COMPUTATIONAL NEURAL NETWORKS FOR THE EVALUATION OF BIOSENSOR FIA MEASUREMENTS

A computational neural network based evaluation method is presented, w hich enables a reliable quantification of enzyme field effect transist or (EnFET) flow injection analysis (FIA) signals from samples with cha nging pH values. Two FIA systems, one for glucose and the other for ur ea determination, are employed to test the evaluation method. Measurem ent signals were obtained from samples with different glucose concentr ations (3, 4, 5, 6 and 7 g/l) and urea concentrations (1, 1.25, 1.5, 1 .75 and 2.0 g/l) at various pH values (5.5, 5.75, 6.0, 6.25 and 6.5). These signals cannot be evaluated based on the peak height, width or i ntegral. Using a large set of measuring signals for training the artif icial neural network (12 samples, each measured fivefold (=60) signals ) the error of analyte prediction from test signals are 3.2% and 2.5% for glucose and urea respectively. With a reduced training set of five measurement signals the error of prediction of the test set increases to 4.5% and 5.5% for glucose and urea respectively. In this investiga tion it will be demonstrated that computational neural networks are ab le to evaluate FIA signals, which cannot be evaluated reliably by FIA standard methods.