Sleep staging is crucial for diagnosing sleep-related disorders. The heavy and time-consuming task of manual staging can be released by automatic techniques. However, the automatic staging model would have a relatively poor performance when working on unseen new data due to individual differences. In this research, a developed LSTM-Ladder-Network (LLN) model is proposed for automatic sleep stage classification. Several features are extracted for each epoch and combined with the following epochs to form a cross-epoch vector. The long short-term memory (LSTM) network is added into the basic ladder network (LN) to learn the sequential information of adjacent epochs. The developed model is implemented based on a transductive learning scheme to avoid the issue of accuracy loss caused by individual differences. In this process, the labeled data pre-trains the encoder, and the unlabeled data re-fine the model parameters by minimizing the reconstruction loss. The proposed model is evaluated on the data from public database and hospital. Comparison experiments were conducted where the developed LLN model achieved rather satisfied performance while dealing with the unseen new data. The obtained results demonstrate the effectiveness of the proposed approach in addressing individual differences. This can improve the quality of automatic sleep staging when assessed on different individuals and has strong application potential as a computer aided approach for sleep staging.
Automatic interpretation of the EEG has so far been faced with significant difficulties because of a large amount of spatial as well as temporal information contained in the EEG, continuous fluctuation of the background activity depending on changes in the subject's vigilance and attention level, the occurrence of paroxysmal activities such as spikes and spike-and-slow-waves, contamination of the EEG with a variety of artefacts and the use of different recording electrodes and montages. Therefore, previous attempts of automatic EEG interpretation have been focussed only on a specific EEG feature such as paroxysmal abnormalities, delta waves, sleep stages and artefact detection. As a result of a long-standing cooperation between clinical neurophysiologists and system engineers, we report for the first time on a comprehensive, computer-assisted, automatic interpretation of the adult waking EEG. This system analyses the background activity, intermittent abnormalities, artefacts and the level of vigilance and attention of the subject, and automatically presents its report in written form. Besides, it also detects paroxysmal abnormalities and evaluates the effects of intermittent photic stimulation and hyperventilation on the EEG. This system of automatic EEG interpretation was formed by adopting the strategy that the qualified EEGers employ for the systematic visual inspection. This system can be used as a supplementary tool for the EEGer's visual inspection, and for educating EEG trainees and EEG technicians.
16 April 2010 in International Journal of Control, Automation, and Systems
Monitoring neurological disorders is crucial for the early detection of neurodegeneration and abnormal neural activity of the human brain. This study combines a feature-based random forest (RF) machine learning model with an image-based convolutional neural network (CNN) deep-learning approach, forming a hybrid random forest-convolutional neural network (RF-CNN) model to detect neurological disorders such as mild cognitive impairment (MCI), Alzheimer's disease (AD), and epilepsy (Ep) using electroencephalography (EEG) signals. EEG data from 19 channels were segmented into delta, theta, alpha, and beta frequency bands, generating power-based features, spectral topographic maps, and continuous wavelet transform (CWT) based scalograms, as inputs for cerebral pattern analysis. The experimental results demonstrated detection accuracy of 88 % and F1-score of 84.85 % with RF, accuracy of 97.58 % and F1-score of 95.16 % using scalograms, accuracy of 98.39 % and F1-score of 97.64 % using spectral maps, and an outstanding 99.19 % accuracy and 98.32 % F1-score with hybrid RF-CNN model. Unlike previous models that relied solely on feature-based machine learning or image-based deep learning, this approach enhances disorder detection with greater accuracy by integrating both features and images. Features like power asymmetry increase with cognitive decline, indicating hemispheric imbalance, while a declining cognition index reflects interhemispheric communication loss. Additionally, images including spectral topographic maps and CWT-based scalograms provide a comprehensive view of spatial power distribution and time-frequency characteristics. The hybrid RF-CNN approach enhances more reliable analysis of altered non-linear brain dynamics and transitional phases, making it a valuable tool for detecting neurological disorders.