Prediction of medical examination results using radial-basis function networks

This paper proposes a method of predicting future medical examination measurements given the past values. The medical examinations considered in this paper are blood sugar level, low and high blood pressures, and cholesterol level. This paper uses a specific type of artificial neural networks, radial-basis function network (RBFN), to approximate mapping from the past medical measurements to that of the upcoming year, in order to help the subjects be aware of the signs of unusual health states without consulting with doctors. Experimental results show that the RBFN-based estimation is superior to the conventional linear regression in terms of prediction accuracy of the future examination measurements. The proposed method is expected to be implemented in a handy consumer electronic devices such as Smartphones without adding extra hardware parts provided that the history of the medical examination measurements are available.