M.V. Vyucheiskaia - Post-graduate Student, Department of Applied Mathematics and High Performance Computing, Northern (Arctic) Federal University named after M.V. Lovonosov (Arkhangelsk). E-mail: maria.vmv@mail.ru I.S. Biriukov - Junior Researcher Scientist, Laboratory of Neurophysiology and Higher Nervous Activity, Northern (Arctic) Federal University named after M.V. Lovonosov (Arkhangelsk). E-mail: bis4937@mail.ru

Impairments of cognitive functions are one of the most common neurological disorders. At present early diagnosis of cognitive impairment with a view to the earliest appointment of therapy to prevent or delay the onset of dementia in the elderly becomes important. The current dissatisfaction with the solution of the problem determines the urgency of developing new approaches to the early diagnosis of cognitive impairment. At present artificial neural networks take on special significance, especially their application for the automation of scientific research and applied tasks, in particular to address the medical and biological diagnostics tasks. The aim of our study was to develop artificial neural network, which allows diagnosing of cognitive impairment in elderly patients (50-80 years). The neural network problem was the correct classification of patients into the following groups: the patient is healthy; the patient has a mild cognitive impairment; the patient has a mild cognitive impairment; the patient has severe cognitive impairment. The initial sample size of 127 people was divided into training (70 people) and control (57 people) group. As the parameters of the neural network were: age (the number of complete years), the average time of a complex sensory-motor choice reaction to visual stimulus (ms), the number of errors during the test of the difficult choice of sensorimotor response, peak latency of P300 occipital derivations (O1, O2). Multilayer perceptron was chosen for the construction of the diagnostic system. To train a neural network was used back propagation algorithm. This algorithm minimizes the mean square error of the neural network by using the gradient descent method in the space thresholds and weighting coefficients. When testing the neural network correctly classified patients by more than 92% of cases. Thus, developed neural network complex reduces diagnostic time, and helps identify early cognitive decline, which ultimately leads to improved quality of life of elderly and senile age.

 

1. Zakharov V.V., JAkhno N.N. Kognitivnye rasstrojjstva v pozhilom i starcheskom vozraste. Metodicheskoe posobie dlja vrachejj. M., 2005. 71s.
2. Zakharov V.V. Diagnostika i lechenie umerennykh kognitivnykh narushenijj: Nauchnaja statja po specialnosti «Klinicheskaja medicina» // Nevrologija, nejjropsikhiatrija, psikhosomatika. 2009. № 2. S. 14.
3. KHajjkin S. Nejjronnye seti: polnyjj kurs. Izd. 2-e: Per. s angl. M.: Izdatelskijj dom «Viljams». 2006. 1104 s.
4. Akhtjorov A.V., Kirilchenko A.A. Osnovy teoreticheskojj robototekhniki. Iskusstvennye nejjronnye seti. (Obzor). M. 2008. URL: http://www.keldysh.ru/papers/2008/prep02/prep2008_02.html.