Statistical shape models that predict native glenoid width based on glenoid height are inaccurate in their current form: a cross-sectional study

Abstract

The extent of measurement errors of statistical shape models that predict native glenoid width based on glenoid height to subsequently determine the amount of anterior glenoid bone loss is unclear. Therefore, the aim of this study was to (1) create a statistical shape model based on glenoid height and width measured on 3-dimensional computed tomography (3D-CT) and determine the accuracy through measurement errors and (2) determine measurement errors of existing 3D-CT statistical shape models. A retrospective cross-sectional study included all consecutive patients who underwent CT imaging before undergoing primary surgical treatment of traumatic anterior shoulder dislocation between 2007 and 2022 at the Tohoku University Hospital and affiliated hospitals. Patients were included when instability was unilateral and CT scans of both the injured and contralateral uninjured shoulder were available. 3D segmentations were created and glenoid height and width of the injured and contralateral uninjured side (gold standard) were measured. Accuracy was determined through measurement errors, which were defined as a percentage error deviation from native glenoid width (contralateral uninjured glenoid), calculated as follows: measurement error = [(estimated glenoid width with a statistical shape model - native glenoid width) / native glenoid width] × 100%. A linear regression analysis was performed to create a statistical shape model based on glenoid height according to the formula: native glenoid width=a × glenoid height+b. The diagnosis and procedure codes identified 105 patients, of which 69 (66%) were eligible for inclusion. Glenoid height demonstrated a very strong correlation (r=0.80) with native glenoid width. The linear regression formula based on this cohort was as follows: native glenoid width=0.75 × glenoid height - 0.61, and it demonstrated an absolute average measurement error of 5%±4%. The formulas by Giles etal, Chen etal and Rayes etal demonstrated absolute average measurement errors of 10%±7%, 6%±5%, and 9%±6%, respectively. Statistical shape models that estimate native glenoid width based on glenoid height demonstrate unacceptable measurement errors, despite a high correlation. Therefore, great caution is advised when using these models to determine glenoid bone loss percentage. To minimize errors caused by morphologic differences, preference goes to methods that use the contralateral side as reference.