D, three dimensional. Across individual 2D measurements (proximodistal, dorsopalmar/plantar, and axial–abaxial), all variables showed significant associations with condylar 3D sclerosis volume (Tables 2, 3, 4, 5). Dorsopalmar/plantar measurement showed the strongest association with condylar 3D sclerosis volume (r = 0.76, R 2 = 0.58, 95% CI for r = 0.69–0.81) (Figure 5), followed by proximodistal measurements (r = 0.63, R 2 = 0.40, 95% CI for r = 0.53–0.71) (Figure 6), and axial–abaxial measurements (r = 0.50, R 2 = 0.25 and 95% CI for r = 0.38–0.60) (Figure 7). TABLE 2Pearson's correlation coefficients and coefficients of determination between two‐dimensional (2D) linear and three‐dimensional (3D) sclerosis measurements.Condylar 3D sclerosis volumeTotal 3D sclerosis volumeCharacteristic r R 2 95% CI p value r R 2 95% CI p value2D volume of sclerosis (mm3)0.790.620.73–0.85<0.0010.600.360.46–0.71<0.001Dorsopalmar/Plantar sclerosis (mm)0.760.580.69–0.81<0.0010.610.370.47–0.72<0.001Proximodistal sclerosis (mm)0.630.400.53–0.71<0.0010.550.300.39–0.67<0.001Axial–abaxial sclerosis (mm)0.500.250.38–0.60<0.0010.390.150.21–0.54<0.001Abbreviation: CI, confidence interval.John Wiley & Sons, Ltd. Pearson's correlation coefficients and coefficients of determination between two‐dimensional (2D) linear and three‐dimensional (3D) sclerosis measurements. Abbreviation: CI, confidence interval. TABLE 3Bland–Altman analysis of agreement between two‐dimensional (2D) and three‐dimensional (3D) volumetric measurement of palmar/plantar condylar sclerosis.ComparisonBias (mm3)95% LoA (mm3) β (proportional bias) R 2 2D volume vs. 3D volume−29.74−2372.5 to 2313.30.42 (p < 0.001)0.29 Notes: Bland–Altman analysis was used to assess agreement between methods. Mean bias represents the average difference between 2D and 3D volumetric measurements (2D–3D). Limits of agreement (LoA) were calculated as mean bias ± 1.96 standard deviations of the differences. Proportional bias was assessed using linear regression of the differences against the mean of the two methods. Statistical significance was set at p < 0.05.Abbreviations: LoA, limits of agreement.John Wiley & Sons, Ltd. Bland–Altman analysis of agreement between two‐dimensional (2D) and three‐dimensional (3D) volumetric measurement of palmar/plantar condylar sclerosis. Notes: Bland–Altman analysis was used to assess agreement between methods. Mean bias represents the average difference between 2D and 3D volumetric measurements (2D–3D). Limits of agreement (LoA) were calculated as mean bias ± 1.96 standard deviations of the differences. Proportional bias was assessed using linear regression of the differences against the mean of the two methods. Statistical significance was set at p < 0.05. Abbreviations: LoA, limits of agreement. TABLE 4Standardized fixed‐effects coefficients from a linear mixed‐effects model for predictors of condylar three‐dimensional (3D) sclerosis volume.Condylar 3D sclerosis volumeCharacteristicStandardized β SE95% CI t valueInterpretation2D sclerosis volume0.3410.110.13–0.563.07StrongDorsopalmar/Plantar sclerosis0.2850.090.11–0.463.30StrongProximodistal sclerosis0.1770.060.06–0.292.85ModerateAxial–abaxial sclerosis0.1490.050.05–0.252.88Moderate Notes: Condylar 3D sclerosis volume was modeled as the dependent variable, with 2D sclerosis volume and 2D linear measurements (dorsopalmar/plantar, proximodistal, and axial–abaxial) included as fixed effects. Horse was included as a random intercept to account for clustering of condyles within horses. All variables were standardized prior to analysis. β coefficients represent the independent association of each predictor with 3D sclerosis volume after adjustments for other predictors and clustering within horse.Abbreviations: 2D, two dimensional; SE, standard error.John Wiley & Sons, Ltd. Standardized fixed‐effects coefficients from a linear mixed‐effects model for predictors of condylar three‐dimensional (3D) sclerosis volume. Notes: Condylar 3D sclerosis volume was modeled as the dependent variable, with 2D sclerosis volume and 2D linear measurements (dorsopalmar/plantar, proximodistal, and axial–abaxial) included as fixed effects. Horse was included as a random intercept to account for clustering of condyles within horses. All variables were standardized prior to analysis. β coefficients represent the