Finite element and multivariate random forests modelling for stress shield attenuation in customized hip implants

• Published in: Forces in mechanics 2023-02, Vol.10, p.100151, Article 100151
• Main Authors: Shehata, Merna Ehab, Mustapha, K.B., Shehata, E.M.
• Format: Article
• Language: English
• Subjects: Finite element analysis; Hip implant; Random forests modelling, Data-driven model; Stress shielding; Total hip replacement
• ISSN: 2666-3597; 2666-3597
• DOI: 10.1016/j.finmec.2022.100151

•Developed and assessed the potential of using computational intelligence tools towards an optimized design of primary normal stem hip implant.•Designed over 100 hip implant CAD models with varying critical geometrical parameters from the base model of a commercial implant.•Analysed intact and implanted femurs using finite element method to retrieve critical von Mises stress during walking and stair-climbing.•Implemented multivariate random forests schemes that exhibit excellent predictive strengths for the inverse modelling of hip implant performance. Primary total hip replacement surgery has an undisputable reputation as a widely successful orthopaedic operation, but it is beset by a phenomenon known as stress shielding. The cause of stress shielding is multifaceted. However, its reduction is reported to be hinged on the optimal design of prosthetic implants. Yet, to date, the design of a hip implant profile that behaves biomechanically similar to the natural physiological load-bearing zones of the femur remains an open problem. Along this vein, this paper instantiates an inquiry into the development of a framework that couples the capability of the finite element analysis (FEA) with that of machine learning methods toward the discovery of optimal design parameters for a customized hip implant. First, premised on the properties of a commercial normal-stem hip implant, a baseline computer-aided design (CAD) parametric model was created. From the baseline CAD model, a database of 120 hip implant profiles is established from the perturbation of the lateral edge, lateral angle, and the ratio of the radial cross-sectional areas of the implant. Next, the validation of the developed finite element procedure was conducted on a healthy intact femur and detailed numerical simulations were undertaken to assess the stress shielding (SS) attributes of all hip implants in the established database. The ensuing stress and strain data from the FEA is then deployed to ward a data-driven inverse model based on the random forests machine learning algorithm. Results-wise, the validation of the static analysis on the intact femur yielded von Mises stresses that matched those reported in published studies. Moreover, other results from the FEA revealed that a rectangular cross-sectioned hip implant resulted in the highest SS in the four zones of the proximal femoral compared to the trapezoidal cross-sectioned implant. Further, the inverse RF model exhibited excellent predictive