Machine learning models for predicting tibial intramedullary nail length.

Area of Science:

• Orthopedic surgery
• Biomedical engineering
• Data science in medicine

Background:

• Tibial intramedullary nailing (IMN) is standard for tibial shaft fractures.
• Accurate preoperative prediction of IMN length remains challenging.
• Current methods often rely on intraoperative data, increasing surgery time and radiation exposure.

Purpose of the Study:

• To evaluate machine learning (ML) models for predicting tibial IMN length using anthropometric measurements.
• To compare the efficacy of different ML models in this prediction task.

Main Methods:

• Retrospective analysis of 163 patients undergoing tibial IMN.
• Collected anthropometric data: height, shoe size, olecranon-to-5th metacarpal distance (OM), tibial tuberosity-to-medial malleolus distance (TTMM).
• Employed and evaluated four ML models: linear regression, random forest, decision tree, and XGBoost using MSE and R-squared values.

Main Results:

• Linear regression achieved the best performance (R-squared=0.89, MSE=117.53).
• TTMM showed the strongest correlation with IMN length (r=0.911), followed by height and OM.
• Shoe size had a weaker correlation and negatively impacted prediction accuracy; non-linear models offered limited improvement over linear regression.

Conclusions:

• Combining anthropometric data with linear regression models accurately predicts tibial IMN length.
• This approach can optimize preoperative planning, reducing intraoperative measurements, surgery time, and radiation exposure.
• Further validation on larger, diverse datasets is recommended.