Alcohol addiction diagnosis on the basis of the polysomnographic parameters

Abstract

Abstract Alcoholism is one of the most widely occurring addiction in the world. In this paper, we proposed the method of addiction detection based on polysomnography. We have got the sleep records which were described by numerical parameters calculated from standard processed records of polysomnography signals. The database used in the experiments consisted of 172 examinations: 50% of healthy and alcohol-addicted patients, and 50% males and females, with normal-like age distribution. For the diagnosis, we have used the decision system built on an artificial neural network. In our investigations, we have optimised the input set of parameters and the network structure. To verify the correctness of the diagnosis we have used the “leave one out” validation method. Finally, we have obtained over 97% correctness of alcohol addiction diagnoses for different, optimised sets of data for men and women. we got the 8 parameters described men and 11 for women where only 5 has been common. What must be underlined such a positive result was obtained by dividing the data base. For the whole base, we have got only about 89% correct diagnoses.