[Application of wavelet transform-radial basis function neural network in NIRS for determination of rifampicin and isoniazide tablets].

A calibration model (WT-RBFNN) combination of wavelet transform (WT) and radial basis function neural network (RBFNN) was proposed for synchronous and rapid determination of rifampicin and isoniazide in Rifampicin and Isoniazide tablets by near infrared reflectance spectroscopy (NIRS). The approximation coefficients were used for input data in RBFNN. The network parameters including the number of hidden layer neurons and spread constant (SC) were investigated. WT-RBFNN model which compressed the original spectra data, removed the noise and the interference of background, and reduced the randomness, the capabilities of prediction were well optimized. The root mean square errors of prediction (RMSEP) for the determination of rifampicin and isoniazide obtained from the optimum WT-RBFNN model are 0.00639 and 0.00587, and the root mean square errors of cross-calibration (RMSECV) for them are 0.00604 and 0.00457, respectively which are superior to those obtained by the optimum RBFNN and PLS models. Regression coefficient (R) between NIRS predicted values and RP-HPLC values for rifampicin and isoniazide are 0.99522 and 0.99392, respectively and the relative error is lower than 2.300%. It was verified that WT-RBFNN model is a suitable approach to dealing with NIRS. The proposed WT-RBFNN model is convenient, and rapid and with no pollution for the determination of rifampicin and isoniazide tablets.