Fuzzy Optimal Associative Memory for Background Prediction of Near-Infrared Spectra

Abstract

A fuzzy optimal associative memory (FOAM) has been devised for background correction of near-infrared spectra. The FOAM yields improved predicted background scans for calculation of near-IR absorbance spectra of glucose in plasma matrices from single-beam data. The FOAM is an enhanced optimal associative memory (OAM) that uses a fuzzy function for encoding the spectra. The FOAM can predict a matching reference spectrum for a near-IR absorbance spectrum with low glucose absorbances by using second-derivative spectra. Glucose concentrations were predicted from calibration models furnished by partial least-squares (PLS). The FOAM stored reference spectra obtained from either water/phosphate buffer or plasma/glucose solutions. Both of these associative memories were evaluated. The standard error of prediction (SEP) for glucose concentration from an optimal PLS calibration model based on FOAM-corrected spectra was 0.60 mM for the water/phosphate buffer spectra. For FOAM-corrected spectra from plasma/glucose reference spectra, the SEP was 0.68 mM. The SEP of conventionally corrected double-beam second-derivative spectra was 0.81 mM. FOAM-corrected spectra generally furnish improved calibration models.