Optimal Implant Sizing Using Machine Learning Is Associated With Increased Range of Motion After Cervical Disk...

Area of Science:

• Spine surgery
• Biomechanical engineering
• Artificial intelligence in medicine

Background:

• Cervical disk arthroplasty (CDA) preserves motion but relies on subjective implant sizing.
• Accurate prediction of postoperative range of motion (ROM) is crucial for optimizing CDA outcomes.
• Current methods for implant selection lack precision, potentially limiting ROM.

Purpose of the Study:

• To develop interpretable machine learning (IML) models for predicting postoperative ROM after CDA.
• To identify optimal implant sizes that maximize ROM in patients undergoing CDA.
• To enhance the precision of implant selection in cervical disk arthroplasty.

Main Methods:

• Retrospective analysis of adult patients undergoing single-level CDA (2012-2020).
• Development and comparison of IML models (bagged regression tree, bagged MARS, k-NN) against linear regression.
• Assessment of model performance using root mean square error (RMSE) and variable importance analysis.

Main Results:

• IML models achieved an average RMSE of 7.6° for ROM prediction, significantly outperforming linear regression (15.8° RMSE).
• Graft size, patient age, and preoperative caudal disk height were key predictors in the best-performing IML model.
• An implant size 110% of the adjacent healthy disk height was identified as the optimal cutoff for improved ROM.

Conclusions:

• IML models provide reliable ROM change prediction after CDA with an average error of 7.6°.
• Sizing implants to match adjacent healthy disk dimensions may be key to maximizing postoperative ROM.
• This approach offers a data-driven method for optimizing implant selection in CDA.