A collaborative approach of finite element method and machine learning algorithms for biomechanical analysis of implants used in tibial shaft fractures.

backgroundTibial fractures are among the most common complex orthopedic injuries. The mechanical strength and biomaterial properties of implants used in the treatment of such fractures directly affect the healing process. In this study, the mechanical effects of different implant designs and biomaterials on obliquely fractured tibia were analyzed. In addition, it was aimed to evaluate the data obtained from finite element analysis (FEA) with machine learning (ML) algorithms.

methodsSeven implant models for tibial shaft fractures were analyzed using static structural simulations in Ansys Workbench. Implants and cortical screws were made of Ti–6Al–4 V alloy or 316 L stainless steel (SS), and axial loads of 600, 800, and 1000 N simulated single-leg stance. A dataset of 1008 points, including maximum stress and total displacement, was generated and used to train Multilayer Perceptron (MLP), Support Vector Machine (SVM) and Decision Tree (DT) models in WEKA.

resultsThe mechanical behaviors of different implant and biomaterial combinations were compared, and the maximum stress value in implants with 316 L SS material properties was higher than the maximum stress value in Ti-6Al-4 V alloy implants. When the total displacement values ​​in the tibia fracture region were examined, 316 L SS implants gave better results. In the machine learning estimations, the SVM model outperformed the MLP and DT algorithms. For maximum stress prediction, SVM achieved an mean absolute error (MAE) of 0.24 and 0.41 for the training and test sets, respectively, while MLP and DT showed higher errors (3.27/4.02 and 10.99/14.90, respectively). Similarly, for total displacement prediction, SVM showed the lowest errors with MAE values of 0.0003 and 0.0015 for the training and test sets, whereas MLP and DT had higher MAE (0.0032/0.0040 and 0.0058/0.0072, respectively).

conclusionThis study evaluated the effects of different implant designs and biomaterials on oblique tibial fractures and demonstrated that finite element results can be accurately predicted using machine learning models. The SVM algorithm showed superior performance, with prediction errors of approximately 0.24–0.41% MAE and 0.27–0.49% root mean square error (RMSE) for maximum stress, and 0.03–0.15% MAE and 0.03–0.23% RMSE for total displacement in the training and test sets, respectively. In comparison, MLP and DT exhibited higher errors. These findings highlight the potential of data-driven approaches in biomechanical analyses and their contribution to developing clinical decision support systems.