Machine learning–assisted optimization of a terahertz photonic metamaterial absorber for blood cancer detection

Blood cancer originates in the bone marrow and disrupts the body’s normal hematopoietic processes. The rapid progression of this disease highlights the need for accurate and sensitive detection to improve treatment outcomes. Here, we present a novel, compact, multi-resonant terahertz photonic metamaterial absorber for blood cancer detection. The compact structure, with dimensions of 0.41λ0 × 0.41λ0 × 0.029λ0, integrates multiple circular resonators with a rectangular patch, supporting strongly confined resonances that achieve near-unity absorption of 97.6%, 99.9%, and 99.6% at 3.48 THz, 4.95 THz, and 6.01 THz, respectively. The sensing performance was evaluated by introducing an analyte layer representing normal and cancerous blood cells, resulting in remarkable sensitivities of 1.28 THz/RIU, 3.00 THz/RIU, and 2.14 THz/RIU at the three resonance frequencies. The corresponding quality factor (Q-factor) values are 15.03, 20.21, and 24.5, and the figure of merit (FOM) values are 5.54, 12.24, and 8.73 for the three resonance peaks, supporting its reliability in sensing. Moreover, the electric field, magnetic field, and surface current distributions were analysed, and an equivalent circuit model was also developed and validated against the simulated results. Several machine learning models were also employed for design prediction, with Gradient Boosting demonstrating excellent performance and enabling up to a 60% reduction in optimization time. The combination of a multi-band, high-absorption design and ML-assisted approach provides a robust, ultrathin, and high-sensitivity platform, offering a promising route toward next-generation terahertz biophotonic sensors for accurate and sensitive blood cancer detection.