Analyzing the Sensitivity, Reliability, and Classification Accuracy of Vibration-Based Metrics for Stability Assessment of Percutaneous Osseointegrated Transfemoral Implants

Abstract

Vibration methods noninvasively assess the stability of percutaneous osseointegrated implants by correlating natural frequencies with bone-implant interface (BII) condition. However, these frequencies are sensitive to geometric variation, limiting their use as absolute stability metrics. A previously proposed method uses a one-dimensional (1D) finite element (FE) model to estimate BII stiffness (k), enabling geometry-independent stability assessment for press-fit transfemoral implants. This study extends the 1D model to incorporate variability in stem length and diameter, adapter length, and connector presence. A tapered element formulation was introduced to better represent components with nonuniform cross-sections. Synthetic clinical signals were generated using a three-dimensional (3D) FE model across 48 scenarios involving variations in BII stiffness, geometry, and damping. The 1D model predicted k values of 3.58±0.36×106, 2.96±0.25×108, and 1.97±0.26×109 N/m for the LOW, INTERMEDIATE, and HIGH BII conditions, respectively. Statistical analysis confirmed that k and the first mode frequency (f1) differentiate between BII groups, while the second mode frequency (f2) did not. Unlike k, f1 exhibited a monotonically increasing coefficient of variation, indicating that its reliability decreases for stiffer configurations and that accounting for geometry is a non-trivial scaling matter. Classification accuracy further supported this: 100% for k, 83% for f1, and 17% for f2. Clinically, misclassifications can lead to incorrect diagnosis of BII condition. These results support stiffness prediction via the 1D FE model as a clinically viable tool for implant stability assessment.