independent association of each predictor with 3D sclerosis volume after adjustments for other predictors and clustering within horse. Abbreviations: 2D, two dimensional; SE, standard error. TABLE 5Comparison of dependent correlations between three‐dimensional (3D) condylar sclerosis volume and two‐dimensional (2D) condylar volume and linear sclerosis measurements (dorsopalmar/plantar, proximodistal, and axial–abaxial) using Steiger's test.Comparison with 3D volume–2D volume correlationCorrelation comparison z p valueInterpretationDorsopalmar/Plantar vs. 3D volume1.350.178Not significantly differentProximodistal vs. 3D volume4.76<0.001Significantly weakerAxial–abaxial vs. 3D volume6.49<0.001Significantly weaker Note: z = test statistic from Steiger's test for comparing dependent correlations. Positive z values indicate that the correlation between 3D and 2D sclerosis volume was stronger than the comparator correlation. Statistical significance was defined as p < 0.05.John Wiley & Sons, Ltd. Comparison of dependent correlations between three‐dimensional (3D) condylar sclerosis volume and two‐dimensional (2D) condylar volume and linear sclerosis measurements (dorsopalmar/plantar, proximodistal, and axial–abaxial) using Steiger's test. Note: z = test statistic from Steiger's test for comparing dependent correlations. Positive z values indicate that the correlation between 3D and 2D sclerosis volume was stronger than the comparator correlation. Statistical significance was defined as p < 0.05. Scatterplot depicting the relationship between linear dorsopalmar/plantar sclerosis measurements and total 3D volume of condylar sclerosis. Each point represents an individual condyle. The blue line indicates the linear regression fit, with the coefficient of determination (R 2) annotated to show the strength of correlation. 3D, three dimensional. Scatterplot illustrating the relationship between linear proximodistal sclerosis measurements and 3D volume of condylar sclerosis. Each point represents an individual condyle. The blue line indicates the linear regression fit, with the coefficient of determination (R 2) annotated to show the strength of correlation. 3D, three dimensional. Scatterplot illustrating the relationship between linear axial–abaxial sclerosis measurements and 3D volume of condylar sclerosis. Each point represents an individual condyle. The blue line indicates the linear regression fit, with the coefficient of determination (R 2) annotated to show the strength of correlation. 3D, three dimensional. Bland–Altman analysis was performed to assess agreement between 2D and condylar 3D sclerosis volume measurements (Table 3, Figure 8). The mean bias was −29.74 mm3. The limits of agreement were wide, ranging from −2372.75 to 2313.28 mm3. Linear regression of the differences against the mean demonstrated a significant proportional bias (p < 0.001), with a coefficient of determination of R 2 = 0.29. Bland–Altman analysis comparing 2D and 3D measurements of sclerosis volume. The plot displays the mean of 2D and 3D sclerosis volume (mm3) against the difference between methods (2D–3D). The solid blue line represents the mean bias (−29.74 mm3), whereas the dashed red lines indicate the 95% limits of agreement (−2372.5 to 2313.28 mm3 ± 1.96 × standard deviation of the differences). 2D, two dimensional; 3D, three dimensional. Standardized fixed‐effects coefficients from the linear mixed‐effects model are presented in Table 4. After adjustment for the remaining predictors and accounting for clustering of condyles within horses, 2D sclerosis volume remained the strongest independent predictor of 3D sclerosis volume (β = 0.341), followed by dorsopalmar/plantar (0.285), proximodistal, and axial–abaxial measurements. Comparisons between dependent correlation coefficients using Steiger's test are presented in Table 5. The association between dorsopalmar/plantar measurements and condylar 3D sclerosis volume did not differ significantly from that of total 2D sclerosis volume (z = 1.35, p = 0.178), whereas proximodistal and axial–abaxial measurements demonstrated significantly weaker associations (z = 4.76 and 6.49, respectively; both p < 0.001). Intra‐ and interobserver agreement for 2D and 3D sclerosis measurements was assessed using a randomly selected subset of 40 condyles from the total dataset of 192 