Nonunion scaphoid bone shape prediction using iterative kernel principal polynomial shape analysis

Area of Science:

• Orthopedic surgery
• Medical imaging
• Biomechanical modeling

Background:

• The scaphoid bone is crucial for carpal stability.
• Accurate scaphoid shape estimation is vital for treating scaphoid nonunion.
• Traditional statistical shape models (SSM) using principal component analysis (PCA) have limitations in linearity and mode adjustment.

Purpose of the Study:

• To develop an iterative kernel principal polynomial shape analysis (iKPPSA)-based SSM for predicting the pre-morbid scaphoid shape.
• To enhance the accuracy and robustness of scaphoid shape prediction models.

Main Methods:

• Trained an SSM using 65 scaphoid image sets.
• Validated the model on 9 image sets with simulated defects (tubercle, proximal pole, avascular necrosis).
• Evaluated iKPPSA-SSM against PCA, PPSA, and KPCA using mean error, Hausdorff distance, and Dice coefficient.

Main Results:

• The iKPPSA-based SSM demonstrated robustness with 0.264 mm generality and 0.260 mm specificity.
• It captured 99% of variability within the first seven principal modes.
• Outperformed PCA-based models for proximal pole fractures, reducing errors by up to 25.2% and improving Dice coefficient by 0.35%.

Conclusions:

• The iKPPSA-based SSM effectively leverages data nonlinearity for high reconstruction accuracy.
• This model can be integrated into preoperative planning for scaphoid fracture management.
• It also supports morphology-based biomechanical modeling of the scaphoid.