Prediction of forearm bone shape based on partial least squares regression from partial shape

Computer-assisted corrective osteotomy using a mirror image of the normal contralateral shape as reference is increasingly used. Instead, we propose to use the shape predicted by statistical learning to deal with cases demonstrating bilateral abnormality, such as bilateral trauma, congenital disease, and metabolic disease.

Computed tomography (CT) scans of 100 normal forearms were used in this study. The whole bone shape was predicted from its partial shape based on statistical learning of the other 99 bones. Accuracy was evaluated by average symmetric surface distance (ASD), and translational and rotational errors.

ASDs for predicted shapes were 0.71-1.03 mm. Mean absolute translational and rotational errors were 0.48-1.76 mm and 0.99-6.08°, respectively.

Normal bone shape was predicted with an acceptable accuracy from its partial shape using statistical learning. Predicted shape can be an alternative to a mirror image, which may enable reduced radiation exposure and examination costs.