Hybrid Neural Network for the Differential Diagnosis of Pancreatic Diseases with Descriptors Determined by Bioimpedance Spectroscopy

We developed a method for forming descriptors based on the results of bioimpedance spectroscopy, which forms the basis of a hybrid classifier. The hybrid classifier contains five macro layers, the first of which is based on probabilistic neural networks, and four are based on fuzzy logical inference. The structure of the device, which allows to provide the formation of the space of informative features for the hybrid classifier of pancreatic diseases, is presented. Experimental studies of software and clinical trials of a medical decision support system built on the basis of the proposed structural and functional organization of a hybrid neural network have shown that, in terms of diagnostic sensitivity and diagnostic specificity, the proposed technical solutions achieve acceptable for medical practice quality indicators for risk classification of pancreatic diseases.