Optimizing preoperative planning for total hip arthroplasty using random forest models to predict stem size and...

Area of Science:

• Orthopedic Surgery
• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Preoperative planning is crucial for successful total hip arthroplasty (THA).
• Accurate selection of femoral stem size and ensuring compatibility with patient anatomy are vital to prevent complications like distal fixation.
• Existing deep learning methods for preoperative planning require further validation regarding optimization of stem size and compatibility.

Purpose of the Study:

• To develop and evaluate supervised machine learning models for estimating optimal femoral stem size and compatibility.
• To provide tools that assist surgeons in preoperative planning for THA, focusing on stem selection.

Main Methods:

• Developed two Random Forest (RF) models: one for stem size estimation (seven-class classification) and one for stem compatibility (binary classification).
• Trained models using computed tomography (CT) scan data from 320 hips, including 10 femoral measurements and derived ratios.
• Validated models on data from 109 THA patients, assessing performance using accuracy, F1 score, precision, and recall.

Main Results:

• Models achieved improved accuracies of 92.7% for size estimation and 87.2% for compatibility estimation when ratio features were included.
• Feature importance analysis identified distal medullary cavity diameter as critical for size and overall femoral dimensions for compatibility.
• Initial models without ratio data showed accuracies of 89.0% (size) and 85.3% (compatibility).

Conclusions:

• Developed accurate machine learning models to estimate critical factors (size and compatibility) for THA stem selection.
• These models show potential to support preoperative planning for the Accolade II stem.
• Further validation is needed to confirm applicability across different implant systems.