condyles. Intraobserver reliability for 2D linear and volumetric measurements is summarized in Table 6. Volume of 2D sclerosis measurements showed excellent agreement (ICC = 0.982; 95% CI: 0.967–0.991), with minimal systematic bias (61.95 mm3; 1.64%), with 95% limits of agreement ranging from −640.45 to 764.35 mm3. Individual 2D linear measurements also demonstrated high repeatability, with negligible mean differences for dorsopalmar/plantar (0.11 mm; 0.82%), proximodistal (0.07 mm; 0.38%), and axial–abaxial (−0.05 mm; −0.37%) measurements. Limits of agreement were narrow for dorsopalmar/plantar (−0.45 to 0.67 mm) and proximodistal (−1.19 to 1.33 mm) measurements and slightly wider for axial–abaxial measurement (−1.86 to 1.75 mm). Bland–Altman analysis demonstrated minimal systematic bias across all 2D volumetric and linear measurements, with no evidence of clinically meaningful proportional error. TABLE 6Intraobserver reliability of two‐dimensional (2D) linear sclerosis measurements assessed using intraclass correlation coefficients (ICC) and Bland–Altman analysis.ParameterICC (95% CI) p valueMean difference (Bias)% BiasSD differenceLower 95% LoAUpper 95% LoA2D volume (mm3)0.982 (0.97–0.99)<0.00161.951.64358.37−640.45764.35Dorsopalmar/Plantar (mm)0.997 (0.99–0.99)<0.0010.110.820.29−0.450.67Proximodistal (mm)0.982 (0.967–0.991)<0.0010.070.380.64−1.191.33Axial–abaxial (mm)0.947 (0.902–0.971)<0.001−0.05−0.370.92−1.861.75Abbreviation: LoA, limits of agreement.John Wiley & Sons, Ltd. Intraobserver reliability of two‐dimensional (2D) linear sclerosis measurements assessed using intraclass correlation coefficients (ICC) and Bland–Altman analysis. Abbreviation: LoA, limits of agreement. Interobserver agreement for 3D sclerosis volume measurements is summarized in Table 7. Interobserver reliability was excellent (ICC = 0.969; 95% CI: 0.942–0.983), with minimal systematic bias (5.4 mm3) and narrow limits of agreement (−685.1 to 695.9 mm3). The mean absolute interobserver difference was 274.5 mm3, corresponding to a mean percentage difference of 8.2%. TABLE 7Interobserver agreement of three‐dimensional (3D) volumetric sclerosis measurements assessed using intraclass correlation coefficients (ICC) and Bland–Altman analysis.ParameterICC (95% CI) p valueMean difference (Bias)Mean difference (%)Mean absolute differenceSD differenceLower 95% LoAUpper 95% LoA3D volume (mm3)0.969 (0.942–0.983)<0.0015.48.2274.5352.3−685.1695.9Abbreviation: LoA, limits of agreement.John Wiley & Sons, Ltd. Interobserver agreement of three‐dimensional (3D) volumetric sclerosis measurements assessed using intraclass correlation coefficients (ICC) and Bland–Altman analysis. Abbreviation: LoA, limits of agreement. Discussion The presence and severity of subchondral bone sclerosis in Thoroughbred palmar/plantar distal MCIII/MTIII is of increasing interest to the equine veterinary community and racing industry, particularly in the context of chronic stress remodeling resulting in performance limiting pathologies and prodromal changes that could indicate increased risk for fracture [5, 10, 11, 20, 21, 22, 23, 24, 25]. Sclerosis within the palmar/plantar MCIII/MTIII condyles in racing Thoroughbreds has been shown to arise from stress‐induced microfracture and cumulative fatigue damage. During high speed training and racing conditions, the distal condyles of MCIII/MTIII are subjected to repetitive cyclical loading that generates high compressive and shear forces at the articular surface and subchondral bone [8, 25, 26, 27, 28]. Over time, this mechanical demand may exceed the bone's capacity for micro‐damage repair, leading to the accumulation of microcracks within the subchondral bone plate and adjacent trabecular bone. These microstructural changes represent early features of structural fatigue injury, whereby the accumulation of microfracture and incomplete repair initiates maladaptive subchondral bone remodeling and sclerosis. Sclerosis forming in the Thoroughbred palmar/plantar distal MCIII/MTIII condyles as a response to chronic stress remodeling typically develops in a dome to conical/triangular shape, extending proximally from the subchondral bone toward the physeal scar region [5]. It is therefore logical that an obliquely oriented frontal plane image would enable more accurate measurement of the proximodistal sclerosis as it extends