Comparison efficiency of the artificial intelligence methods for the diagnosis of Acid - base and anion gap disorders.

Diagnosis of the most complicated disorders in acid-base status and accompanying electrolyte balance creates a lot of troubles for practicing physicians. The purpose of our study was to create and compare: 1) an artificial neural network, 2) genetic program, 3) fuzzy-neural system that can diagnose acid-base disorders, based on a set of laboratory gasometric and electrolyte measurements. We took into account 7 single acid-base disorders, 11 double acid-base disorders and 6 triple complicated disorders with accompanying anion gap alterations. We prepared a set laboratory measurements consisting of 250 results for training and the same number of results for testing the program. Finally, the efficiency of presented artificial intelligence (AI) methods has been described and compared.

Genetic programming in medical diagnosis.

The paper describes the results obtained for differential diagnostics of substantially complex disorders in the acid-base status. Since the first attempts of applying the genetic programming to this problem, presented by the authors at MIE 2000, proved promising, the investigation has been continued. The goal of the present study is to create computer genetic programs which, while taking into account a set of laboratory gasometric and electrolyte measurements, would be able to assist the diagnostics of acid-base disorders. Seven (7) single acid-base disorders, eleven (11) double acid-base disorders and six (6) triple complicated disorders with accompanying anion gap alterations are approached in the study. A set of simulated laboratory measurements has been prepared, providing 250 results for the evaluation of the fitness function of the designed genetic computer programs, plus some additional results for testing. The final results are presented in the form of a confusion matrix for the testing data, which shows that the developed system may be helpful in clinical practice.