Quantitative classification of HbA1C and blood glucose level for diabetes diagnosis using neural networks


AUSTRALASIAN PHYSICAL & ENGINEERING SCIENCES IN MEDICINE, vol.36, no.4, pp.397-403, 2013 (SCI-Expanded) identifier identifier identifier


In this study, artificial neural network structures were used for the quantitative classification of Haemoglobin A1C and blood glucose level for diabetes diagnosis as a non-invasive measurement technique. The neural network structures make inferences from the relationship between the palm perspiration and blood data values. For this purpose, feed forward multilayer, Elman, and radial basis neural network structures were used. The quartz crystal microbalance type and humidity sensors were used for the detection of palm perspiration rates. Total 297 volunteer's data is used in this study. Three quarters of the data was used to train the neural networks. The remaining data were used as test data. The best results for the quantitative classification were obtained from the feed forward NN structure for the detection of the glucose and HbA1C level quantities. And, the performances of all neural networks for the HbA1C value were better than the performances of these neural networks for the glucose level.