Optimization of in situ gold nanorod synthesis via a combined RSM-BP neural network approach: Application to SERS detection of tetracycline

The presence of trace tetracycline (TC) in aqueous environments demands sensitive and scalable detection methods, while conventional techniques often suffer from poor selectivity and high detection limits. Herein, a synergistic optimization strategy combining response surface methodology (RSM) and backpropagation (BP) neural network was developed to precisely tailor the localized surface plasmon resonance (LSPR) properties of gold nanorods (AuNRs) for constructing a high-performance surface-enhanced Raman scattering (SERS) platform for TC detection. One-way experiments were used to investigate the effects of individual parameters on AuNR structure, followed by Box-Behnken design to build a composite response value integrating LSPR peak position and absorption intensity ratio, enabling optimization of CTAB, AgNO 3 , AA, and NaBH 4 concentrations. A BP neural network was applied to further enhance parameter prediction accuracy. The optimal synthesis conditions yielded AuNRs with good morphological uniformity and an LSPR peak at 767 nm, confirmed by UV–Vis and TEM. The resulting SERS platform achieved a detection limit of 8.5 × 10 −10  M (S/N = 3) for TC, about six orders of magnitude more sensitive than conventional Raman. In real water samples, recoveries ranged from 91.0 % to 103.0 %, with RSD < 5 %, indicating excellent reproducibility and anti-interference capability. This study demonstrates a generalizable route for precision nanostructure synthesis and offers a reliable platform for ultrasensitive antibiotic detection in complex aqueous matrices, showing strong promise for environmental and food safety monitoring.