Predicting Mini-Mental State Examination Scores Using Electroencephalography Signal Features

Abstract

Purpose: The purpose of this study is to use linear and non-linear features extracted from Electroencephalography (EEG) signal to predict the Mini-Mental State Examination (MMSE) test score by machine learning algorithms. Materials and Methods: First, the MMSE test was taken from 20 subjects that were referred with the initial diagnosis of dementia. Then, the brain activity of subjects was recorded via EEG signal. After preprocessing this signal, various linear and non-linear features are extracted from it that are used as input to machine learning algorithms to predict MMSE test scores in three levels. Results: Based on the experiments, the best classification result is related to the Long Short-Term Memory (LSTM) network with 68% accuracy. Conclusion: Findings show that by using machine learning algorithms and features extracted from EEG signal the MMSE scores are predicted in three levels. Although deep neural networks require a lot of data for training, the LSTM network has been able to achieve the best performance. By increasing the number of subjects, it is expected that the classification results will also increase.

Keywords

• Mini-Mental State Examination
• Electroencephalography Signal
• Machine Learning Algorithms
• Electroencephalography Feature Extraction