Evaluation of the quality of functioning of artificial neural network with logic outputs in the diagnosis of diseases of hepatopancreatoduodenal zone

Aim. To develop a set of information methods to improve the quality of neural network diagnosis of diseases of hepatopancreatoduodenal zone. Methods. The study involved 385 patients with peptic ulcer, cholecystitis and pancreatitis undergoing in-patient treatment in medical organizations of the city of Kursk. For data mining internally developed software «System of Intellectual Analysis and Diagnosis of Diseases» was used which is an environment for the creation, adjustment, training and practical clinical application of an artificial neural network, such as a multilayer perceptron with an activation function - hyperbolic tangent. Results. Hyperbolic tangent (activation function) of the output layer’s neuron takes the value OUT ∈ ℝ ∧ OUT ∈ (-1; 1) which requires an interpretation. For logic network gates, for example, presence/absence of a disease, it can be performed by comparison with an arbitrarily assigned threshold yB ∈ (0; 1). In this approach, the values are interpreted as false (if y ≤-yB), undefined if y ∈ (-yB; yB), or true (if y ≥yB). Network operation control includes calculation of sensitivity, specificity, false positive and false negative results, for which the comparison of arrays of pairs of calculated and empirical values is carried out. In case of artificial neural network use for diagnosing diseases of hepatopancreatoduodenal zone, the optimal mode was achieved assigning yB≈0.3 as a threshold of the output neuron activation function. Conclusion. Assessing the quality of the ability of artificial neural network with logic outputs to diagnose hepatopancreatoduodenal zone diseases, as well as its controlled setting, is most effective by evaluation of sensitivity, specificity, frequency of false positive and false negative results at the threshold value yB≈0.3; the demonstrated sensitivity (83-94.7%) and specificity (83-97.8%) levels are comparable to the traditionally used diagnostic methods.

artificial neural networks, prognosis of diseases, peptic ulcer disease, cholecystitis, pancreatitis, diagnosis, artificial neural networks, prognosis of diseases, peptic ulcer disease, cholecystitis, pancreatitis, diagnosis