<p>The femoral stem is a critical implant in total hip arthroplasty; however, conventional designs often suffer from mechanical mismatch with the host bone. This study aimed to propose a rapid design strategy for re-entrant metamaterials with tailored mechanical properties, which was applied to the development of patient-specific femoral stems. A total of 300 reentrant lattice structures (16 mm × 16 mm × 16 mm) were developed by varying the re-entrant angle (<i>θ</i>: 50°–90°), the side length of the strut cross section (<i>t</i>: 0.36–1.09 mm), and compression direction (<i>d</i>: <i>y</i> or <i>z</i> axis). Stress-strain curves from quasi-static compression tests were used to train a back propagation neural network (BPNN) model, where 31 characteristic stress values and Poisson’s ratio served as input parameters, and <i>θ, t</i>, and <i>d</i> as output parameters. Predictive performance of the BPNN model was evaluated using the root mean square error (RMSE) and the coefficient of determination (<i>R</i>²). The well-trained BPNN model was then applied to match the mechanical properties of the cancellous bone obtained from subject-specific finite element analysis, thereby identifying the adaptive re-entrant structure for designing the lateral configuration of the femoral stem. The BPNN model achieved high predictive accuracy (RMSE ≤ 3.09°, <i>R</i>² ≥ 0.92 for <i>θ</i>, RMSE = 0.04 mm, <i>R</i>² ≥ 0.96 for <i>t</i>, and 100% for <i>d</i>). Implantation of the mechanically adaptive femoral stem produced distributions of stress and strain energy density (SED) in the six Gruen zones (G1-G5 and G7) comparable to those of the intact femur, with average von Mises stresses and SEDs ranging from 80% to 120% of the corresponding intact bone values. The BPNN-based inverse design framework enables precise tailoring of lattice structures for mechanical compatibility, effectively mitigating stress shielding and promoting osseointegration in patient-specific orthopedic implants.</p>