STATISTICAL SHAPE MODELING OF COMPUTED TOMOGRAPHY-DERIVED CARPAL BONES REFLECTS SCAPHOLUNATE INTEROSSEOUS LIGAMENT INJURY

Abstract

INTRODUCTION

Injuries to the scapholunate interosseous ligament (SLIL) are among the most common upper extremity injuries. Early, accurate diagnosis is essential to minimize progression of scapholunate advanced collapse (SLAC)-pattern radiocarpal OA ¹. SLIL injuries widen the SL interval and contribute to the scaphoid palmar flexion and lunate extension of dorsal intercalated segment instability ². Radiographs are limited by carpal overlap and sensitivity to forearm pronosupination angle ³. Volumetric imaging-derived three-dimensional (3D) bone models can be used in statistical shape modeling (SSM) to compare joint alignment and morphology, mitigating challenges of planar imaging.

OBJECTIVE

The objective of this study is to compare 3D carpal alignment in wrists with and without SLIL injury using a multi-level (shape and alignment), multi-object (three bone) (MLMO) SSM. We hypothesize that (1) there will be differences in the 3D morphology of the radius, scaphoid, and lunate between wrists with versus without SLIL injury and (2) these differences will affect joint space width.

METHODS

Twenty-one participants (14.3% female, median [25^th-75^th percentile] age 42.0 [26.8-50.0] years, 57.1% dominant hand injury) with arthroscopically-confirmed, unilateral SLIL injuries were recruited to a prospective clinical trial evaluating the role of CT in detecting SLIL injuries ⁴. Bilateral wrist CT images were acquired (SOMATOM Force and NAEOTOM Alpha, Siemens Healthineers, Germany) using published acquisition parameters ⁴. The radius, scaphoid, and lunate were segmented from static CT with semi-automated algorithms (Analyze Pro, Mayo Foundation for Medical Education and Research, Rochester, MN). Segmentation maps were used to generate 3D stereolithography meshes of each bone. Left-handed images were reflected to right-handed anatomies. MLMO SSM was performed (ShapeWorks v6.3.2 ⁵) ⁶. Linear discriminant analysis (LDA), a form of supervised machine learning for dimensionality reduction and class separation, was used to compare uninjured and injured morphologies ⁷. Discriminant scores between wrists were compared with a Wilcoxon signed rank test. SSM-derived bone surface particles of the mean uninjured and mean injured bones were used to calculate interosseous proximities, a metric approximating joint space width, at the SL interval and radioscaphoid joint using k-nearest neighbor methods within distance thresholds of 5.0 mm and 2.5 mm, respectively ^8,9. Interosseous proximity distributions were compared using two-sided Kolmogorov-Smirnov (KS2) tests. Significance was defined as α=0.05 with Bonferroni corrections as appropriate.

RESULTS

There was a significant difference in LDA joint shape and alignment discriminant scores between uninjured (-1.23 [-2.07-0.17]) and injured (1.45 [-1.20-3.31]) wrists (z=-3.146, p=0.002). There was a significant difference between uninjured versus injured scapholunate (KS2 =0.54, p<0.001) but not radioscaphoid (KS2 =0.08, p=0.894) proximity distributions (Figure 1).

CONCLUSION

SSM, a robust methodology to analyze 3D anatomies, revealed differences in scaphoid and lunate alignment between uninjured and injured wrists, with injured-sided wrists having descriptively more palmar-flexed scaphoid and extended lunate bones. These alignment differences were reflected in scapholunate but not radioscaphoid mean interosseous proximity distributions. CT-derived 3D osseous anatomies reflect injury, with implications for detection and intervention before SLAC OA progression.