Viktor Sapunov

This paper discusses the promising areas of research into machine learning applications for the prevention and correction of neurodegenerative and depressive disorders. These two groups of disorders are among the leading causes of decline in the quality of life in the world when estimated using disability-adjusted years. Despite decades of research, the development of new approaches for the assessment (especially pre-clinical) and correction of neurodegenerative diseases and depressive disorders remains among the priority areas of research in neurophysiology, psychology, genetics, and interdisciplinary medicine. Contemporary machine learning technologies and medical data infrastructure create new research opportunities. However, reaching a consensus on the application of new machine learning methods and their integration with the existing standards of care and assessment is still a challenge to overcome before the innovations could be widely introduced to clinics. The research on the development of clinical predictions and classification algorithms contributes towards creating a unified approach to the use of growing clinical data. This unified approach should integrate the requirements of medical professionals, researchers, and governmental regulators. In the current paper, the current state of research into neurodegenerative and depressive disorders is presented.

This paper presents the results of a study of the optimal architecture of a neural network for depressive disorder recognition based on harmonized EEG datasets. Based on the re-sults of the comparison, two architectures were chosen: EEGNet and a two-dimensional fully-convolutional neural network. Using the Bayesian optimization method, the optimal model parameters were selected to achieve maximum accuracy on the test samples. The average accuracy of depressive disorder classification on the test set using cross-validation was 0.68.

In this work harmonization technique was used to combine multiple depression-related datasets with EEG data into one. A simple CNN model was trained on each individual dataset and combined one to solve the same binary classification problem. Comparison of obtained metrics allowed to back up the necessity to combine data from different sources. Additionally, the inadequacy of employing regular shuffle split instead of per-patient split to the problem under consideration was demonstrated.

The existing emotion recognition techniques based on the analysis of the tone of voice or facial expressions do not possess sufficient specificity and accuracy. These parameters can be significantly improved by employing physiological signals that escape the filters of human consciousness. The aim of this work was to carry out an EEG-based binary classification of emotional valence using a convolutional neural network and to compare its performance to that of a random forest algorithm. A healthy 30-year old male was recruited for the experiment. The experiment included 10 two-hour-long sessions of watching videos that the participant had selected according to his personal preferences. During the sessions, an electroencephalogram was recorded. Then, the signal was cleared of artifacts, segmented and fed to the model. Using a neural network, we were able to achieve a F1 score of 87%, which is significantly higher than the F1 score for a random forest model (67%). The results of our experiment suggest that convolutional neural networks in general and the proposed architecture in particular hold great promise for emotion recognition based on electrophysiological signals. Further refinement of the proposed approach may involve optimization of the network architecture to include more classes of emotions and improvement of the network’s generalization capacity when working with a large number of participants.