Artificial neural networks approach in diagnostics of Polycythemia Vera

Results of a new methodology for Polycythemia Vera (PT) diagnostics, which is based on artificial neural network technology, are presented in this paper. Our initial hypothesis has supposed that a trained artificial neural network, known as a robust modeling system, gives the solution for classification of "gray" PV diagnostic zones that occur in practice, and therefore it obtains high quality of decision process for PV diagnostics using lab and other clinical findings. Initial set of 522 patient records with 10 lab and other clinical findings were included into our study. Because of the large amount of missing data for some patients, several preprocessing phases are performed and initial set was transformed into 1654 "virtual" records for artificial neural network training and testing processes. Trained network with ten inputs, four hidden nodes, and one output for classification gave the best diagnostic results. Significant differences (p<0.001) were found in comparison of correct diagnostic classification of patients using a trained ANN (97.1%), and using PVSG diagnostic criteria (73.2%). A new set of lab parameters, defined in this study as ANN inputs, gives a basis for better prediction of PV diagnosis than the set used in PVSG criteria These results are an important step in specification of new PVSG diagnostic criteria applicable in clinical practice.