A 2D Convolutional Recurrent Neural Network model incorporating scalograms achieved a classification accuracy of 88.25% for 2-second epochs in distinguishing individuals with MCI from healthy controls.
Do deep learning models using scalograms and spectrograms from EEG data accurately classify individuals with Mild Cognitive Impairment?
A 2D CRNN model using EEG-derived scalograms can accurately classify individuals with mild cognitive impairment versus healthy controls.
Electroencephalography (EEG) is a valuable tool for studying brain function and identifying neurological disorders. This study aimed to analyze EEG data using various techniques for feature extraction and classification. The data was preprocessed by applying filters and dividing it into epochs. Feature extraction techniques, including Fast Fourier Transform (FFT) in the frequency domain and Continuous Wavelet Transform (CWT) in the time-frequency domain, were applied to convert the EEG signals into scalograms and spectrograms. The primary objective was to classify individuals with Mild Cognitive Impairment (MCI) and Healthy Controls (HC) using the scalograms and spectrograms with 2D Convolutional Neural Networks (CNN) and 2D Convolutional Recurrent Neural Networks (CRNN). The classification results obtained from epochs of different durations (5 seconds and 2 seconds) were compared. The analysis revealed that the 2D CRNN model incorporating scalograms achieved the highest classification accuracy of 87.79% for 5 sec epochs and 88.25% for 2 sec epochs. This demonstrates the effectiveness of using scalograms and spectrograms in combination with deep learning models for accurately classifying individuals with MCI and HC with EEG data.
Pathapati et al. (2024) studied Mild Cognitive Impairment (MCI). Scalograms and spectrograms with deep learning models (2D CNN and 2D CRNN) was evaluated on Classification accuracy for Mild Cognitive Impairment vs Healthy Controls. A 2D Convolutional Recurrent Neural Network model incorporating scalograms achieved a classification accuracy of 88.25% for 2-second epochs in distinguishing individuals with MCI from healthy controls.