toward the physeal scar region. This is in contrast to a true frontal plane image aligned with the cortices of the MCIII/MTIII diaphysis, which research has indicated, may underestimate proximodistal sclerosis [7]. Previous studies have suggested that sclerosis is most frequently identified within the palmar/plantar aspect of the condyles, consistent with site‐specific loading patterns [4, 29]. CT and histological studies further demonstrate regional variation in subchondral sclerosis density across dorsal and palmar regions, supporting a nonuniform distribution of sclerosis within the condyle [4, 29, 30]. The present study demonstrated that CT‐derived 2D linear measurements of subchondral bone sclerosis were strongly associated with condylar 3D sclerosis volume within the distal MCIII/MTIII condyles. Among the evaluated linear measurements, dorsopalmar/plantar sclerosis demonstrated the strongest relationship with condylar 3D sclerosis volume (r = 0.76, R 2 = 0.58, 95% CI for r = 0.69–0.81), exceeding proximodistal and axial–abaxial measurements. It was initially hypothesized that proximodistal sclerosis would demonstrate the strongest relationship with condylar 3D sclerosis volume, given the expected propagation of sclerosis within the condyle. However, this was not supported by the findings of the present study. The superiority of the linear dorsopalmar/plantar sclerosis measurement may reflect both the geometric distribution of condylar sclerosis and the interaction between lesion orientation and imaging plane. Previous studies have used 45° angle dorsal oblique reconstructions vertical to the MCIII/MTIII, others perpendicular to the articular surface of the palmar/plantar aspect of the condyle for assessing subchondral bone changes [5, 7, 31]. These examples show that although there is great interest in the industry regarding the presence and severity of condylar sclerosis, we do not yet have a standardized method for assessing the severity of this pathology. In the present study, dorsopalmar/plantar measurements derived from parasagittal reconstructions showed the strongest and most consistent association with condylar 3D sclerosis volume. Dorsopalmar/plantar measurements obtained from sagittal or parasagittal reconstructions likely perform better because they capture a broader representation of sclerosis across the condyle and are less sensitive to slice selection and localized variation. As a result, the dorsopalmar/plantar measurement more closely approximates total 3D sclerosis volume. In contrast, proximodistal and axial–abaxial measurements may be more sensitive to localized changes in slice position and misrepresent focal regions of sclerosis impacting overall sclerosis volume. Variation in subchondral bone sclerosis and sclerosis distribution across medial and lateral condyles has previously been demonstrated using histological techniques, supporting the concept that sclerosis is not uniformly distributed at the condylar level [29, 30]. Although dorsopalmar/plantar measurements demonstrated the strongest association among linear measurements, the relationship between 2D measurements and condylar 3D sclerosis volume requires careful interpretation. Correlation describes the strength of association between measurements but does not demonstrate equivalence or interchangeability. In the present study, 2D sclerosis volume showed the strongest association with condylar 3D sclerosis volume (r = 0.79, R 2 = 0.62; 95% CI for r = 0.73–0.85), indicating that a substantial proportion of variability in condylar 3D sclerosis volume was explained by 2D volumetric assessment. Furthermore, approximately 38% of variability remained unexplained, suggesting that 3D volumetric assessment captures additional spatial information not fully represented by 2D techniques. Bland–Altman analysis further demonstrated minimal fixed bias but wide limits of agreement with significant proportional bias indicating increased disagreement at higher sclerosis volumes. These findings indicate that 2D and 3D measures of sclerosis volume are strongly related but not directly interchangeable across the entirety of the continuum for sclerosis. The mixed‐effects model further supported this interpretation, with 2D sclerosis volume remaining the strongest independent predictor of condylar 3D sclerosis volume (β = 