AI-Assisted 3D-Printed Meander-Line Antenna for Non-Invasive Bone-Tumor Sensing

Abstract

This study presents a compact 3D-printed meander-line antenna optimized using Bayesian optimization for non-invasive bone-tumor sensing. The antenna geometry was optimized by adjusting feed and patch dimensions, rotation angle, and substrate permittivity. A simple multilayer bone-mimicking phantoms with tumor-like inclusions were fabricated to enable controlled and reproducible measurements. Experimental results obtained using PLA, ABS, and resin substrates demonstrate clear material-dependent electromagnetic responses. PLA- and resin-based antennas show noticeable resonance deepening and small frequency down-shifts in the 2-3 GHz range in the presence of a tumor inclusion, indicating higher dielectric sensitivity. In contrast, the ABS-based antenna exhibits a more stable resonance behavior near 5.5 GHz with smaller amplitude variations, suggesting improved structural stability but lower sensitivity. To model the nonlinear relationship between dielectric loading and antenna response, a pyramidal deep regression network (PDRN) was trained using measured |S11| data. The proposed framework provides a low-cost, data-efficient, and reproducible approach for evaluating tumor-induced electromagnetic perturbations under controlled phantom conditions.