Application of pattern-recognition techniques in wavelength selection for instrumentally read reagent strips.

Abstract

Abstract Pattern recognition techniques--discriminant analysis and principal component analysis--are utilized in selecting the wavelengths for monitoring, by reflectance spectroscopy, color-generating reactions involving uric acid and cholesterol in serum. The data base we used was accumulated by a rapid-scanning reflectance spectrophotometer that measured reflectance at 16 wavelengths every 5 s after the reaction was initiated. The data were then analyzed in multidimensional space mainframe computer with commercial statistical software packages. The most appropriate wavelengths were those that yielded the largest generalized distance between analyte concentration by discriminant analysis and the largest weighting coefficient by principal component analysis. For uric acid, taking the ratio of reflectance measured at two wavelengths instead of at a single wavelength much better separates the clinically significant concentrations. For cholesterol, the initiated. The data were then analyzed in multidimensional space hemoglobin, can be clearly demonstrated y the "pattern" generated with principal component analysis. generalized distance between analyte generalized distance between analyte concentration by discriminant