Use of Machine Learning Methods to Analyze Patterns of Brain Activity during Assessment of the Self and Others

• Published in: Neuroscience and behavioral physiology 2023-09, Vol.53 (7), p.1210-1218
• Main Authors: Knyazev, G. G., Savostyanov, A. N., Rudych, P. D., Bocharov, A. V.
• Format: Article
• Language: English
• Subjects: Activity patterns; Affective disorders; Behavioral Sciences; Biomedical and Life Sciences; Biomedicine; Brain mapping; Brain research; Cortex (cingulate); Functional magnetic resonance imaging; Information processing; Learning algorithms; Machine learning; Medical imaging; Medical research; Neurobiology; Neuroimaging; Neurosciences; Neurosis; Personality; Personality traits; Prefrontal cortex; Principal components analysis
• ISSN: 0097-0549; 1573-899X
• DOI: 10.1007/s11055-023-01517-2

Studies of cerebral activity during the processing of self-referential information, in comparison with the processing of information related to other people, are based on use of mass univariate analysis with the assumption that activity in one region does not depend on activity in other regions. Recent times have seen an increase in interest in the use of neuroimaging in studies of spatially distributed information using multidimensional approaches such as multi-voxel pattern analysis (MVPA). We report here the use of MVPA to analyze fMRI data recorded during a task involving assessment of the self and other people of different degrees of closeness. Testing of the patterns identified by machine learning showed that these brain activity patterns predicted what the subject was assessing self or other in 75–88% of cases. Prognostically significant structures were widely distributed in different areas of the brain and, in addition to the cortical median structures making the greatest contribution, included areas of the visual, lateral prefrontal, and many other cortical areas. The most informative areas for the selection of the Self variant on classifying self/other were the ventral regions of the medial prefrontal and cingulate cortex, while for selection of Other the most informative were the parietal and occipital median areas. Principal components analysis revealed a combination of brain structures, including the anterior cingulate gyrus and the bilateral amygdalas, whose factor scores correlated positively with the psychometric reward sensitivity scale and negatively with the neuroticism scale. Activity in this combination of structures can be regarded as a protective factor against affective disorders. In general, the results obtained here demonstrate the productivity of using machine learning methods for analysis of data from experiments of this type.