ISSN 10546618, Pattern Recognition and Image Analysis, 2015, Vol. 25, No. 1, pp. 132–160. © Pleiades Publishing, Ltd., 2015.

INTRODUCTION

The possibilities of using the method and tools of mathematics and computer science in brain research and interpreting their results are as diverse and broad as the subject matter itself of this research. It is widely knowledge that the role of mathematical and informa tional methods in natural sciences consists in the fol lowing: (1) a formal statement of problems; (2) automation of processing, analysis, and inter pretation of the experimental results; (3) construction of mathematical and imitative models of the studied objects and informational, bio logical, physiological, physical, chemical, and other material and energetic processes; (4) computer experiments with mathematical and imitative models; (5) adoption of intellectual solutions based on analysis of research and simulation results; (6) extraction from experimental data and models, structurization, and formal description of new knowl edge.

These mathematical procedures have a starkly pro nounced dual character, since the results immediately become a departure point for formulating and solving new research problems [4].

The goals and tools of mathematical and informa tional brain research can be formulated as follows: (1) goals: • identification of the principles and mechanisms defining development, organization, and information processing, and intellectual capabilities of the nervous system, • automation of information and knowledge extraction from experimental data, • simulation at the level of individual neurons, neural networks, and parts of the brain; (2) tools: • development of methods and software for analy sis and simulation, • development and theoretical and experimental study of models of the nervous system and processes in it, • development of methods, tools, databases (DBs), and knowledge bases (KBs) of the neuro sciences on two levels of analyzing the mechanisms of brain activity and its functions.

Mathematical and informational approaches are widely used in today’s neurosciences, in particular, in such essential sectors as molecular and cellular neuro science, behavioral neuroscience, systemic neuro science, developmental neuroscience, cognitive neu roscience, theoretical and computational neuro science, neurology and psychiatry, neural engineering, neurolinguistics, and neurovisualization. As well, brain research itself basically has a theoretical, algo rithmic, and structural character, and it is primarily

APPLIED

PROBLEMS

On Basic Problems of Image Recognition in Neurosciences and Heuristic Methods for Their Solution

I. B. Gurevich, A. A. Myagkov, Yu. O. Trusova, V. V. Yashina and Yu. I. Zhuravlev

Dorodnicyn Computing Center, Russian Academy of Sciences, ul. Vavilova 40, Moscow, 119333 Russia email: igourevi@ccas.ru, aem.istranet@gmail.com, ytrusova@ccas.ru, werayashina@ccas.ru, zhur@ccas.ru

Abstract—The paper describes the possibilities and main results of mathematical and informational approaches to automating the analysis, recognition, and evaluation of images in brain research. The latter are conducted in such essential sectors of neuroscience as molecular and cellular neuroscience, behavioral neu roscience, systemic neuroscience, developmental neuroscience, cognitive neuroscience, theoretical and computational neuroscience, neurology and psychiatry, neural engineering, neurolinguistics, and neurovisu alization. An important direction in simulating diseases, including diseases of the brain and their diagnoses, is the obtaining, storage, processing, and analysis of data extracted from digital images. The theoretical and methodical basis of automating the processing, analysis, and evaluation of experimental data obtained in brain research consists of the mathematical theory of image recognition and mathematical theory of image analysis. The paper presents examples of mathematical and informational approaches to automate the pro cessing, analysis, and evaluation of microimages of neurons for constructing preclinical models of Parkin son’s disease.

Keywords: biomedical image analysis, neurodegenerative diseases, microscopical images of brain sections, dophaminergic neurons, microimages of neurons, image recognition, automation of biomedical image analysis.

DOI: 10.1134/S105466181501006X

Received October 21, 2014

PATTERN RECOGNITION AND IMAGE ANALYSIS Vol. 25 No. 1 2015

ON BASIC PROBLEMS OF IMAGE RECOGNITION IN NEUROSCIENCES 133 conducted at the following structural levels (this pri marily pertained to works on brain stimulation): • the brain as a whole, • specific brain systems (e.g., the visual system), • superlarge neural networks, • small neural networks, • neurons, • ion channels and synapses, • molecular processes.

An important direction in simulating diseases, including brain diseases and their diagnoses, is the obtaining, storage, processing, and analysis of data extracted from digital images. An image is one of the most informative and widespread forms of representa tion, transmission, and storage of information; today images are actively being used as a means of presenting the results of biological and clinical research in the main divisions of the medical sciences and practical medicine. Thus, e.g., for an indepth understanding of pathogenesis, diagnosis, and treatment of neurode generative diseases (NDD), in particular, Parkinson’s disease (PD), their experimental simulation is extremely important. The development of models could be greatly accelerated and made economically more efficient by reducing the time and material expenditures on morphological research by automat ing and optimizing methods for processing and ana lyzing experimental material using modern algebraic methods of the mathematical theories of image analy sis and recognition.

The state of the art of the mathematical theories of image analysis and recognition makes it possible to pose and solve problems related to the creation of standardized replicated information technologies, the versatile highly scientific algorithmic software (ASW) supporting them, and specialized versions thereof making it possible to automate information extraction from medical images and create an objective basis for optimizing diagnostic solutions adopted by medical practitioners and researchers in interactive modes.