Computational analysis of L4-L5 interspinous process devices and interbody fusion spacers using ceramic and polymeric materials via finite element modeling and artificial intelligence.

Lumbar spinal stenosis involves pathological narrowing of the spinal canal, whereas disc degeneration refers to the progressive deterioration of intervertebral disc structure and function. Interspinous process devices (ISPs) are commonly used to manage spinal stenosis, whereas anterior lumbar interbody fusion (ALIF) spacers are employed for conditions such as spondylolisthesis and disc degeneration. Although titanium remains the standard material for spinal devices, ceramics and polymers have emerged as promising alternatives because of their favorable biomechanical and biological properties. This study investigated the biomechanical performance of ceramic- and polymer-based materials in ALIF and ISP devices via finite element method (FEM) simulations and further explored the potential of artificial intelligence (AI) models to predict biomechanical responses on the basis of material characteristics. In the first phase, FEM models were used to evaluate spinal devices made of ceramics (zirconia, alumina), ceramic composites (ZTA, ATZ), polymers (PEEK, PEKK, PMMA), fiber-reinforced composites (CFR-PEEK, GFR-PEEK), and hydroxyapatite-reinforced PEEK (HA-PEEK) under physiological loading in both normal and osteoporotic bone conditions. In the second phase, eighteen machine learning regression models were trained on FEM-generated datasets to predict key biomechanical metrics, including stress, strain, and range of motion (ROM). The results demonstrated that for ISP devices, stiffer materials such as ceramics and CFR-PEEK 60% reduced stress and strain in adjacent vertebrae compared with softer polymers, thereby lowering fracture risk. Conversely, for ALIF spacers, more compliant materials, such as PEEK and PEKK, offered better performance in osteoporotic spines because of improved load-sharing capabilities. Among AI models, multilayer perceptron (MLP), support vector regression (SVR), and XGBoost achieved high predictive accuracy (R2 > 0.95; MAE < 5%), enabling efficient estimation of spinal biomechanics across diverse material and loading scenarios.