350 rub
Journal Biomedical Radioelectronics №9 for 2014 г.
Article in number:
Hybrid neural network with macrolayers for medical application
Authors:
E.N. Korovin - Dr.Sc. (Eng.), Professor, Voronezh State Technical University
О.V. Shatalova - Ph.D. (Eng.), Associate Professor, Southwest State University
V.V. Zhilin - Ph.D. (Eng.), Associate Professor, Kursk State Agricultural Academy n.a. professor I.I. Ivanov
Abstract:
It is suggested to use the intellectual structure of multiagent systems for the diagnosis and prognosis of cardiovascular disease (CVD) in difficult situations. In this structure the expert fuzzy information about the presence or absence of a pathology provides important information for managing, as preventive and diagnostic studies. Found that to implement multi-agent structure using probabilistic neural network (PNN) with makrosloyers. Classification of network PNN, for expert assessments of probabilities of alternative hypotheses, Bayes method should be used. Implementing the method of probability density function is evaluated for each target class. Proposed measure error rates of classification. The main strategic task to select a neural network architecture. In modeling neural networks use neural network block type to allow classification of the object of study classes. The article is a block diagram of the probabilistic neural network block type, which implements this approach. Health of the considered method was checked when each input data set had six symptoms. The main aspects for the training set that characterizes the neural network architecture. The final decision to the probabilistic neural network layer is added makrolayer FNN by fuzzy technology. This layer is based on fuzzy logic decision making. A distinctive feature of the proposed approach is that each module classification contains the concatenated block PNN and FNN. The PDS makes the final decision on the basis of the analysis of probability from the outputs of the blocks of FNN. The article describes the structure of the neural network with two makroslayers PNN and FNN. Found that the quality of decisions, based on the proposed method outperforms the known methods by an average of 10 to ...16%.
Pages: 32-37
References

  1. Zhuravlev Ju.I., Nazarenko G.I., Ryazanov V.V. i dr.  Novy'j metod analiza riska razvitiya ishemicheskoj bolezni serdcza na osnovanii genomny'x i komp'yuterny'x texnologij // Kardiologiya. 2011. T. 51. № 2. S. 19 - 25.
  2. Tomakova R.A., Filist S.A., Zhilin V.V., Gorbatenko S.A. Strukturno-funkczional'ny'e resheniya nechetkix nejronny'x setej dlya intellektual'ny'x sistem analiza raznotipny'x priznakov // Fundamental'ny'e i prikladny'e problemy' texniki i texnologii. 2011. № 1. S. 85-91.
  3. Filist S.A., Zhilin V.V., Al'-Mualemi Vail Abdulkarim Gibridny'j sposob klassifikaczii biosignalov na osnove texnologij nechetkoj logiki prinyatiya reshenij i nejronny'x setej // Biomediczinskaya radioe'lektronika. 2009. № 5. S. 77-82.
  4. Burmaka A.A., Volkov I.I., Ivanov V.A. i dr. Veroyatnostny'e seti s makrosloyami v sistemax podderzhki prinyatiya reshenij po differenczial'noj diagnostike serdechno-sosudisty'x zabolevanij // Mediczinskaya texnika. 2013. № 4. S. 18-20.
  5. Podval'ny'j E.S. Modeli individual'nogo prognozirovaniya i klassifikaczii sostoyanij v sistemax komp'yuternogo monitoringa. Voronezh: Izd-vo VGTU. 1998. 127 s.