Machine Learning-Aided Ultranarrowband High-Q Metamaterial Absorber for Blood Cancer Detection
Abstract
This article introduces a compact, high-Q, angle-independent, and polarization-independent metamaterial absorber (MMA) designed for sensing applications in the X-band, achieving over 99% absorption efficiency. The cross-shaped structure consists of a copper metallic layer patterned with a four-arm rotationally symmetric bent-slot geometry, mounted on a Rogers RT5880 dielectric substrate. The proposed design maintains absorption above 90% for oblique incidence angles up to 70° under both transverse electric (TE) and transverse magnetic (TM) polarizations, with approximately 85% absorption retained at 80°. The cross-polarized reflection coefficient of the proposed design remains negligible across the operating band, confirming the polarization-independent behavior arising from the structure’s geometric symmetry. The absorption mechanism of the design is examined through surface current distribution, electric-field analysis, and a parametric optimization study, while an equivalent circuit model substantiates the resonance characteristics of the structure and validates the simulated resonance response. A distinct shift in the absorption response occurs with changes in the permittivity of a superstrate layer, enabling the design to function as a refractive-index sensor with a quality factor (<italic>Q</italic>) of 145, a sensitivity of 1.055 GHz/RIU, and a figure of merit (FoM) of 17.58 RIU<sup>–1</sup> over a refractive-index range of 1.0–2.0. Within the biologically relevant range (<inline-formula> <tex-math notation="LaTeX">$\text {n}{\,}={\,}{1}.{35}$ </tex-math></inline-formula>–1.50), the sensor achieves a good sensitivity and a FoM. Using this sensing response, 560 simulated absorption spectra were generated across polarization orientations, incident angles, and two discrete refractive-index conditions representing normal <inline-formula> <tex-math notation="LaTeX">$\text {n}{\,}={\,}{1}.{37}$ </tex-math></inline-formula> and cancerous <inline-formula> <tex-math notation="LaTeX">$\text {n}{\,}={\,}{1}.{39}$ </tex-math></inline-formula> blood samples. The machine-learning algorithms were evaluated using stratified fivefold cross-validation together with an independent hold-out test set. The analysis of the best-performing model achieved high classification accuracy, while feature-importance analysis identified the absorption response at <inline-formula> <tex-math notation="LaTeX">$\phi {\,}={\,}{90}^{\circ } $ </tex-math></inline-formula> as the dominant discriminative signature. These results of the sensor represent a proof-of-concept demonstration of refractive-index-based classification rather than a clinically validated diagnostic outcome. A <inline-formula> <tex-math notation="LaTeX">${2} {\,}~\times ~{\,}{2}$ </tex-math></inline-formula> prototype was fabricated and experimentally characterized, and the measured results exhibit good agreement with simulations, with minor resonance-frequency deviations attributed to fabrication tolerances, finite-array effects, and substrate-parameter variations at high frequencies.