0.341), whereas dorsopalmar/plantar sclerosis demonstrated the strongest effect among individual linear measurements. Overall, these findings suggest that dorsopalmar/plantar measurements provide the most informative linear representation of condylar 3D sclerosis volume when 3D volumetric software is not feasible but do not replace 3D volumetric assessment where precise quantification is required. The strong inter‐ and intraobserver agreement observed for 3D volumetric and 2D linear sclerosis measurements supports the reproducibility of this methodology. Excellent intraobserver reliability (ICC = 0.982), together with minimal systematic bias and narrow limits of agreement on Bland–Altman analysis, indicates measurement stability across repeated evaluations. Interobserver agreement was similarly strong (ICC = 0.969) with low systematic bias and a percentage error of 8.2%. Although minor variability was observed, this is most likely attributable to inherent challenges in delineating the transitional boundary between sclerotic and trabecular bone, which often exhibits gradual rather than discrete changes in attenuation. Although a fixed HU threshold was applied during segmentation, partial volume effects and local heterogeneity in bone density may result in individual voxel variability crossing inclusion thresholds depending on their spatial composition. Additionally, small differences in ROI placement and the complex 3D morphology of condylar sclerosis may contribute to minor variation in the sampled voxel population. Importantly, the magnitude of this variability was small relative to the overall range of sclerosis values observed, suggesting that observed differences in condylar 3D sclerosis volume are likely to reflect true biological variation rather than a systematic measurement error. The clinical relevance of these findings is further highlighted by previous work investigating the relationship between sclerosis measurements and fracture risk. Tranquille et al. have shown that on 35° palmar/plantar oblique plane high field MRI images, each millimeter increase in proximodistal sclerosis is associated with almost a fourfold increase in the odds of a lateral condylar fracture, with a threshold of ≥16 mm identified as clinically significant [10]. In the present study, regression modeling estimated that a proximodistal sclerosis measurement of 16 mm corresponds to an estimated condylar 3D sclerosis volume of approximately 2509 mm3. However, this estimate is derived from CT imaging‐based associations within a retrospective dataset and does not incorporate clinical outcome data. Therefore, the approximated condylar 3D sclerosis volume should not be interpreted as a validated risk threshold. Rather, this value provides a preliminary translational comparison between established 2D clinical measurements and 3D volumetric assessment. Whether volumetric assessment offers improved predictive ability for fracture risk compared with existing linear measurements cannot be determined from the present study and requires prospective validation incorporating clinical outcomes. Nonetheless, these findings support further investigation into the potential role of 3D CT‐derived metrics in refining risk stratification for condylar fracture. Clinically, these findings highlight the importance of careful evaluation of sagittal and parasagittal CT reconstructions when assessing condylar sclerosis. Although frontal and transverse images remain valuable for lesion characterization, dorsopalmar/plantar measurements derived from sagittal or parasagittal reconstructions may be less susceptible to orientation‐related variability and may provide the most reliable assessment for the extent of sclerosis. One of the benefits of CT as a cross‐sectional imaging modality for assessment of racehorse condyles is that it allows the observer to slice the anatomy in multiple orientations after the images have been obtained. This is in contrast to low‐field MRI that requires image orientation to be predetermined and assigned at the prescription stage of the examination. The use of additional sequences to cover the anatomy and pathology in the area of interest, therefore, requires additional acquisition time and increases the length of the overall examination. This further highlights the importance of image orientation and standardization