N.T. Abdullayev1, N.Ya. Mammadov2, A.N. Jafarova3, G.E. Abdullayeva4

1,4 Azerbaijan Technical University (Baku, Azerbaijan)

2,3 Azerbaijan State University of Oil and Industry (Baku, Azerbaijan)

When solving problems of medical diagnostics, artificial neural networks are widely used, with the help of which the possibility of classifying diseases is realized. With interval input parameters (symptoms), the choice of criteria for evaluating the effectiveness of the functioning of neural networks is relevant.

The purpose of the work is to conduct research on the creation of experimental data for the considered diseases of the neuromuscular system, the formation of input parameters for training the selected neural network, as well as the choice of criteria for evaluating the effectiveness of their functioning.

Due to the fact that the numerical values of the symptoms of diseases of the neuromuscular system vary in certain intervals of change, in diagnostic problems using artificial neural networks, first of all, it is necessary to determine which of the input parameters is the most informative, i.e. it is necessary to estimate the significance of features for the three-layer perceptron used in diagnostic and recognition problems. A classification of diseases associated with lesions of peripheral nerves is carried out, such as carpal tunnel syndrome (median nerve) and cubital tunnel syndrome (ulnar nerve). Also considered is demyelinating neuropathy, which is characterized by multiple lesions of the peripheral nerves, leading to muscle weakness.

With interval set parameters of the neural network, to select the optimal structure, the method of sequentially adding neurons to the hidden layer is used, and at each step the error functional (network accuracy) is minimized. The operations are repeated after the next step of changing the interval value. Evaluation of neural network performance criteria is repeated for all interval parameters of previously determined significant symptoms.

The condition for stopping the algorithm is to obtain, according to the criterion of the error functional, the optimal structure of the network.

As a result of the research, the possibility of obtaining the optimal structure of the neural network according to the criteria of minimal error and the significance of input symptoms with interval-defined parameters of the considered myographic diseases has been shown.

Abdullayev N.T., Mammadov N.Ya., Jafarova A.N., Abdullayeva G.E. Application of artificial neural networks with interval input parameters in electromyography. Biomedicine Radioengineering. 2022. V. 25. № 2–3. Р. 73-83. DOI: https://doi.org/10.18127/j15604136-202202-08 (In Russian)

1. Khaykin S. Neyronnyye seti: polnyy kurs. M.: ID "Viliams". 1994. 1104 s. (in Russian).
2. Tarik Rashad. Sozdayem neyronnuyu set. SPb.: OOO "Alfa – kniga". 2020. 274 s. (in Russian).
3. Vorontsov K.V. Lektsii po iskusstvennym neyronnym setyam. M.: Knigagid "Orphus". 2007. 24 s. (in Russian).
4. Kallan R. Neyronnyye seti. Kratkiy spravochnik. M.: "Viliams". 2017. 288 s. (in Russian).
5. Grachev A.M. Metody szhatiya rekurrentnykh neyronnykh setey dlya zadach obrabotki estestvennogo yazyka: Avtoref. diss. na soiskaniye uchen. st. kand. komp. nauk. M.: VShE. 2019. 18 s. (in Russian).
6. Ivanov N.V. Neyronnyye seti v meditsine. Slozhnyye sistemy. 2018. №4(29). S. 46–70. (in Russian).
7. Al-Khaydri V.A., Isakov R.V., Sushkova L.T. Vozmozhnost primeneniya iskusstvennykh neyronnykh setey dlya obnaruzheniya iskazheniy EKG. Biotekhnosfera. 2016. №1(43). S. 8–13. (in Russian).
8. Konnova N.S., Khaperskaya V.Yu. Primeneniye neyrosetey i algoritmov mashinnogo obucheniya s tselyu diagnostiki serdechno-sosudistykh zabolevaniy na osnove signalov SKG. Biomeditsinskaya radioelektronika. 2020. T. 23. № 1. S. 5–20. (in Russian).
9. Dubrovin V.I., Subbotin S.A. Otsenka znachimosti priznakov na osnove mnogosloynykh neyronnykh setey v zadachakh diagnostiki i raspoznavaniya. Tekhnicheskaya diagnostika. 2002. №1(3). S. 66–72. (in Russian).
10. Gekht B.G., Kasatkina L.F., Samoylov M.I. i dr. Elektromiografiya v diagnostike nervno-myshechnykh zabolevaniy. Taganrog: TRTU. 1997. 370 s. (in Russian).
11. Kasatkina L.F., Gilvanova O.V. Elektromiograficheskiye metody issledovaniya v diagnostike nervno-myshechnykh zabolevaniy. Igolchataya elektromiografiya. M.: Medika. 2010. 416 s. (in Russian).
12. Vasilyev A.N., Tarkhov D.A. Matematicheskiye modeli sistem s intervalno zadannymi parametrami na osnove geterogennykh neyronnykh setey. Prodolzheniye temperaturnogo polya – klassicheskaya postanovka zadachi. Neyrokompyutery: razrabotka i primeneniye. 2014. № 3. S. 12–16. (in Russian).
13. Tarkhov D.A. Neyrosetevyye modeli i algoritmy. M.: Radiotekhnika. 2014. 352 s. (in Russian).
14. Ezhov A.A., Chechetkin V.R. Neyronnyye seti v meditsine. Otkrytyye sistemy. 1997. № 4. S. 34–37. (in Russian).
15. Komartsova L.G., Maksimov A.V. Neyrokompyutery. M.: Izd-vo MGTU. 2002. 311 s. (in Russian).