Russian version English version
Volume 14   Issue 2   Year 2019
On Possibility of Machine Learning Application for Diagnosing Dementia by Eeg Signals

Dorovskih I.V.1, Senko O.V.2, Chuchupal V.Ya.2, Dokukin A.A.2, Kuznetsova A.V.3

1"1586 Military Clinical Hospital" of the Ministry of Defense of the Russian Federation, Podolsk, Russia
2Federal Research Center "Computer Science and Control" of the Russian Academy of Sciences, Moscow, Russia
3N.M. Emanuel Institute of Biochemical Physics of the Russian Academy of Sciences, Moscow, Russia

Abstract. The purpose of this study was to investigate the possibility to use electroencephalography for early diagnostics of dementia and for objective assessment of disease severity and neurometabolic treatment results. The study was based on application of machine learning methods for computer diagnosis of dementia by the energy spectra of EEG signals. Effectiveness of various machine learning technologies was investigated in order to separate different groups of patients with varying severity of dementia from healthy ones and patients with pre-dementia disorders according to the vectors of spectral indicators. Applying of cross-validation procedure showed that separation of the group with dementia from the group of people with normal physiological aging and groups of young people reaches 0.783 and 0.786, respectively by parameter ROC AUC. The results of the study allow to make an assumption, that the algorithmic assessment of dementia severity by EEG corresponds to the actual course of the disease. So, the number of cases with algorithmically identified positive dynamics significantly exceeds the number of cases with algorithmically detected negative dynamics after neurometabolic therapy in the group with mild dementia. In a combined group with both average and heavy severity of the disease such an increase was not observed.

Key words: machine learning, diagnosis, dementia, ňlectroencephalogram, cerebrolysin.

Table of Contents Original Article
Math. Biol. Bioinf.
doi: 10.17537/2019.14.543
published in Russian

Abstract (rus.)
Abstract (eng.)
Full text (rus., pdf)


  Copyright IMPB RAS © 2005-2024