across the various cross‐sectional imaging modalities, not only at the stage of acquisition but also at the stage of image review and interpretation. The observation of greater mean sclerosis in the medial condyle within our study aligns with known regional heterogeneity in subchondral bone sclerosis between the medial and lateral condyles of distal MCIII/MTIII. Previous studies have demonstrated nonuniform distribution of bone density across medial and lateral condyles [29, 30]. These findings suggest that both medial and lateral condyles of distal MCIII/MTIII are subject to region‐specific adaptive remodeling rather than uniform structural change. This study had several limitations. Its retrospective design and the population included actively training or racing Thoroughbred racehorses, in a single jurisdiction and obtained at a single timepoint, limiting generalization to other equine populations and the ability to assess dynamic changes in sclerosis over time. Although a standardized imaging plane was used to improve consistency, small differences in imaging orientation or measurement placement could impact the accuracy of 2D linear measurements, particularly for proximodistal and axial–abaxial measurements. 3D volumetric quantification relies on semi‐automated segmentation, which, although more objective than manual methods, may still be influenced by software threshold settings and the operator's interpretation of sclerotic margins. This could contribute to variability in the volumetric data, especially in early or subtle cases of sclerosis. Although sclerosis was appreciable on FBCT, histological confirmation was not performed; therefore, the biological significance of sclerosis remains presumptive. The simplified geometric approach used to derive 2D sclerosis volume may not fully capture the complex and variable 3D morphology of condylar sclerosis, which often exhibits individual variation in morphology. Inter‐ and intraobserver variability for 2D linear and 3D volumetric measurements were assessed on a randomly selected subset of 40 out of 192 condyles included in the study population. This sampling approach was intended to provide a representative assessment of measurement reliability while maintaining feasibility for repeated measurements and observer evaluation across a large dataset. The authors acknowledge that this approach may reduce precision for estimated limits of agreement despite appropriate sampling. In conclusion, CT‐derived 2D linear measurements of subchondral bone sclerosis demonstrate strong associations with 3D volumetric assessment in the distal palmar/plantar MCIII/MTIII condyles of Thoroughbred racehorses. Among these, dorsopalmar/plantar measurements provided the most consistent representation of condylar 3D sclerosis volume across statistical approaches. Although 3D volumetric analysis remains the most comprehensive method for sclerosis volume quantification, its limited availability and increased time required to undertake the analysis restrict routine clinical use. Dorsopalmar/plantar measurements demonstrated strong and consistent associations with condylar 3D sclerosis volume and may provide a practical 2D linear alternative when volumetric analysis is not available, highlighting the value of simple, reproducible linear measurements for monitoring condylar sclerosis in Thoroughbred racehorses. However, given the presence of proportional bias and wide limits of agreement, 2D and 3D volumetric measures of sclerosis should not be considered interchangeable, particularly in condyles that have greater volumes of sclerosis. Author Contributions Annabelle C. Trainor: conceptualization, methodology, data curation, investigation, formal analysis, writing – original draft, writing – review and editing, validation, visualization, project administration. Benjamin J. Ahern: methodology, supervision, writing – review and editing, validation, funding acquisition. Frederik E. Pauwels: conceptualization, methodology, writing – review and editing, validation. Alex C. Young: conceptualization, methodology, validation, writing – review and editing, supervision. Ekaterina V. Strounina: methodology, software, validation, visualization, formal analysis, resources. Disclosure The authors have nothing to report. Conflicts of Interest The authors declare no conflicts of interest.