Determination and analytical validation of creatinine content in serum using image analysis by multivariate transfer calibration procedures

The main objective of this study was to explore the feasibility of image analysis (RGB, HSI and gray intensity histograms) and partial least squares regression using a calibration model transfer technique in the quantitative analysis of creatinine in serum samples by the use of two different devices: a desktop scanner and a cell phone camera. In addition, a multivariate validation model based on linearity, accuracy, sensitivity, bias, prediction uncertainty and beta-expectation tolerance intervals was estimated. The colorimetric reaction for creatinine was carried out in a 96-microwell plate format with flat-bottomed 250 mu L microwells. The results achieved separately for both devices were very significant compared to the reference method, showing no statistical difference at a confidence level of 95%. When the calibration model based on the scanner was used directly to predict the concentration for cell phone data, it produced an unsatisfactory prediction with the RMSEP = 0.79 mg dL(-1). However, the prediction was greatly improved after the calibration was transferred based on DS (RMSEP = 0.14 mg dL(-1)). The same trend was observed when the scanned data were predicted by the calibration model based on the cell phone, where the initial RMSEP = 0.25 mg dL(-1) was reduced to RMSEP = 0.10 mg dL(-1), after calibration transfer. These results show the transferability of the calibration transfer technology applied to the image data, where efficient calibration transfer to other devices was clearly demonstrated with all devices in the study effectively giving similar results on a transfer set.