The aim of this study was to identify points on the scapula that can be used to predict the anatomy of the native premorbid glenoid.

Forty-three normal scapulas reconstructed in 3D and positioned in a common coordinate system were used. Twenty points distributed over the blade of the scapula (portion considered normal and used as a reference) and the glenoid (portion considered pathological and needing to be reconstructed) were captured manually. Thirteen distances (X) between two points not on the glenoid and 31 distances (Y) between two points of which at least one was on the glenoid were then calculated automatically. A multiple linear regression model was applied to calculate the Y distances from the X distances. The best four equations were retained based on their coefficient of determination (R²) to explain a point on the glenoid being reconstructed (p<0.05). In the first scenario, the glenoid was modeled assuming it was completely destroyed. In the second scenario, only the inferior portion of the glenoid was worn.

For a completely destroyed glenoid, the mean error for a chosen distance for a given point on the glenoid was 2.4 mm (4.e-3mm; 12.5mm). For a partially damaged glenoid, the mean error was 1.7mm (4.e-3mm; 6.5mm) for the same distance evaluated for a given point on the glenoid.

The proposed statistical model was used to predict the premorbid anatomy of the glenoid with an acceptable level of accuracy. A surgeon could use this information during the preoperative planning stage and during the actual surgery by using a new surgical assistance method.